Sleep and Suicide Prevention: Advancing Innovation and Intervention Opportunity-Day 2
REBECCA BERNERT: Hello, everyone. Welcome back to day two of our NIMH virtual workshop on Sleep and Suicide Prevention: Advancing Innovation and Intervention Opportunity. On behalf of myself and my co-chair, Dr. David Leitman, I'm pleased to co-chair this session today, and I'd like to thank you all for joining us, and to all of our incredible speakers throughout the workshop.
In review of our sessions yesterday, we provided a state of the science overview introducing this area as an emerging subfield in suicide prevention, followed by a focus on lifetime and timing mechanisms, sleep neurobiology and neuroplasticity, and the role of hyperarousal and inflammatory markers at the intersection of risk, whereas today we'll explore neurocognitive factors and affect regulation, novel therapeutic frameworks and treatment development, followed by technology innovation and digital medicine.
We'll end with a general discussion today among all of our speakers and panelists to guide innovation and insights in this area and building on the workshop findings as a whole. With this in mind, I'm delighted to introduce our first session chair and my esteemed colleague and collaborator, Dr. Andrea Goldstein, from Stanford University. Dr. Goldstein.
Session 1: Neurocognition, Learning and Affect Regulation
ANDREA GOLDSTEIN-PIEKARSKI: Good morning, everyone. So I am Andrea Goldstein-Piekarski. I am the session chair for session one, day two, from Stanford. We have a great lineup today. Gina Poe, from UCLA, will be starting us off, followed by Dr. Kelly Zuromski, from Harvard, Dr. Erin Wamsley from Furman, and Dr. Eti Ben Simon, from Berkeley. And of course, it's Dr. Gina Poe, as well.
It is an early morning here, and of course sleep always comes difficultly before a big talk, so I'm going to go ahead and pass it off to Dr. Gina Poe now. Thank you.
Agenda Item: Sleep and Memory
GINA POE: Thank you so much. It's such a pleasure to be here, everyone, and I want to start my talk, dedicating it to my Uncle Danny, who, when he returned from Vietnam, his wife and best friend ran off together and took his five-year-old daughter with them, and he closed his bank accounts, gave away all of his things, and then drove his car off the highway at 70 miles an hour. I know the family would like to think that it wasn't suicide, but my mother, one of his six sisters, thinks that it definitely was. And I want to talk about -- this is his daughter, leaving his daughter, my grandma, and six sisters and three brothers, and my brother and I behind.
He was a gentle spirit. He played the flute for us. He taught us how to ride bikes. He was a wonderful man, and I can't help but think that if we had had a cure for PTSD, he would have had the resilience to be able to withstand the subsequent insult of his family leaving him behind.
So what happened? Why did he fail to resolve that emotional memory? He couldn't put the past in the past. This is our model. He couldn't integrate this memory into his whole schema of who he was, he couldn't prune specific memory circuits associated with that emotion, all of those emotions, when he was trying to resolve them, and he couldn't detach that somatic and emotional experience from his current day-to-day reality.
I know I've been through trauma myself. When my first husband left me it was so painful, I wanted to throw up, and in fact I did throw up. For 40 days, I wrestled with it, and grappled with it, and slept. Thankfully, I was able to sleep, and probably that sleep was adaptive, because after the period of about 40 days, I was able to find some peace and move on. Of course, my insult wasn't nearly as bad as his, and I didn't have PTSD to start with.
What I think the key is, and my research has pointed toward, is the key is good adaptive sleep. We all know what sleep looks like or feels like, and today we're going to talk about the signatures of sleep, which are slow waves in that deep slow-wave sleep state that we go into fairly early in the night, and sleep spindles, which are about one-and-a-half second long, 10- to 15-hertz squiggles that go on in our brain, electrical squiggles that go on in our brain, during this N2 transition to REM sleep state, and the theta waves that go on during rapid-eye-movement sleep, which is our dream state. And I'll be talking about these last two more than any others.
So what happens in various stages of sleep? What we know now is that slow wave sleep serves to set up our brain, restore the energy stores, of scaffold, of things that we've just built during the day. It clears our brain of the trash, and all of this is really important to set up our brains so that we can do the important cognitive work of integrating and consolidating our memories through those sleep spindles, to transfer information from short-term to long-term stores in an integrated fashion, into our schema.
And then during REM sleep, to clean our brain of those synapses that we no longer need, to weaken those old connections -- for example, the emotional connections of our semantic and episodic memories, and strengthen new insights that we can gain through all the grappling that we do during the day, with the trauma or the thing that just happened to us.
So the neurotransmitter milieu is really also important in these various stages of sleep. First of all, acetylcholine really helps set up our brain to be plastic, and it sets up our hippocampus with this theta rhythm, which sometimes goes into the cortex as well, and theta is also really important for forming new memories and also in my research, to erase old memories. So when theta is high, which is during waking and REM sleep, that's when our plasticity is also very, very high. The absence of acetylcholine sets up slow waves; in fact, there have been many studies to show that you actually don't get slow waves in the brain unless your acetylcholine is absent.
Norepinephrine is our fight-or-flight stress axis. It's also really important for learning. So a little bit of stress is really good for learning, and it's high whenever we're awake. But when we go into that transition to REM-N2 stage of sleep, and REM sleep, then norepinephrine levels are really low, and in fact, they need to be low. Both acetylcholine, norepinephrine, and serotonin all need to be low to set up sleep spindles, thalamocortical sleep spindles.
When it's present, it also makes the theta that comes from the hippocampus even larger. Serotonin, the absence of serotonin -- we're going to touch on briefly today -- is also extremely important during the stage of REM sleep to depotentiate novel circuits in order to divorce that recent traumatic memory and all of those emotions associated with it from the semantic and episodic long-term memory stores. So these two neurotransmitters, especially, are what I'm going to be concentrating on today.
So what sleep features are important for emotional health? Is it slow waves, spindles, theta, acetylcholine in REM sleep, or all of the above? What would you choose? The answer is all of the above, and more.
This is the outline that I'm going to be talking about today, of sleep for emotional wellness. After cleaning and restoring the brain, which is essential to set up the brain for everything else that we have to do during the night, and after our new memories are consolidated in that N2 transition to REM spindle sleep state, our memories can become familiar by strengthening the familiarity circuits in our brain, and that familiarity can set up a process whereby the novelty tags of the new memory can be removed, and it doesn't have to stay salient and prominent in our novelty inputting pathways.
Shame and other emotional circuits can be detached from the semantic and episodic recall, and that is through a disconnect between our emotional activation and our prefrontal cortex, which normally happens during REM sleep.
So here's just a schematic of it. Our spindles incorporate new information into existing neural circuits, so when we acquire new information, this is show a schema here of the world, we can, through sleep spindles, connect that information. The hippocampus and prefrontal cortex have special connectivity during those sleep spindles, where information can be transferred and synapses can be strengthened, and then, during REM sleep, during the theta activity that goes on during REM sleep, my research is showing that that is really important for pruning the redundant connections that no longer need to be supported once we have consolidated the memories into the neocortex, so it can be downscaled from our novelty, high-salience circuits, through REM sleep theta. So that new schema can be sown in, new information can be sown into our schema in a coherent fashion.
If our sleep becomes maladaptive, for example, in PTSD, then schema cannot be properly incorporated, because one of the problems with maladaptive sleep is the sleep spindle state isn't normal, and we don't get to incorporate that new information in a coherent fashion. What's more, one of the things that many research has shown, is that norepinephrine stays high in sleep, including REM sleep, and what that would do is serve to only restrengthen, in fact, those traumatic memories, those old things that stay salient. It only keeps it salient in the structure.
When I was a graduate student, John Lisman came to UCLA when I was at UCLA for graduate school, and he showed us a new paper that he and Patricio Huerta had found, that reiterated what Constantine Pavlides had already shown, and that is when he added acetylcholine to a slice of the hippocampus and stimulated at the peaks of that theta wave that resulted, he got really beautiful long-term potentiation, with very physiological levels of stimulation.
Furthermore, what was really exciting is when he stimulated at the troughs of theta in this slice or REM-like neurochemical environment -- there's no norepinephrine or serotonin added, just lots of acetylcholine -- he would get a reliable depotentiation of what was previously potentiated, and this was the first you can get physiological LTP and depotentiation reliably.
So when I went to my postdoc at the University of Arizona, I was able to access a bunch of rats that had electrodes implanted into their hippocampus, and I knew already that when rats are learning in their environment, there's lots of theta, lots of acetylcholine in their brains, and the cells that are associated with learning are firing primarily at the peaks of those theta waves. So that was great, because that's consistent with long-term potentiation.
But when they went into REM sleep, what we found was that the cells switched their phase to fire at theta troughs, consistent with depotentiation. So the hypothesis that Francis Crick and Graeme put out there that REM sleep is for forgetting seemed to be right. That was really exciting to me. Maybe REM sleep was really for erasing all of those synapses that no longer need to be supported, once they're consolidated.
Furthermore, we looked at the time course of that switch, and what we found is that all those cells continue to fire always at theta peaks when animals were experiencing the maze and learning session, even once it became familiar. During REM sleep, it switched from firing at theta peaks, consistent with maintaining those memories, to theta trough over the course of a week, and that week is also the same time course of memory consolidation from the hippocampus to the cortex. So it was really exciting to see that cells could go from theta peaks to theta troughs in a manner that was consistent with that consolidation of that memory.
I have spent some time trying to chase how that happens. Around the same time, Mark Thomas in Tom O'Dell's lab at UCLA found that when norepinephrine remains present, you actually can't get depotentiation, no matter what you do. No matter what kind of stimuli you give, you cannot get depotentiation when you have norepinephrine present. And interestingly, the time that the locus coeruleus, which provides norepinephrine to the forebrain, is off is during REM sleep and during those periods of time in that N2 transition to REM sleep state, when sleep spindles appear. That's when the locus coeruleus is off, norepinephrine is absent, and erasure is possible. So this is the firing pattern of the locus coeruleus neurons across sleep-waking states, both that was measured by Gary Aston-Jones in 1981, and by many other labs since, and my lab recently.
So the locus coeruleus, providing norepinephrine all over the brain, is also, I just wanted to say, recently found to be modular. It's not just one structure. It projects to different places at different times, including the prefrontal cortex, as different from other areas.
So what happens when you can't downscale that norepinephrine level during REM sleep and the transition to REM sleep? Insomnia is actually a state of disturbed sleep, and we learned a lot yesterday about how insomnia predicts suicidality. Insomnia is characterized by reduced sleep spindles and more arousals from REM sleep and that transition to REM sleep spindle state. We have a heightened sympathetic drive, fight-or-flight; that's probably why it's difficult to actually fall asleep or stay asleep. The locus coeruleus never really seems to rest. And we know that insomnia really does associate with depression, very strongly with depression, and other anxiety disorders.
There was a recent paper, a couple of years ago, where they found that sleep resolves emotional distress. And this principle is going to be talked about by Eti Ben Simon later in today's session. What happens is that these patients with insomnia seem to be haunted by the past. The old emotions remain salient in insomnia disorder, and what they did is they exposed normal sleepers and insomniac sleepers to a new insult. They had to listen to themselves singing karaoke at the top of their lungs without any background music, and this was an embarrassing thing that elicited embarrassment and emotional distress in all of the listeners. But those who were normal sleepers had the activity of all of these self-conscious and emotional areas. Those who had insomnia disorder had the same areas, but stronger. Larger areas are involved in that same embarrassing experience.
Furthermore, so when people were recalling an embarrassing, emotionally embarrassing situation that happened years ago, people with insomnia disorder had similar areas of the brain lighting up again, as though it just happened yesterday or a few days ago, rather than years and years ago. Whereas normal sleepers had resolved that and their same emotional areas hadn't lit up. Also, the galvanic skin response and other signs of emotional distress continued to light up in people with insomnia disorder, whereas it didn't in normal sleepers.
So this is just the bold change image change in these people with insomnia, that just had heightened activity that never resolved, even after years of living with this memory.
One thing that happens during REM sleep and also during slow wave sleep is the prefrontal cortex, the area that's involved in judgment and decision-making, is relatively hypoactive. It doesn't really light up. The only time that the prefrontal cortex is active during sleep is during those sleep spindles when memories are being sown into the schema, new memories. But the emotional circuit is really, really active in REM sleep. The amygdala, the hippocampus, the cingulate cortex, all of that is super active. So there's a disconnect between activity in the prefrontal cortex and the emotional system.
And in the model that we're working with right now, that disconnect is critical to divorcing of the emotional memories from the semantic memory of what happened. So when we're awake and experiencing a traumatic or emotional event, we are encoding it, the locus coeruleus is on, it helps us learn really quickly and learn really well, what all of this things are happening when we're awake. When we go into sleep, during that transition to REM spindle state, the prefrontal cortex is uniquely connected to all of this memory and limbic circuitry, and the locus coeruleus is on a little bit, at times, at the end of each sleep spindle, it turns on, helping us to consolidate that memory into our old schema.
And then in REM sleep, what happens is that the prefrontal cortex is off while that emotional circuitry is still really highly active, and what that does is it sets up what's called heterosynaptic depotentiation. So when a postsynaptic target is off while the presynaptic target is really, really on, presynaptic input is really highly on, then especially in the absence of locus coeruleus activity you can get depotentiation through that heterosynaptic activity, and that is what we think is critical to the divorcing of that semantic memory, that episodic memory, from the emotionality, allowing us to not feel the same emotions when we recall that traumatic memory. Now I can remember my divorce with sanguinity, if that's the word, and it doesn't bring up all the same emotions that it did at the time. Although I remember being emotional, I just don't feel that way.
So here is a circuit-level description of how it's possible that the emotion can be hollowed out and the saliency of that emotion can be reduced. What happens is within the absence of serotonin, once this familiar memory gets strengthened out here in the temporoammonic circuitry, through the sleep spindles, these distal inputs are the ones that are the familiarity encoding circuits. Then, without serotonin, the dendritic spikes can reach the axon hillock and cause the cell to fire, and because that firing is occurring when input through the novelty encoding circuit is not happening, then that firing can actually cause this circuit to disappear and reduce its activity.
So the novelty encoding circuits can be cleared in the absence of norepinephrine during REM sleep and REM sleep only. So in posttraumatic stress disorder, we have heightened sympathetic drive, lower REM sleep theta -- it's not something I've talked about, but that's what norepinephrine does, its presence lowers REM sleep theta, and that has been found in people with PTSD, and we have insomnia and disturbed sleep and nightmares, all of that is probably caused by too much norepinephrine during sleep. People get stuck in the past, they can't contextualize the fear or shame or guilt, and can't detach from their emotions and the memory stays salient and novel, and that's probably because our hippocampal novelty memory circuit is saturated and probably happens also in the cortex, which is set up the same way.
Just one more last word, which is why it's really important that norepinephrine and serotonin be absent during REM sleep and how that would contraindicate the administration of antidepressants for people with an anxiety disorder, at least during sleep. SSRIs and selective noradrenergic reuptake inhibitors would not allow this downscaling of norepinephrine and serotonin during REM sleep, and in fact it prevents at least adaptive REM sleep from happening. So we still stay haunted by our past, overdriven by the present stressful situation, and we have a dysfunctional reasoning and system because our prefrontal cortex is unable to divorce from that emotionality, and that is a very dangerous combination that could lead to suicide.
I want to thank my lab, the heroes here, that are working on this project in sleep and PTSD and thank the NIMH for its continued support for the last 20-plus years. Thanks, everyone.
Agenda Item: Neurocognitive Factors in Suicidal Behaviors
KELLY ZUROMSKI: Thank you so much. That was a really interesting talk.
Good morning, or good afternoon for most of you. My name is Kelly Zuromski. I am a research scientist at Harvard University, and I'm going to be talking today about neurocognitive risk factors and suicide. This is a pretty broad topic and extensive area to cover in a brief 15-minute talk, so I've outlined a couple goals that I'm hoping to accomplish and get through today.
First, I'll provide you with some background research, reviewing the state of the literature on neurocognitive factors, including general factors, and some suicide-specific neurocog factors that I'll get a little more into later on. Then I'll present some of my recent research findings from this work, specifically from my recent K award that's ongoing, and then I'll get into some quick, brief future directions to wrap things up.
When we're talking about neurocognitive factors, perhaps the earliest that we can label as examining these types of factors was the work of Neuringer in the 1960s, who studied rigid thinking, and this idea that suicidal individuals may have difficulty developing new or alternative solutions when facing high distress.
Of course, in the last 60 years or so, we've significantly expanded upon our definition of neurocognition beyond cognitive flexibility or rigid thinking, and nowadays neurocognitive factors really refer broadly to a set of cognitive functions that are closely linked to neural pathways in the brain. A number of these neurocognitive factors have been studied in the context of suicide risk, including decision-making, cognitive control, inhibition, shifting or switching, and a bunch of other ones as well.
I'll note that these factors have been operationalized and grouped and named in a lot of different ways. For example, a chunk of these factors listed here generally are grouped together under the umbrella term of executive functioning, and others are thought to measure impulsivity, typically decision-making and inhibition, or some combination thereof. And one of the most recent frameworks that's been used that might be helpful to organize this broad set of risk factors is of course RDoC, NIH's research domain criteria framework, where most of these factors would fall into the cognitive systems domain, and to a lesser extent some in positive valence systems.
Like I mentioned, this is a fairly complex literature base that we'll only be able to scratch the surface of today, but for the purposes of simplicity, I'll just use the RDoC framework today to dive into some of the most well-studied of these factors, namely the first few that are listed here: decision-making, cognitive control, inhibition, and shifting or switching, which the latter two are typically thought of as aspects of cognitive control.
These kinds of neurocognitive processes are often assessed using task-based measures, so organizing this using the RDoC framework that I just mentioned, I'll briefly highlight some of the most well-studied neurocog factors in the context of suicide, which would be cognitive control and reward valuation, which also include a number of other subconstructs listed here, and you can see I've just tried to outline a couple of the corresponding tasks that might be used to measure each of these constructs.
So for example, for inhibition, the go/no-go task is something that's used pretty commonly for studying inhibition; for response selection, the Iowa Gambling Task, for example. This is just a small subset of the measures that do exist, and I'll focus on the ones that are listed here are the ones that have really been studied the most frequently in relation to suicide, and they've been used in a number of studies. So, just to summarize the state of that literature base of using these neurocognitive tasks concisely, I'll just briefly review the results of a couple recent meta-analyses examining these tasks in relation to suicide.
For cognitive control, we're generally seeing small-to-medium effect sizes for measures of cognitive control and suicide attempts, the association there, including the go/no-go, the CPT, the Iowa Gambling Task, the Stroop, and these effects notably have only been found for suicide attempts. Suicide ideation has been examined in a couple of the metas listed here, but it was not significant, or in some cases, it wasn't examined at all. And death by suicide was not examined due to the lack of studies with that outcome.
For reward valuation, we are generally seeing slightly larger effects, in the medium-to-large range, for the association between delay-discounting tasks specifically and both suicidal ideation and attempt. But importantly, all the studies that were included in these meta-analyses, using the tasks, were cross-sectional, so we have some helpful information that these constructs may be associated or related to suicide risk, mostly in case control studies comparing suicide attempters to non-attempters, but what we don't know so far at least is whether or not these types of neurocognitive factors can prospectively predict suicide risk.
There has been one prospective meta-analysis, actually conducted by Cassie Glenn, who might be on the call, in our lab a few years ago, using the RDoC framework, so what they found was that for cognitive control, looking at prospective studies, they found small pooled-effect sizes for cognitive control, which was by-and-large assessed via self-report measures of impulsivity like the Barratt impulsiveness scale for predicting suicide attempt.
And these self-report measures did not significantly predict suicidal ideation or death. And further, no prospective studies were found that looked at reward valuation, so clearly a lot more work is needed there. For these two constructs, all the studies that were included in this meta, were using self-report measures like I just mentioned of impulsivity, for example, so we haven't even been able to evaluate the tasks that I listed on the prior slide in prospective studies.
Shifting gears slightly, in addition to all of the general neurocognitive factors I just talked about, things like cognitive control and reward valuation, I'd also just like to briefly highlight some suicide specific neurocog factors because we've seen that these so-called general factors tend to be fairly weak predictors of suicide. So there's a whole other body of literature that sought to develop tasks that are specific to suicide, to better understand, for example, how people are thinking about suicide or to focus on psychological processes that might be a lot more relevant to suicide than some of the neurocognitive factors that I've mentioned so far.
One example of that would be the suicide implicit association test, which was a task that was designed to try to tap into attitudes about self-harm that may be operating at a more automatic or implicit outside-of-conscious-control level. And looking at results of the suicide IAT, which we don't have time to get into a lot, but we do see that self-harm related implicit associations are robustly associated with suicide attempts, and this is an effect that has been replicated in several studies.
Another example of a suicide specific task is an escape decision making task that was designed by one of my colleagues at Harvard, Alex Millner. And this is a reinforcement learning task that was developed to try to examine this hypothesis that suicide represents an effort to escape an aversive psychological state. So results for studies where he's used this task have shown that indeed suicidal participants are showing a bias on this task for making responses to escape aversive states, compared with psychiatric controls.
So these are just a couple suicide-specific neurocognitive tasks that have been developed, and I bring these up because these types of tasks might map on pretty clearly to the types of processes that we know are relevant in suicidal states, so using these types of tasks might help us be a lot more specific when trying to understand how and why suicide happens.
To summarize what I've said so far, the most studied neurocognitive factors, general factors, tend to have pretty modest associations with suicide attempts only, and the majority of these studies using the neurocog tasks have been cross-sectional, and there's been some self-report research showing that neurocognitive factors like impulsivity can prospectively predict suicide attempts, although small effect again. And there's also another literature base focused on suicide-specific neurocognitive factors that may be more relevant, ultimately, to understanding suicidal states.
In terms of next steps in this research area, one of the most important things that's needed is more longitudinal research, which seems to be a common limitation in suicide research, but for all the neurocognitive factors that I've shown today, more longitudinal research is needed, including for the suicide-specific ones.
Of course, another thing to point out is that I've been talking all about neurocog factors in suicide specifically, just for the purposes of this talk, but we know of course that sleep problems are closely linked and are strongly associated with deficits in neurocognitive functioning, mood and affect are also really strongly related. So being able to examine the interplay between all these factors is another important area, and one method that might be helpful to do so is using ecological momentary assessment, which is of course an increasingly used method that might be helpful to try to examine risk factors like these in real time, to try to capture short-term fluctuation in factors that may recede or may coincide with when people actually are feeling suicidal.
A lot of the work that I've been doing the past couple of years at Harvard and related to my K award that I'm in year two of now, has been in this area of real-time assessment of suicide risk. So I'll now present on some brief recent findings from the work that I've been doing, that maybe can shed some more light on the dynamics between neurocognitive factors, suicide, and sleep, briefly talking about two studies.
The first is the ongoing project that's funded by a couple grants, including my K23, and a U01 that my K mentor, Matt Nock, has. And this is a real-time monitoring study, where we're following suicidal adults and adolescents for six months after they're discharged from psychiatric hospitalization. Participants are monitored in this study in a couple of ways. One is that they complete EMA smartphone surveys throughout the day, completing six surveys a day for the first three months of the study, and then daily for the last three months, and in these surveys, and each survey takes about 90 seconds or so to complete, we ask questions about mood, suicidal thoughts and behaviors, and sleep. And we're also collecting wearable biosensor data, but for the purposes of this presentation I won't be talking about those data, but they do exist, and we'll be using them in the future.
What I'll show you today are some preliminary results in which we are examining the association between one neurocognitive factor, impulsivity, which as you may remember from earlier was the only neurocognitive factor in the literature thus far that's been shown to prospectively predict suicide attempt. So looking at self-reported impulsivity, suicide risk, and sleep, using the following EMA items on a 0 to 10 scale to do so: how much do you feel impulsive, how strong is your urge to kill yourself, focusing on urge specifically given the conceptual overlap with urges and impulsivity, and then asking about sleep quality.
I think, based on a recent paper that I just read a few days ago on momentary assessment of impulsivity, I think this is the first look at a momentary assessment of impulsivity in suicide in real time. So let's take a look.
Here's what we did. This is going to be results from a multilevel mediation analysis, and we see that lower sleep quality is associated with higher levels of impulsivity the next day. So this is prior night sleep quality. And the higher impulsivity is associated with higher suicidal urges. And not surprisingly, see that lower sleep quality is associated with higher suicidal urges as well.
But we wanted to know whether impulsivity was mediating this association between sleep quality and suicide urge, and we did find that there was a significant and direct effect, so it appears to be that people are reporting that they're not sleeping well, and then the next day they're feeling more impulsive, and they're also feeling more suicidal or having higher suicidal urges.
This has been a pretty interesting effect to see, kind of highlighting this dynamic association between all three of these variables. We're going to dig into this further and see if whether or not this effect replicates in our adult sample. This is only in our adolescent sample at this point. But pretty interesting initial findings.
Knowing that sleep, suicide, and impulsivity are all related, we were also curious just to see what participants' suicidal urges looked like across the day, expecting that you might see higher levels of suicidal urges late at night, perhaps when people weren't sleeping or having trouble sleeping. So as you can see here, this is a smoothed average of participants' self-reported urge to kill themselves over here on the y-axis, and then hour of day down here on the x-axis. And what you can see is that suicidal urges are typically highest later in the day, and midnight, which is over here. And that they tend to rise throughout the day starting in the afternoon, up until the evening.
We wanted to dig a little bit deeper on this idea of urges for self-destructive behaviors like suicide, potentially varying based on time of day or across the day, and so to do so I'll just present very briefly on one other study, preliminary data from a study led by Daniel Coppersmith, a graduate student of Matt Nock's, and he conducted a study, it was also focused on suicidal individuals, suicidal adults, who were recruited online and who completed a bunch of EMA smartphone monitoring over a six-week period, and he assessed for a bunch of urges as well, in addition to what we did in the study that I just presented, urge to kill yourself -- also assessing urge for NSSI, drugs and alcohol, et cetera. So we wanted to look at how these urges varied across the day as well.
Again, this is a lot to look at all of a sudden, but the urges are on the y-axis and hour of day is on the x-axis, so kind of just eyeballing it, you can see that participants are consistently reporting higher urges at nighttime across the board. For urge to kill themselves, urges to self-injure, to drink, to use drugs, to smoke, and to binge eat.
This is just a preliminary look at the trajectory of some of these urges over the course of a day, but I know that even just putting this figure together for this talk, and I'll be talking a lot about this interesting finding, thinking about what could be going on here, what might be driving these effects, and there's probably a lot of things at play, including sleep pressure and people being really tired at night, maybe some suicide-specific factors, but one hypothesis that we have been thinking about is whether nocturnal wakefulness, or just being awake at night, in itself might be a risk factor for suicide.
Right now, what we're doing for future directions is we're working on some research building off the work of folks like Michael Perlis and others who have posited that just being awake at night when you're not biologically prepared to be awake could itself confer suicide risk, and possibly through a mechanism of reduced frontal cortex activity. And there are some work suggesting that we could be onto something here. Suicide is three times more likely to occur at night compared to any other time of the day, so we're excited about this new direction and collaborating on this project with Michael and Beth Klerman at MGH to try to examine this hypothesis and understand the interplay between neurocognitive factors, sleep, and suicide.
So I'll leave you with that idea and just do some brief acknowledgements of my research group that I work with, particularly Matt Nock and Evan Kleiman, who are the co-mentors on my K award, Daniel Coppersmith for allowing me to share his data today, the rest of the Nock lab. Several of the studies that I talked about today were funded by the NIMH and of course we'd like to thank our clinical partners, Franciscan Children's and MGH, with whom we worked to recruit participants for our studies.
With all that being said, I will now turn it over to Dr. Wamsley.
Agenda Item: Sleep, Dreams, and Learning
ERIN WAMSLEY: Thanks, Kelly. Very fascinating work. So, my research is in the area of memory consolidation during human sleep, and also how the content of dreams relates to memory consolidation. And that's what I'll be speaking about today.
A critical question related to this workshop is whether dreams might reflect cognitive and emotional disturbances seen in psychiatric disorders, and in turn whether dreaming could help us to understand suicide risk. I'd argue first that it's now well-known that the consolidation of memory after learning happens during sleep, and there's emerging evidence that dreaming itself seems to be related to this process. Work from my own lab and others suggests that the content of our nightly dreams is influenced by the reactivation and consolidation of memory.
This could relate to understanding suicide risk, because memory consolidation and emotion regulation processes are disrupted in depression and some other psychiatric disorders, and disruption of these functions may actually be evident reflected in patients' dream content.
Behavioral research dating back decades shows that when participants sleep after learning, their memory for that just-learned information is improved later on, relative to control groups who remain awake after learning. And interestingly, this has really been shown to be true across all forms of learning and memory that have been studied, very diverse forms of memory systems, including, for example, verbal learning, maybe this classic word-pair associates task, spatial navigation, in humans and animals, and novel environments, and also including memory for vivid emotional material. Of course, during sleep, at the same time as the consolidation of memory is thought to be occurring, we are also dreaming.
And interestingly, the actual content of dreams appears to at least in part reflect this memory processing that's happening in the sleeping brain. A first piece of evidence here is that although it's actually surprisingly difficult to get research participants to dream about something on command, it can be done, and participants especially will tend to dream about novel interactive learning experiences that are relevant to them.
Early studies of dreaming back in the 1960s, for example, tried to influence participants' dreams by showing them static images or films of various kinds, but that was largely unsuccessful, really. However it was noticed very early on that participants are extremely likely to dream about the novel experience of the sleep laboratory environment itself. The electrodes, the researchers, sleeping in a strange place with people watching you -- because that experience of sleep in the lab with this equipment is much more impactful and meaningful of a learning experience than most of the particular stimuli that researchers had come up with.
But more recently, our lab and others have had success in manipulating dream content using not static images and films, but engaging interactive videogame-like experiences that participants are apt to become highly invested and absorbed in. And we've leveraged that method to test whether dreaming about a recent laboratory introduced learning task might actually predict the memory benefit that's conferred by sleep.
In this work, we've used, among other things, a virtual maze navigation task like that pictured here, and in this task, essentially, participants are placed in this complex maze environment, and are asked to repeatedly navigate to a specified goal point across a series of trials. After an exploration period there would be a series of trials where from different randomized starting locations in this novel environment participants must escape the maze, escape the maze, escape the maze.
After encoding, we would test the effect of post-learning state on consolidation by experimentally manipulating whether participants enter a period of sleep or not just after learning, prior to retesting them on their retention of memory for that environment later on. So for example, in one study using this task, we had participants report to the laboratory around noon to train on this virtual maze task. And just after that, participants were randomly assigned either to lie down for a 90-minute nap opportunity or else to remain awake during that time, and then at the end of the day, we test participants' retention for the spatial environment layout that they learned earlier. So the main dependent variable in this study and similar studies is really going to be the extent to which participants retained memory from the end of training to the delayed test.
What we found, not surprisingly, first of all, is that napping did lead to improved memory for the maze environment, such that participants that were randomly assigned to remain awake showed some forgetting across the day. But those were able to take a nap after learning actually improved their navigation performance later on.
But during this study and some similar ones, we also asked participants both in the sleep condition and in the wake condition, actually, to report on their subjective experiences. So during the nap or the wake period, we prompted participants up to three times to just report everything that was going through their mind just before we called them.
Then what we did was classify participants into two subgroups. One group of participants were those who in fact did report thinking about or dreaming about or imagining either this particular maze task, or a maze, even if it did not resemble this particular one.
And we saw that across wake it didn't really matter much whether participants reported thinking about the maze task. But across sleep, those who reported sleeping and dreaming about the maze task improved their memory ten times more than participants who slept and did not report a dream about the maze task. Solid green bar are those in the sleep condition who reported a dream about the maze. The hatched bar are those in the sleep condition who did not report a dream about the maze. Y-axis is improvement in speed to solve the maze.
We've more recently replicated this observation in an overnight study, seeing the same thing. And there have recently been a few similar papers published by other labs, all showing that participants who report dreaming about a learning task show a greater benefit from post-training sleep for their memory, compared to those who do not report recalling a dream about the learning task.
So it may be that memory consolidation during sleep is actually expressed in the specific content of our nightly dreams, and that could be relevant to the topic of this workshop, because we know that memory consolidation during sleep is in fact impaired in several different psychiatric disorders, including depression.
As an example, this 2018 study by Harrington et al argued that specifically for negative emotional material, overnight memory consolidation is impaired in participants with depression symptoms by sleep deprivation. So as you can see on the graph in this study, Harrington looked at two subgroups of participants, one that was high on their scores on the Beck Depression Inventory and one that was low on their scores on the Beck Depression Inventory, and the dependent variable here is the extent to which participants retained memory for negative emotional images across the night. The experimental manipulation was sleep deprivation, with the solid black bars representing participants who underwent a full night of sleep deprivation and the open bars representing participants who slept normally.
So as you can see, Harrington et al observed that memory was more affected by sleep deprivation in participants who were high on depression symptoms relative to those who were low on depression symptoms. So this is one of the few pieces of evidence suggesting that emotional memory processing during sleep specifically could be impaired in depression.
So is this relevant to dreaming? Well, you know, it's a viable hypothesis, if memory consolidation is expressed in dreams and if emotional memory processing during sleep specifically is impaired in depression, could it be that dream content itself can tell us something useful about the disruption of emotional processing in depression and other psychiatric disorders potentially?
Now, perhaps surprisingly and perhaps not, there's not a lot of research specifically addressing this question in terms of the dreaming level of analysis. But we do know a few things. First, we know that the ability to even remember dreams at all appears to be impaired in depression. Even in patients who are not under treatment with antidepressant medication, persons with depression symptoms report unusually low rates of dream recall in comparison to healthy controls or participants who score low on depression symptom measures.
Second, there are a few longitudinal studies showing that as patients' depression symptoms improve over time, also concomitant with that the affect of their dreams becomes more positive. We see fewer negative emotional dreams.
In fact, a handful of studies has gone further and actually associated the content of patients' dreams to changes in depression symptoms over time. Most famously, pioneering sleep researcher Rosalind Cartwright, pictured here, who unfortunately passed away this year, she conducted a series of studies over the years looking at depression symptoms and dream content in women who had recently gone through a divorce, and among Cartwright's observations were that remission from depression one year following divorce was robustly predicted by a couple of features of dreaming, and those were, one, patients whose depression was in remission a year later tended to show dream affect that became more positive across the night, meaning that earlier in the night there was more negative emotion, less positive emotion, but dreams that occurred in the late morning as participants moved toward wakefulness showed increasingly positive emotion.
And that's a pattern that's been described in healthy controls as well in that it's typical for the affect, the emotions reported in dreams, to progressively become more positive across the night, and so that pattern predicted remission from depression. But also remission from depression was seen in a couple of studies to be predicted by actually dreaming of the ex-spouse more frequently or in a particular way that connected the ex-spouse to other memories.
So Cartwright and others have interpreted these findings in light of an emotion adaptation theory of the function of sleep, which Dr. Poe actually reviewed some of the mechanistic possibilities here, and according to this kind of emotion adaptation theory of sleep, dreams as well as sleep are proposed to help us process negative emotional experiences across the course of a night.
So based on these kinds of observations about dreaming, it's been speculated that depression is associated with the failure of this emotional memory processing function of sleep, and in fact that the content of dreams may reveal or reflect this failure of emotional memory processing across the course of a night.
So to summarize, I'd emphasize just a few points. One is that we know that dreaming seems to be influenced by memory consolidation processes that are happening in the sleeping brain and that there's evidence that consolidation processes and perhaps particularly emotion memory consolidation processes may be impaired in depression, also in some other psychiatric disorders as well, but not too much time to go into that, and that in fact features of dreaming may reflect this impairment of emotional memory processing in psychiatric patients.
Really the upshot that I'd like to leave you with is that the study of dreaming could potentially provide useful information about impaired cognitive and emotional processing in patient populations. So it's a really different level of analysis than examining behavior. It's a really different level of analysis than examining neural mechanisms. But the level of analysis of subjective experience is a meaningful one in this topic that we should be paying attention to, in my opinion.
And that is all I have. Thank you to the researchers and funding sources that have supported this work, including of course NIMH. And I'm now going to pass it over to Dr. Ben Simon.
Agenda Item: NREM, Sleep, and Affective Processing in Healthy Samples
ETI BEN SIMON: Thank you very much. Thank you for getting us back on time as well, Erin. Hi, everyone. I am Eti Ben Simon, and I am very happy to be here and take part in this wonderful workshop.
I want to talk about the studies we have done looking at the effects of sleep loss on social and emotional behavior in humans. We know that sleep loss has a dramatic effect on how humans feel. In two recent meta-analyses, sleep loss, either sleep deprivation or impairments in sleep quality, have been reliably associated with reductions in positive mood and increases in anxiety, and these findings are really in line with what we see clinically as well, where people who have disturbed sleep have more than double the risk to develop depression and anxiety down the line, and patients that have depression or anxiety, 70 to 80 percent of them would report troubles either falling asleep or staying asleep.
And this profile determines the mood following sleep loss has also recently been expanded to include changes in social behavior as well. In a really new meta-analysis pioneered by Melanie Hom, one of our panelists today, sleep loss including insomnia has been associated with increased feelings of loneliness, which is a subjective sense of being socially disconnected from others, and also the mere desire to interact with other people has been shown to be influenced by sleep loss.
So for instance here, a night of sleep deprivation has caused participants to report that they are more likely to want to be alone, and this is in a linear relationship with how tired they were. Here are the impact. And also reduction in their desire to be with friends. So both a sense of social disconnection from others, also social withdrawal, is affected by lack of sleep.
This also holds clinical relevance as loneliness is increasingly recognized as a salient killer, and people who report high levels of loneliness have 45 percent increased risk of all-cause mortality and also as we have heard from Dr. Van Orden yesterday, loneliness can increase suicide ideation and cases of suicide in older adults.
So in our studies, we really wanted to dig deeper into how sleep loss affects or contributes to these changes in both social and emotional behavior. Specifically, we wanted to know what is the neural basis of these social impairments, what changes in the brain that might lead to people becoming more socially withdrawn and more anxious, and also we really wanted to emphasize the causal role of sleep in triggering these impairments.
So often when we have some sort of disruption in social emotional function, we typically have that accompanied by disruptions in sleep, either insomnia or poor sleep quality. But can we also say the opposite? If we take healthy individuals and we disrupt their sleep or take away their sleep, do we see impairments in social emotional function, and if this is true, we can really offer sleep not just as a symptom of disruptive social emotional function, but also as a contributing factor to the development and the maintenance of such disorders?
So this is why most of our studies are done on healthy samples where we take away their sleep and try to examine the social emotional behaviors.
The two studies that I have to go through today that we have done in this context, one looking at the effects of sleep loss on social withdrawal and loneliness, and the other one on anxiety.
I'll start with the first one on social withdrawal, and this is moreover now after we've been through a year of social distancing, but this is actually done before. So the setup for this design was we had healthy participants visit the lab twice, once they were allowed to sleep normally, and the other time they stayed awake for the entire night, so total sleep deprivation. And we then wanted to get a measure of their desire for social interaction with others. We used the social distance task. In this task, the participant, here in blue, and the experimenter is walking towards them, and they have to tell the experimenter when to stop. If they're interested in social interaction, they let the experimenter come in close. And if they're less interested, they ask them to stop farther away.
We also wanted to probe the neural correlates of social approach. So we had participants view videos. I'll show you an example of other individuals that they're not familiar with, walking steadily towards the camera, and this was by gender. And they were watching these videos also indicating their preferred level of social distance.
And then at the end of the session, we also filmed an interview with them.
So I'll start with the behavioral results. We found that after sleep deprivation, here in orange, participants preferred to keep greater social distance from other. In fact, there was a range of 18 to 60 percent greater distance after one night of no sleep relative to the same distance these participants preferred to keep after a night of sleep. And we also have evidence that sleep loss affected ratings of loneliness as well in another study that we did where we tracked people's sleep from night to night. So both social withdrawal and loneliness were affected by lack of sleep.
When we looked at the brain, we mostly focused on two networks that are relevant for social approach. The first you can see here in red is a network that is consisted of sensorimotor regions and is known to monitor social approach. So whenever a human, but not an object, anything of social value comes close, and what we found is that these regions are more sensitive or more active following sleep deprivation relative to a sleep rested state.
The second network we focused on was the theory of mind network that is very relevant for prosocial behavior, and here we found a deactivation, so less activity in this network, following sleep deprivation. Together this profile could suggest that the brain is more sensitive to social approach, while at the same time be less able to consider other people's motivations or desires, overall leading to greater social withdrawal.
Indeed, we found that greater activity in these regions that monitor personal space was associated with greater distance that the participants preferred to keep from others. These findings tell us that the sleep-deprived individual here in red is less interested in social interaction following a night of sleep loss.
We also wanted to probe the other side of the social interaction. Do other people care if they have come into contact with someone who is sleep deprived? Does it change their social behavior as well?
To answer this question, we used the interviews I told you about. We put them online and we asked external judges to look at the videos and tell us what they think about the individuals they see. We didn't mention that some of the videos were taken after a night of no sleep. We only include that later in the analysis.
What we found is that our external judges rated sleep deprived individuals as lonelier and also indicated that they are less interested to socially interact with them or to collaborate with them on a work project. So these findings really emphasize that sleep loss affects both sides of the social interaction, and it could indicate that sleep loss can start off a cycle of social isolation, and social withdrawal, by not allowing the sleep-deprived individual to even reconnect with others even if they wish.
In the second part of my talk, I want to tell you about the second study we've done looking at the effects of connection between sleep loss and anxiety. Here we had a similar design. Participants visited the lab for two sessions, one with no sleep and the other one with normal sleep. That was also recorded. We measured anxiety before and after at each session, and we also had our participants view emotional and mutual videos in the center so we could trigger activity in the limbic process(?).
Behaviorally, what we found, and you will hear more about this effect of sleep loss on anxiety from Dr. Goldstein right after my talk, is a significant increase in anxiety following sleep deprivation, and in fact, 50 percent of our participants reach clinical levels of anxiety after just one night of sleep loss, but then again, they're undergrads at Berkeley. So maybe their baseline levels are already high.
And we then focused on the brain specifically in regions of the medial prefrontal cortex bordering on the anterior cingulate cortex, because there is a lot of evidence showing that activity in the medial prefrontal cortex and specifically the connectivity of the medial prefrontal cortex to the amygdala is impaired in patients with anxiety disorders as well as in individuals that have high levels of anxiety, indicating impairments in emotional regulation.
So looking at the medial prefrontal cortex, we did find a significant decrease following sleep deprivation (indiscernible) the activity in this region following a night of sleep loss, and we also saw that the greater the impairment in medial prefrontal activity the more anxious participants were reporting. So here you can see for instance that larger impairments in medial prefrontal activity coincided with greatest levels of change in anxiety from sleep deprivation to sleep rested.
Supporting a model of impaired emotion regulation, we find impairments in the connectivity between the medial prefrontal cortex and the amygdala following a night of sleep loss. So these findings can support a model in which sleep loss triggers a state of emotion-based regulation and that in return increases anxiety. If this is true, we would expect that sleep, a night of sleep, somehow stops that process from happening and helps us feel less anxious state.
To try to answer this question, we focused on the night of sleep that participants had, and we wanted to know whether we can see any association sensitivity to anxiety that would correlate with any one of the sleep stages that they had, either non-REM or REM, and what we found is that the time they spent in deep non-REM sleep and N3 or N4 was actually the only stage sensitive to the levels of anxiety they felt the next day, having spent more time in deep non-REM sleep was associated with less anxiety the next morning.
And we also found the electrical quality of non-REM characterized by the power in slow wave activity had a similar result. The participants that had greater power in the delta band during non-REM specifically in central and posterior electrodes, which was very interesting, was associated with reduced anxiety the next day. So this could suggest that deep sleep specifically helps restore emotion regulation and, in fact, we found that having more delta power, the same measure of delta power, was also predictive of greater activity in the medial prefrontal cortex the next day.
So overall, we can conclude this part of the study by suggesting that lack of sleep targets the same regions that make us vulnerable to anxiety to begin with, and that would trigger a state of emotion dysregulation, while a night of good sleep and specifically deep sleep can act as a natural anxiolytic and restore mechanisms that are relevant for emotion regulation.
And overall, the message I'd like to leave you with is that sleep shares a causal and bidirectional relationship with how we feel and with our social emotional function. It can affect how we feel about ourself, how we feel about others, also how others feel about us. There is a lot of work that needs to be done, trying to clarify the mechanisms that really support this overall effect on our social emotional function, and also it's very relevant; that's why I am really happy to be part of this workshop, to put sleep on the map as an intervention that can reduce the risk factors to develop these affective disorders and also to prevent future tragedies.
So you may have heard suggestions to keep calm and go to sleep. I would actually rephrase to say that you should go to sleep and that will keep you calmer.
I want to thank my colleagues at the Center for Human Sleep Science, specifically Professor Matthew Walker, for all the support, and I will move my virtual mike over to Dr. Goldstein. Thank you very much.
Agenda Item: Sleep, Mood, and Neuroimaging
ANDREA GOLDSTEIN-PIEKARSKI: Thank you, Eti. I am very honored to come after such a wonderful list of speakers on this topic and hope to kind of expand the discussion, particularly following from Dr. Ben Simon's talk, to give a little bit more of a background as to why sleep disturbance may play a causal role in depression and anxiety or at least what evidence supports that.
In addition to discussing a potential candidate brain mechanism, which is informed both by the neuroimaging literature of sleep literature, but also in depression and anxiety. Then I want to talk about some of the gaps in our understanding, even despite these huge advances that we've made over the past decade in these research realms, as well as talk about a few specific projects that my lab is hoping to address which will help answer some of these unanswered questions.
So Dr. Ben Simon has done a wonderful job already hinting at this very important relationship between sleep and emotional function, and specifically highlighting evidence demonstrating this causal relationship between sleep disturbance and emotional dysregulation that may underlie disorders such as depression and anxiety.
Just elaborating on these findings a bit more than Dr. Simon has already presented, we have a number of different disturbed sleep factors that have been associated with the behavioral and autonomic or physiological markers of emotional dysregulation. This includes difficulty in recognizing emotional expressions, our responses to mild stressors, the way in which we respond or utilize emotion regulation strategies, as well as the autonomic responses themselves.
And this potential causal pathway of sleep disturbance leading to psychological distress is further supported by evidence demonstrating that psychiatric disorders, particularly depression and anxiety, tend to have greater sleep difficulties. Now this perhaps isn't surprising at all, considering that most of these disorders, particularly depression and anxiety, actually include sleep disturbance in part of their diagnostic criteria themselves.
However, not only is it that depression and anxiety have more sleep disturbances, but we've also heard about that sleep disturbance can often precede these other symptoms. For example, this meta-analysis found that having insomnia can increase the odds of developing depression and anxiety disorders by almost three times compared to when insomnia is absent, and we've also heard from Dr. Wamsley today that even day-to-day changes -- I'm sorry, Dr. Zuromski -- that even day-to-day changes in sleep have been associated with a next-day suicidality, with worse sleep quality both objective and subjective, being associated with next-day increases in suicidal ideation and risk.
But perhaps the most convincing evidence of this causal relationship between disturbed sleep and psychological distress is under experimental conditions when we actually take sleep away, and we've also heard from Dr. Ben Simon that even 24 hours of sleep loss can lead to substantial increases in negative affect and mood states, including depression, anxiety, psychosis, stress, and anger, even in individuals who aren't normally experiencing these types of symptoms.
So when taken together, these findings really do suggest that disturbed sleep through a number of different pathways might directly contribute to the development and maintenance of depression, anxiety, as well as related symptoms, which is important considering this current workshop, because these all may play as risk factors for suicidality itself.
Which then leads to the important question of whether or not sleep is an important potential treatment target, and again, we've heard from several talks already hinting at kind of new evidence that intervening in this disrupted sleep cycle can actually improve mood, and these recent findings are also supported by at least one meta-analysis that's looked at CBT-I as a way of improving depressive symptoms, and they did find promising findings in the kind of joint study. So while some of these individual studies may not necessarily have had significance at the individual study level, the meta-analysis found generally medium to large effect sizes for CBT-I improving mood.
And other similar studies have shown the same effects for other wellbeing measures, including anxiety. So in our lab most recently, and in response to the COVID-19 pandemic and its associated increases in insomnia risk factors, we actually launched a study, a very small pilot study, to examine whether an early sleep intervention using a brief remote delivered sleep treatment or CBT-I would not only improve insomnia symptoms in the acute form but also impact the mood-related disturbances that were brought on during the pandemic.
So what we did in the study is we looked at 49 individuals who were experiencing insomnia symptoms with an ISI greater than 10 that had begun or came on during about the pandemic. These individuals were then randomized to receive either a brief four-session cognitive behavioral therapy for insomnia that was delivered over telehealth or to a waitlist control.
We then asked them to repeat the same sleep disturbance mental health outcome measures following the treatments, and what we ended up finding, not surprisingly, was that our CBT-I intervention dramatically reduced insomnia symptoms here in blue for the CBT-I group compared to the control in red, and this was paralleled by a similar decrease in depressive symptoms in the CBT-I group compared to the waitlist.
So at least in this small trial, we do have evidence again that CBT not only can improve kind of the pandemic-induced insomnia symptoms in the acute form but also can potentially improve the depressive symptoms that are coinciding with it.
But perhaps what I'm most excited about is that we found that the change in insomnia symptoms that occurred during the treatment, so how quickly people responded and to the degree which they responded during that treatment phase of those first four sessions, significantly predicted or was associated with the subsequent improvement in depressive symptoms at the posttreatment stage. So these findings are really exciting, because it suggests that insomnia symptom improvement as a result of this early intervention was actually mediating the subsequent improvement in depressive symptoms later on.
So when taking all of these pieces together, it really does seem like there's strong evidence that sleep may actually be a modifiable treatment target, not only for insomnia and related sleep disturbance, but also these negative mood-related pieces as well.
Well, what are the potential brain mechanisms underlying these relationships? We've already heard from a number of speakers kind of highlighting the medial prefrontal cortex and amygdala, this fronto-limbic system, as being important to disturbed sleep, and this was first kind of identified back in 2007 when we found that an acute night of sleep deprivation, just 24 hours, increased amygdala reactivity to negative emotional images while at the same time decreasing the medial prefrontal cortex connectivity.
And this relationship has now also been extended not just to sleep deprivation itself, but also to sleep restriction. For example, sleep restriction was associated with decreased amygdala medial prefrontal cortex connectivity, and importantly, those reductions in connectivity were also associated with greater anxiety. Similarly, these findings have been extended beyond experimental sleep manipulations but also just looking at habitual sleep quality with, again, worse sleep being associated with this hyperactive amygdala and disconnected amygdala with the prefrontal cortex.
But there is good news. So it does seem that sleep may actually help reset these systems, meaning that this may be reversible. So for example, in this study, they found that sleep when compared to wake caused a reduction in amygdala reactivity to the emotional images as well as reestablishing the connections between the medial prefrontal cortex and amygdala.
So an interim summary of the brain findings really does suggest that disturbed sleep across a number of different profiles are all associated with this kind of stereotypical hyperactive amygdala and decreased amygdala medial prefrontal cortex connectivity.
So why is this important within the context of depression and anxiety? Well, it's the same hyperactive and disconnected fronto-limbic profile of responding that's seen when you look at depression and anxiety compared to healthy controls. So poor sleep seems to produce a very similar brain profile that mimics those with anxiety and depression.
But it even goes beyond just looking at the comparison between depression and anxiety to healthy controls. These regions are also seemingly important in the response to treatment in both depression and anxiety. So we found in one study or two studies actually, in two separate years, that looked and found that amygdala activation alone or in combination with early life stress was able to predict treatment outcomes to commonly used pharmacotherapies for depression, and more recently, we found that amygdala activation or the degree in amygdala activation change as a result of a behavioral intervention for depression was mediating the depressive improvements that happened through treatment.
So taking kind of this whole presentation so far together, evidence really does point to the amygdala and its connection with the medial prefrontal cortex as important candidate neural pathway through which sleep disturbance may be contributing to psychological distress. However, despite these advances with these studies, there are still many large gaps in our understanding. For example, while there have been some basic research in understanding the impact of sleep deprivation on emotion processing, using these more advanced neuroimaging and sleep measurements, which again was highlighted in Dr. Ben Simon's talk just before mine, there have been very few studies really to focus on these relationships within clinical populations themselves. They've mostly been done in healthy controls.
Similarly, these studies have tended to be relatively small sample sizes, just due to the expense of scanning and enlarging these largescale studies. Moreover, prior studies have tended to look at just one modality or relationship at a time. So it similarly remains unknown how changes in sleep physiology as a result of treatment specifically, rather than, say, removing sleep, may be interacting with downstream changes in brain functioning that may impact functional outcomes.
So many of these relationships or limitations, the points I'm highlighting now, I find seem to be due to kind of the stark separation between the more basic science emphasis, which focuses on brain/behavior relationships but in healthy samples, and that of the clinical research which includes clinical samples and clinical intervention, but at least in the context of sleep research tends not to include these more advanced neuroimaging or EEG assessments of complete techniques.
So it's really the hope of my lab to help kind of address some of these unanswered questions, particularly by combining these neuroimaging and more advanced measurements of sleep together with clinical trial frameworks to better understand the relationship between sleep disturbance and depression and anxiety across the lifespan. Specifically, my research encompasses understanding how the relationship between sleep disturbance and affective brain function and emotional dysregulation may contribute to the development and maintenance of psychopathology, as well as to identify novel biomarkers of both vulnerability, as well as resilience to the affective consequences of sleep loss.
The other side of my research really focuses on manipulating different parts of this pathway in an effort to understand how restoring sleep might actually lead to psychological recovery, and to understand whether sleep may be acting on specific brain networks that aid in emotion regulation to mediate these improvements, as well as to identify which factors may predict which individuals would benefit solely to a sleep intervention but to improve emotional or psychological distress.
And the ultimate goal of this research really is to develop a mechanistically informed intervention for mental health disorders, but through targeting sleep.
So now I actually want to take a few minutes just to talk about two studies that we're launching in our lab that are kind of a direct test of this new model, and these two studies were recently funded by NIMH. One is in a depressed sample of adults who are also experiencing sleep disturbance, and the second study extends that to include individuals who are also including memory impairment.
These studies really aim to determine whether a sleep intervention, in this case CBT-I, will engage or improve fronto-limbic regulation of emotion, as well as determine whether these treatment-induced improvements in fronto-limbic regulation might be mediating the subsequent improvements in depressive symptoms. And then the third aim is to really determine predictors of response and to identify which factors can be used to identify which individuals would benefit from a sleep intervention in order to treat mood.
So this study, or both of these studies, utilize very similar approaches in which we do deep phenotyping. So we include an overnight 25-electrode array of sleep recording, which the next morning they come into the lab and we do an array of fMRI tasks that focus on both emotion regulation as well as cognitive control. We also conduct structural imaging as well as broad assessments relating to cognitive and current functioning as well as emotion distress and mood. The participants are then randomized to either receive the cognitive behavioral therapy for insomnia or a credible control treatment, before doing the same kind of baseline assessments again after treatment.
The hope for these studies is that by causally manipulating sleep within this mechanistic trial framework, we'll hopefully uncover how different aspects of sleep disturbance may be contributing to depression across these multiple units of analysis, and these results will really help advance our understanding of how sleep disturbance and fronto-limbic brain function or dysfunction may be underlying depressive symptoms, and these results really would be a necessary first step in the development of a sleep-based mechanism-focused preventative strategy and treatment that can be personalized for the individual.
So as an overall conclusion, or the big takeaways that I'd like you to walk away with, is that sleep disturbance really does appear to be a causal factor in the development and maintenance of depression and anxiety, and the flip side of that is it does appear to be a really valuable potential treatment target. Finally, the fronto-limbic dysfunction is a strong candidate neural mechanism underlying this relationship between sleep and mood, but there's still lots of work to do.
On that, I will say thank you and go ahead and exit out of my screenshare here and ask that everyone in the session can turn on their cameras, and we will go into a brief question and answer session.
Agenda Item: Q&A
ANDREA GOLDSTEIN-PIEKARSKI: So we have about ten minutes. So I will just go ahead and jump right in, and I see Gina, you are here. Maybe we'll start with you, because I know that you are balancing multiple things right now.
So we have a great question. Dr. Poe, can you please speak to how depression may fit into the model with the typical profile we see of increased REM percentage?
GINA POE: Yeah, so if your sleep is maladaptive and your REM sleep isn't able to do its job, I would imagine that your homeostatic mechanism is signaling that you need more of it. It's not going to be more good REM sleep, but your body doesn't know that. Also, the increased, the shorter latency to REM sleep might also be due to the fact that your body is signaling a homeostatic need for more of it, and could also be bad news, because that increased -- that sooner REM sleep could come at a time when your locus coeruleus is maybe still active from waking processing of the stressors that you were going through for the day.
Can I also answer Dan Buysse's wonderful question? I'll state it, and that is he asked about in insomnia, spindle density doesn't seem to be reduced, and how does that fit in with the model, and so when we stimulated the locus coeruleus in rats across sleep, actually spindle density did not go down either, but instead, spindle length went down, and so in fact, the body seemed like it was trying to generate spindles, but the spindles were really, really short-lived, because whenever this locus coeruleus fired, it ended the spindles. So I would predict that in insomnia, one of the problems is of course that people wake up a lot out of that transition to REM spindle state. But also that the spindles are less effective, that they are not able to -- and probably more local, not able to be as global in doing the job in the cortex as they're supposed to do.
So I would take a closer look at the amplitude and the length of spindles, because actually after learning, what really happens in the prefrontal cortex, is the spindle length, the long spindles, are the ones that are selectively increased.
ANDREA GOLDSTEIN-PIEKARSKI: We have a question for Dr. Zuromski. So homicide rates are also highest at late night and early morning, undoubtedly due to multiple factors. Do you have any comments vis-à-vis suicide?
KELLY ZUROMSKI: Sure. I thought that was a great point, and I'm happy someone brought that up. We actually do have data from the first study that I presented my case study on people's urges to hurt other people in addition to themselves so we could actually kind of get a little bit closer to looking at the direct comparison for self-harm versus harm toward others, but I think it's a great point. I'm sure that there are a lot of factors that are overlapping there, including what I was talking about impulsivity, self-control, decreasing at night and the associated neural activity there. So it's a great point and something that I'm hoping to look more into specifically even in my projects moving forward.
ANDREA GOLDSTEIN-PIEKARSKI: A development question for Dr. Wamsley. So is dream lucidity related to emotional memory process, i.e., lucid dreaming versus not lucid dreaming?
ERIN WAMSLEY: That is a great question. There has not been a lot of research on lucid dreaming and memory consolidation, because it's just so hard to capture lucid dreaming in the lab. For those who aren't familiar, lucid dreaming just means a person is having a dream and while they are having a dream, they actually have the cognition that this is a dream. They realize that it is a dream, and this is corroborated. This is actually a thing.
But it's an interesting question, because you might have noticed that my work and actually all work really showing that dreaming is related to memory consolidation, it's all correlational. We're just splitting participants into people who happen to dream about what they recently learned, and people who did not happen to report a dream about what they recently learned.
So there are a couple of preliminary small sample studies by Michael Schredl from Germany, looking at, well, what happens if you get someone who is able to lucid dream and can dream about something on command to rehearse a learning task during sleep? Much as might have someone do a mental inventory rehearsal of what they learned, it is possible, although difficult, to tell someone that while you're in the dream, rehearse the finger tapping task.
So he does have one study showing that participants who reported successfully having a dream in which they rehearsed the motor sequence finger tapping task showed more improvement than lucid dreamers who did not rehearse the task. You see the same effect that you might expect from a waking mental imagery rehearsal there. It's not necessarily the same as spontaneously having dream content related to what you recently learned, right? Because in a typical dream about our experimental tasks, the dream itself might not be very similar to veridically rehearsing the exact thing that you need to remember, whereas in these lucid dreams, the participants are actually rehearsing the finger movements from the task. But it's a fascinating methodology that I would like to see explored more.
ANDREA GOLDSTEIN-PIEKARSKI: All right, and one for Dr. Ben Simon. So insomnia has been associated with relatively increased metabolic rate in theory of mind circuit during non-REM sleep. Sleep deprivation seems to have an opposite effect from your work. Might this relate to CBT-I effects in sleep restriction in particular, and insomnia?
ETI BEN SIMON: That’s a wonderful question. I'll just mention that we have been able to replicate the effect of theory of mind in sleep deprivation in a psychosocial task that we've done. It seems to be quite robust, and specifically found that it's the posterior regions of the network, the precuneus and the bilateral TPJ, that are mostly affected by sleep loss, and this is also where we found the slow wave beneficial effects with more central and posterior electrodes.
So it is an interesting question to think whether insomnia patients sort of come into the world with already an overactive network that needs to be toned down, kind of similar to what we think about the beneficial effect of sleep deprivation on depression that perhaps something about that baseline activity is different and then sleep loss restores it, in a sense. We have -- I really think it comes down to understanding the mechanism of what sleep and specifically slow wave in these regions help restore, and we're looking more into that in healthy samples right now.
GINA POE: I have a follow up question for Dr. Ben Simon in that I was wondering if you think that slow waves might -- the improvement due to slow waves, first of all, in my model would probably be due to the fact that you're clearing the brain and then setting up for the subsequent processes, but also it's possible that some of your slow waves are slow oscillations that are coupled with spindles, and that might be slightly different from the delta waves that -- so have any comments?
ETI BEN SIMON: We actually found that if you go below 1 hertz, if you really go to slow oscillations, you find the opposite effect. There was no beneficial effect of slow oscillation, and actually what we're thinking is that perhaps we're seeing a marker of sleep depth, that perhaps sensory disconnection is more efficient and better in these regions during the night, and that could be a correlate of just having more restful rejuvenated sleep.
This is one of the hypotheses we're looking at. Also another one is that perhaps -- and this is also related to the questions on the theory of mind where maybe there's relevance in activating or restoring parasympathetic nodes in the brain such as the precuneus, which is relevant for both sensory disconnection and also for theory of mind. So this is kind of, we're still looking into that, because I think that the slow wave profiles during deep sleep could probably be characterized by maybe different functions when it comes to affect versus memory.
ANDREA GOLDSTEIN-PIEKARSKI: I think we have time for one more question. So this one is to Dr. Zuromski. The time course data you showed indicating that suicide and other types of urges increase throughout the day and seemed to peak close to midnight are fascinating. There's some evidence for seasonality of suicide risk. Have you considered parsing your data by season or seasonal factors, e.g., by daily temperature, to see whether the daily time course is moderated by season or seasonal factors?
KELLY ZUROMSKI: I think that is a great question, and for those who aren't familiar, the seasonality effect you tend to see in suicide is that suicide tends to spike in the late spring/early summer, and that is something that we could look at in our data, both on a group level, because the study has been ongoing for a couple of years now, but also even at the individual level, because we're tracking people for six months. So that's a great point, along with other factors like temperature and things like that, so I appreciate you bringing that up. It's definitely something to think about.
ANDREA GOLDSTEIN-PIEKARSKI: Wonderful. I think we are going to go ahead and give people a little bit of a break before the next session. It looks like we have a break from 10:40 local time, so I guess that would be 1:40 Eastern time, to 1:50. So I want to thank all of my fellow speakers and the wonderful questions and pass you guys off to your break. Thank you all.
SESSION 2: Novel Therapeutic Frameworks and Intervention Development
REBECCA BERNERT: Welcome back, everyone. I couldn’t be more excited to introduce our next session today which is focused on novel therapeutic frameworks and intervention development, and introduce my esteemed colleagues, Dr. Vaughn McCall from Augusta University, Dr. Elizabeth Ballard from NIMH, Dr. Wil Pigeon from the University of Rochester and Canandaigua VA Medical Center, and Dr. Melanie Hom from McLean Hospital/Harvard University.
Agenda Item: Brief Behavioral Sleep Treatment as a Novel Therapeutic Approach to Suicide Prevention: A Mechanisms-Focused Clinical Trial
REBECCA BERNERT: I will be starting us off today by presenting on the use of two brief behavioral insomnia treatments using a mechanisms-focused clinical trial. As a brief overview, I am going to begin with some background on preliminary studies on sleep and suicide from our group, supporting the rationale for the development of the current trials. I will also be speaking briefly about treatment development, design of two parallel suicide prevention clinical trials conducted among high suicide risk civilians and military veterans. These were named iSleep and Project SERVE.
I will be able only to just skim the surface of what we did. I am very excited to share all of these findings today – and across the two trials, but I am looking forward to reviewing first-line findings supporting feasibility, clinical indications of efficacy and using this mechanisms-focused, system-neuroscience approach to inform future analyses and opportunities.
Regarding background, and as I spoke about yesterday, to address a number of important scientific methodological gaps in the literature, we conducted a host of studies following early NIH funding to evaluate sleep as an evidence-based suicide risk factor. Based on these and in review of my State-of-the-Science Recap, we were able to confirm that sleep was an independent risk factor for suicidal behaviors across diverse populations, study designs, measurement techniques and outcome measures.
We also found that, using a convenient sample and a community behavioral insomnia uncontrolled trial, that modification of sleep therapeutically impacts risk, providing strong rationale for the novel treatment target of sleep within a suicide prevention clinical trial. Important to treatment development in these studies, we found three distinct sleep-focused suicide risk factors that continued to come out across studies, and these included insomnia symptoms, nightmare symptoms, as well as sleep variability or variability in sleep timing.
This was also informed by clinical relevance that I talked about yesterday, so I will only review this briefly. Insofar as poor sleep is modifiable, it is non-stigmatizing, it is visible to others in the weeks and months prior to death, and highly treatable, representing a low-risk strategy for prevention of a high-risk outcome - in particular where the brevity of treatment signaled particular promise for accelerated treatments or brief-acting treatments, given that existing interventions currently remain scarce specifically focused on suicide indication and inaccessible to those highest in need, where treatment lengths anywhere from six weeks for pharmacotherapy or 24 months for specific psychosocial interventions, for example, remain alarmingly mismatched to a suicidal crisis - focused on hours, days.
In addition, regarding military relevance, suicide rates have surged in recent years, and stigma is well documented in the military as a barrier - where sleep problems are overrepresented by comparison among military samples, not explained by PTSD; yet predict poor outcomes for these disorders. This provided strong rationale for the current trials, where transdiagnostic aspects suggest a shared underlying neurobiology, as I talked about yesterday, motivating our desire to use a multilevel, mechanism-focused clinical trial to enable investigation of mechanisms and moderators of treatment response alongside efficacy testing and to inform possible etiology and treatment innovation factors that lie at the intersection of risk.
It also, we hope, will provide opportunity to study testable hypotheses across social, cognitive, behavioral and underlying neural circuity or biological or neurobiological domains with regard to anti-suicidal response.
Regarding our study design and methods, I will review our NIH open-label trial first, which was entitled, iSleep. In this study, we developed a multi-component behavioral insomnia treatment drawing on the existing literature - and of course the treatment trials that exist. This was based in three sleep-focused risk factors that we were attempting to address -- insomnia, nightmares and sleep variability -- and it was manualized into a five-session, abbreviated treatment program. This was offered as an enhancement to treatment as usual, requiring current suicidal ideation symptoms, a depression diagnosis and, of course, clinically-significant sleep disturbances.
Our program was called iSleep - Insomnia Treatment for Improved Wellbeing. We began recruitment for this study, and our aims were to develop, manualize and test the feasibility of an integrated sleep intervention for suicidal behaviors as well as to examine indications of response in lowering suicidal symptoms post-treatment along with a host of other variables. Our primary outcomes here were suicidal ideation symptoms. The secondary outcome and exploratory outcomes included sleep disturbances and mood and stress indices.
Prior to launching this study, it is important to point out that we did have to develop an intensive and necessary infrastructure to safely conduct these studies, given how outpatient suicide prevention clinical trials require a complex infrastructure and really unique ethical, procedural and training considerations to support their conduct. In this case, this included development of a comprehensive safety monitoring protocol for data and safety monitoring, and this included emergency assessment, referral procedures. They were firmly anchored in best practices using routinized risk level categorization, clinical decision tree rules, as well as systematic screening, warm transfers, outpatient safety planning and a long list of other safety and training checks.
Regarding recruitment, you can see this was an ambitious endeavor. We used community flyering, clinic-based referrals and print advertising. We generated a total of 590 new contacts for the trial, of which 490 were phone screened in an approximately one-hour interview, which resulted in 59 participants being assessed in laboratory using a four-hour battery of clinical and diagnostic assessments based on inclusion criteria. This generated a total of 35 participants enrolled that were allocated to treatment.
Here we saw incredible acceptance and treatment engagement with those receiving treatment. We saw that 97 percent completed treatment and stayed in the study through post-treatment. The majority also engaged through our three-month assessment.
Regarding treatment and manual development, I won’t have too much time to go into this, but we were interested in using an integrative approach - with sleep-focused suicide risk factors matched to known existing treatments - accelerated or abbreviated in format, and then manualized based on other clinical considerations drawing from my experiences treating suicidal patients and my training in behavioral sleep medicine.
Insomnia was addressed by way of CBT for insomnia using frameworks developed with my partner, Rachel Manberr. Nightmares were addressed using imagery rehearsal treatment developed by Barry Krakow and modified by Anne Germain into a one-session add-on format. And sleep variability was addressed by way of social rhythms therapy, which was of course guided by Ellen Frank, who has pioneered work in this area.
This was placed into a manualized format across five treatment sessions for which I developed session-by-session PowerPoints, homework assignments, and therapist guide sheets. The sample overall had a high degree of comorbidity and severe psychopathology, as we would expect, and showed moderate to severe baseline clinical severity -with relatively severe insomnia symptoms, and with nightmares commonly endorsed, but a portion qualifying for a nightmare disorder in this study.
We conducted planned comparisons to evaluate clinical indications of response, which revealed large significant post-treatment symptom reductions in our primary outcome - suicidal ideation, which was maintained at one and three months’ follow-up. And this was regardless of how we measured it - so this was from -I am presenting the Columbia as well as the Beck Suicide Scale.
A similar pattern of results for insomnia as well as depressive symptoms according to the QIDS. What we saw and were thrilled to see is that - everything - essentially improved -looking at these outcomes across just preliminary analyses - at post-treatment follow-up and follow-up at three months. We also saw very large, anywhere from moderate to primarily large or very large effect sizes, particularly for suicide ideation and insomnia symptoms, and this was regardless of the type of sleep or the measurement using the PSQI or the ISI.
I am going to jump now to our DOD trial - looking at a sham-controlled trial for a military sample using a similar intervention and mechanisms approach. We aimed to develop and test the efficacy of an integrated intervention for suicidal behaviors here within a sham-controlled clinical trial and to examine indications of efficacy relative to the act of control, which in this case was desensitization treatment for insomnia, which was renamed in this study, but which had previously been found to be a suitable control in past insomnia trials.
We looked at outcomes, of course, across the same format in symptom measures. This study was called Project SERVE, which stood for Sleep Enhancement for Returning Veterans. We used a similar treatment development approach here, but to abbreviate it further, it combined only CBTI and IRT into four treatment sessions. This was manualized in both individual and group formats into session-by-session PowerPoints, homework, and guide sheets - across both treatments and, of course, matched by therapist contact in treatment, homework assignments, and passage of time. All assessments of suicide ideation were likewise conducted by a clinician blinded to treatment assignment.
We thought that the iSleep study was going to be our most ambitious, but this was, by comparison, truly Herculean, and it required comprehensive recruitment using strategies with an array of partnerships, military health agencies regionally, nationally, hospitals and so forth - CalVet, Wounded Warriors and all kinds of other partnerships. But it generated a total, in a very short period of time, 753 new contacts over 400 screenings, which resulted in 112 completed full battery eligibility assessments. A total of 77 veterans were randomized and allocated treatment, similar to iSleep.
We saw incredible retention across this study and no differential dropout across groups. Seventy individuals, due to early discontinuation prior to treatment, received the intervention, either the active control or the active treatment. Early discontinuation occurred after one session only in one case; whereas, retention was otherwise 100 percent to post-treatment. We were really excited to see this and be able to test this.
We also saw good adherence by the majority of individuals receiving the full treatment - over 75 percent, and almost 90 percent receiving three out of four treatment sessions.
Treatment occurred across four 90-minute treatment sessions using this approach. By the way, the active treatment was offered to those assigned to the active controls following unblinding at the end of the study. As I said, this was uniquely matched for therapist contact, treatment duration, and homework assignments with blinded assessment throughout.
In terms of our results, we saw moderate to severe baseline clinical severity in veterans across service branches and eras of service. Based on plant comparisons, we saw similar reductions across the full sample for suicide ideation measures, as well as other secondary and exploratory outcome measures. These effects were extremely large, making it difficult to see comparisons between active treatment and our sham control, but we did find significant differences according to a brief symptom measure, called the DSSI for suicide symptoms, relative to control. We also saw comparisons to control for insomnia and for sleep quality symptoms in the expected direction, suggesting integrity of the intervention.
We saw large improvement, as well, post-treatment in mood disturbances according to the QIDS, and similar to the iSleep trial really across the board, and this also was extended to PTSD or PCL symptoms in the sample, and that is important because we had a high degree of PTSD, which we are excited to look at in future analyses.
Importantly, the effect sizes were large to extremely large, and for both trials, we are interested to look in the future at sleep variability, which of course was a primary focus here. We had actigraphy throughout the study and sleep diaries and so forth, so we will be looking at that post-treatment. We will be able to compare that to sleep variability reductions, as well, following or in comparison with iSleep, which of course had this formal treatment component focused on circadian dysregulation and sleep, eating and other social rhythms.
In closing, our trial we found very quickly and here reported - supported feasibility testing of a novel manualized insomnia treatment for suicidal behaviors in parallel conducted clinical trials. This also included feasibility of testing of a sham-controlled active treatment arm, which was incredibly complex, but we hope provides application to future work and studies.
It also supported safe outpatient conduct among two high-risk populations, the majority of which were unmedicated, without antidepressants, as well as acceptance and retention, unusually high rates of treatment engagement and supporting acceptability of the intervention. We also saw exceptional tolerability and safety, and this is so meaningful in the context of this discussion. For the full duration of both trials, not a single adverse event was reported, despite two incredibly high-risk groups.
We showed therapeutic impact on primary and secondary outcomes and exploratory outcomes through follow-up, and observed anywhere from large to extremely large effect sizes for suicide ideation and insomnia measures, and some differences compared to control for these measures, but also some degree of placebo effects. And so we are really excited and interested to examine this more, given some elegant work by Marta Pecina, and some of the open placebo trials for depression-inducing robust neurobiological changes and really conceptualized as a form of self-healing.
We have the ability to look at this in a number of ways here. We had imaging studies, biological sampling data, but also measures of hope, perceived treatment, treatment engagement, and we are interested to see how just being a part of a sleep-only treatment actually enhances readiness for treatment, possibly as a gateway to treatment or also as a form of resiliency building.
Regarding novelty, the studies were unique in a couple of ways. I won’t go over this, but we are very much looking forward to looking at future mechanisms underlying neural circuits, for example, in this imaging sub-study that we conducted pre and post-treatment with in-scan implicit mood regulation measures, explicit mood regulation measures and a host of other variables. We are also able to look at, because of the active sham control, some of the results that I just recently described.
And there is also novelty in comparing many of these concepts in parallel between our high-risk civilian and military samples to evaluate maybe specificity regarding some of the effects according to different types of risk that may be veteran-specific or not. And then possible translation and adaptation to other age groups and high-risk samples - for example, having built this infrastructure and now knowing its translational significance and portability to other formats and settings.
In closing, our findings support feasibility of the current approach, which addresses a number of gaps - where interventions remain scarce or unacceptable, and where we hope to evaluate a host of unique mechanisms and moderators in the future - at the intersection of risk - and anti-suicidal response to understand the use of a sleep intervention in these and other high-risk populations.
I would just like to end by thanking all of our collaborators and sponsors, consultants and trainees that made these studies possible. These were very, very difficult to do. I can’t possibly thank them all here, but all told, we hired, trained or collaborated with over 100 different personnel - wonderful, amazing people, to build out these programs and execute both trials simultaneously, which we couldn’t have done without the incredible work of my study coordinators in particular, who were the lifeblood of these studies and, of course, the participants taking part in them. I couldn’t be more grateful.
And last, I just want to end with a thank you to the mentors over the years that have supported my interest in these areas and bridging these two highly different fields and their encouragement along the way. Thank you so much for your attention. I am going to sign off and hand us over to our next speaker, Dr. Vaughn McCall.
Agenda Item: Pharmacological Sleep Treatments for Suicidal Behaviors
W. VAUGHN MCCALL: Thank you, Dr. Bernert. That was a terrific and fascinating presentation. I want to thank you and Dr. Leitman for the opportunity to speak to this group.
I am going to just jump right in and first tell you what I am going to be talking about. This is going to be a pharmacological intervention for sleep as a means of moderating suicidal ideation. This is a nice pairing with the psychological interventions that Dr. Bernert was just discussing. And I have a couple of disclosures here.
The last couple of days we have been talking a lot about mechanistic aspects of why there is a relationship between sleep problems, insomnia specifically, and suicide. This encouraged me a few years back to embark upon intervention trials for insomnia in patients who were at risk for suicide, and we defined this principally as people who were experiencing active suicidal ideation.
I want to tell you about two trials. The first is a trial of Prazosin, which is an alpha-blocker sometimes used to treat nightmares and also daytime symptoms of hypervigilance and hyperarousal in patients with PTSD. Murray Raskind really embarked on this in a big way, and we borrowed heavily from his team’s work in designing this study.
I want to acknowledge that authoritative bodies such as the American College of Physicians and the American Academy of Sleep Medicine in general support non-pharmacologic treatments first, similar to what Dr. Bernert was discussing, but I think we all on this call recognize that there is a dearth of accomplished therapists who can deliver a non-pharmacologic treatment even at a rudimentary level. We need to do better in recruiting therapists to do that or help patients learn to do it for themselves.
And because of the lack of availability, or at least widespread access to non-pharmacologic therapies, I feel that there will be a continued need to educate physicians and other prescribing providers on what they can do pharmacologically to treat sleep problems in patients who are thought to be at risk for suicide. And we are first going to talk about Prazosin.
This study was done at the Medical College of Georgia, and I want to recognize all the co-authors that participated in this. This was a single-site study and was intended to be a pilot for, hopefully, what would have been bigger work, and here is the design. We recruited persons who had post-traumatic stress disorder who had nightmares and were expressing suicidal ideation after a baseline phase shown here. I am only going to show you the psychometrics; I am not going to get into the salivary amylase and cortisol as biomarkers for this study today.
As Dr. Bernert was saying, one of the challenges in doing clinical trials with patients who are at risk for suicide is it makes it ethically somewhat difficult to offer them no treatment at all, so we felt compelled that all of these adult patients needed to be on at least an antidepressant. These were particularly SSRIs if they had major depressive disorder. If they came to us with bipolar disorder, we allowed them to stay on a mood stabilizer.
So this was an add-on study. Patients came to us already on stable doses of SSRIs or mood stabilizers, and then we added on either placebo or Prazosin at bedtime. And given their acuity, we decided to see them every week for eight weeks and so the burden of time pressure on the participants was fairly high.
This is Dr. Raskind’s published treatment schedule for using Prazosin and we will just focus on the men because they are right here. What I want you to see is both morning dosing and bedtime dosing. As a proof-of-concept we wanted to demonstrate that simply dosing at bedtime in an effort to affect and reduce nightmares would be enough, and so we did not give any morning doses. These patients got only Prazosin or placebo at bedtime as their add-on; they did not get a morning dose. The treatment schedule here is actually very aggressive. When people see this for the first time it usually sets them back on their heels a bit because these doses are reasonably high.
These were middle-age folk, mostly women; 80 percent had major depressive disorder, 20 percent bipolar disorder. But if you understand something about the CAPS and the PCL, I think you would agree this is a sample that is highly burdened with a great degree of PTSD symptoms. This is the disturbing dreams and nightmare severity index. It also shows a high degree of nightmares, insomnia is high, depression is high. The scale for suicide ideation is moderate range, but, all in all, this was sort of a severe group, and that proved itself pretty quickly. We only had 20. Our goal was to have 10 in each arm; 10 in the Prazosin add-on and 10 in the placebo add-on, but out of those 20 two participants ended up with emergency psychiatric hospitalization.
The next study I am going to show you in a minute, this didn’t happen. But this group was much more severely ill and it occurred to me that having PTSD and major depression with nightmares and suicidality all together is a particularly explosive combination.
The doses we got to with the Prazosin, again, these were bedtime doses and they were moderately high, so the patients in fact got to the target dose we were expecting them to get to, but no difference between Prazosin versus placebo dosing at bedtime.
This is the scale for suicide ideation scores. Similar to what was just said in the prior presentation, it is important to have a control group because if we had given Prazosin alone, that is the dotted lines, and only followed these people for, say, five weeks or short studied for five weeks, you would say oh, my goodness, Prazosin looks particularly effective in helping mitigate suicidal ideation. But now when we layer on the placebo group you can see that they are doing equally well, and, in fact, there is no statistically significant difference over time between the two groups.
What was even more disturbing was when you look at the nightmare scales where a higher score is worse, we actually had worse outcomes rather than better outcomes in reported nightmares, and this became more dramatic as the study went on and the doses went higher and higher. Again, without the control group, if we only looked at Prazosin for a period of a number of weeks, you might be led to think that Prazosin was effective for nightmares in this group when in fact it was not.
The insomnia severity index also showed improvement in both groups, but it was significantly better than the placebo group.
Again, several things we learned from this. Number one is patients with PTSD, major depression, nightmares and expressed suicide ideation are in fact a high-risk group to take care of in an outpatient setting, needing to have safety measures in place. The second is the importance of having that control group because otherwise it’s really impossible to know if you are accomplishing anything or not. We were totally shocked that Prazosin was actually worse rather than better in this particular setting. In the Q&A, if there is interest, I can share with you why I think they were worse rather than better.
The next study I am going to report is the REST-IT study. In this study we had 103 outpatients. These, again, were adults that had major depression, they all had insomnia and they were free of psychotropics at the beginning of the study. We then randomized them to start a placebo versus open and extended release -- this is a hypnotic drug -- or they got open label SSRIs. SSRIs are open label. And then they got blind study drug, either the hypnotic or the placebo, at bedtime followed for eight weeks.
Again, I want to acknowledge all my colleagues who were PIs with me on this, in particular, Ruth Benca, who started off at University of Wisconsin Madison when the study was kicked off, and Andy Krystal who was at Duke at the time and since this study began everybody has moved so I won’t go into everybody’s location at the moment.
Here is the REST-IT design with sleep apnea screening up front. We excluded people with sleep apnea; we don’t want to give them sleeping pills, generally speaking. Everybody got open label SSRI, typically Fluoxetine, and again you can see the randomization, and we followed them up and made sure they had warm hand-offs and were given to somebody to look after them when they left the study.
These are middle-aged folk, mostly women. We were very happy. We had a healthy representation of non-Caucasians, almost 40 percent were not Caucasian. Moderate to severe levels of illness. I also do work in electro-convulsive therapy and can tell you that this level of depression severity would have earned them randomization in the ECT study without a problem. Moderate to severe levels of insomnia and the scale for suicide ideation, again, was moderate to severe range. About one-third of them had had a prior suicide attempt.
It is important to say that I would encourage us all to be brave and consider doing controlled trials with patients who are at risk for suicide. The idea that they are not going to be good patients, they are not going to follow through was definitely not true. Ninety percent of all scheduled visits were completed; 86 percent completed all their visits in the Zolpidem group, and 73 percent of placebo patients completed all their visits, and they were very good about taking their medication.
So, all in all, I would hold up this sample of patients and say they were as good as any depression sample that you would find. So, worrying about patients’ non-adherence was not a problem.
So, what did we find? We had to show that the hypnotic was working as expected and that it was superior to placebo. Again, all these patients are on an SSRI, and some folks listening today may be surprised that just an SSRI alone with placebo add-on, those patients showed their insomnia was getting better, too. Of course, there was a superior response in those that were assigned the Zolpidem.
At this point, the study is over and the study medication is withdrawn. It’s sort of interesting also to see that what is a superior anti-insomnia effect with Zolpidem persists, even into the follow-up phase. I am not going to go into great detail about why that might be but simply to say it appears that there is, at a very minimum, no problem with massive rebound from insomnia medication when we stopped them at the end of the eight weeks.
And then the outcome which will be of greatest interest to the audience today is looking at the suicidal ideation portion of the Columbia Suicide Severity Rating Scale. We saw an advantage for those who were assigned to Zolpidem over the eight weeks of treatment.
I am going to finish just by saying that in terms of looking at the insomnia-suicide risk relationship as transdiagnostic -- and we have had a lot of discussion on depressive disorders -- does this also extend into psychotic disorders? My colleague here at MCG, Brian Miller, and I have been writing a series of papers showing, in fact, that this does appear to be true - that the appearance of insomnia should be a warning sign about suicide risk in patients with psychotic disorders.
But it also draws our attention back to the InterSePT study published several years ago by Herb Meltzer and his group. If you aren’t familiar with this study you really should read it. This looked at patients with primary psychotic disorders, who were all at risk for suicide, either by a prior suicide attempt or with present suicidal ideation, and randomized them to clozapine versus olanzapine and found that clozapine produced a reduction in suicidal behavior over the period of follow-up.
What is fascinating and which the authors never went into but which we sort of dug into ourselves a little bit - is that those assigned to clozapine not only had a better outcome in terms of reduced suicidal ideation, they were sleeping better than the olanzapine group and had fewer adverse events related to complaints of insomnia. They also took fewer sleeping pills. So, better sleep with less sleeping pills and better suicide outcomes in terms of less suicidal behavior.
That got us thinking, well, is this true beyond just clozapine, so we have this study which is under review at the Journal of Clinical Sleep Medicine. Perhaps one of you is a reviewer and will look at it kindly. We searched the publicly available FDA adverse event reporting system and wanted to see could we come up with any sort of more global opinion about a relationship between sleep and antipsychotics and suicide risk in persons with psychotic disorders.
In this first figure, using clozapine as the benchmark for the greatest anti-suicide effect, what you see plotted here are results straight from the Food and Drug Administration showing the reported odds ratio for having suicidal thoughts or behaviors here versus the likelihood of complaining of insomnia here, and look at this regression line. It looks like those medicines which have the least effect on sleep, or to frame it more properly, the most likely complaints of insomnia, also had the highest risk of complaints related to suicidal behavior or thoughts, which is fascinating.
When we look at the next slide at mood stabilizers, we recognized that lithium, besides clozapine and the psychiatric drug most often cited as having an anti-suicide effect, when we select lithium as the reference and compare lithium to other so-called mood stabilizers that are FDA-approved for such and plot the regression line for reports of insomnia versus reports of suicidal ideation and behavior. Of course, this is leveraged tightly by this particular drug here. But it makes you wonder, it makes you think that perhaps there is a relationship between insomnia and suicide risk that extends beyond mood disorders, extends beyond PTSD all the way to psychotic disorders, and perhaps paying attention to sleep and maximizing sleep in patients with psychotic disorders would be a smart thing to do.
In summary, I will just say that, first, we are seeing that pharmacological clinical trials in patients with suicidal ideation are feasible. You can carry them out reasonably safely. The persons that participated in these trials are good patients and they do what they are supposed to do and they adhere to the treatment protocol.
We saw that failure to impact sleep, as was the case with prazosin, also failed to do anything about suicidal ideation in that group. In comparison, the add-on zolpidem study in depressives did show an advantage for the sleeping pill in mitigating suicidal ideation. And again, just to repeat the idea that this insomnia-suicide relationship may in fact be transdiagnostic.
Going forward, what can we expect in terms of the amount of work required? In our little study we took 100 patients to show a difference between our two interventions, placebo versus zolpidem. In the InterSePT study, which is the clozapine versus olanzapine study in psychotic patients, it took 1,000 patients to show a difference in the outcome of suicidal attempt, suicidal behavior, so, moving from ideation to attempts.
It has been estimated that if we insisted on powering clinical trials to detect difference in suicide death, it would be 10,000. I don’t know if that is entirely true but it sounds about right, and I think it raises questions for our field regarding are we going to be happy with clinical trials designed mostly around suicidal ideation. Are we going to insist upon looking at suicidal behavior, and if we insist upon suicide death we need to have a long discussion about how that is going to happen.
I would like to introduce Dr. Elizabeth Ballard for the next talk and thank you for your attention.
Agenda Item: Sleep, Ketamine Treatment, and Suicidal Behaviors
ELIZABETH BALLARD: Thank you so much. I am so excited and honored to be on this panel, in particular just hearing all of these conversations coming from different perspectives on clinical trials of suicide research. I am really looking forward to the discussion particularly about placebo effects in suicide trials. It has been great so far. Special thanks to Drs. Leitman and Bernert for organizing.
I am going to switch gears a little bit and talk about the relationship between ketamine and sleep. I am going to be talking about the off-label use of a medication. I am a federal employee. I do not have conflicts to disclose, but I will state that any opinions expressed are my own and do not represent the opinions of the United States Government.
What I’m going to do first is give a very broad overview introduction to ketamine and the research around that, specifically as related to suicide. Then I am going to drill down to findings related to ketamine and sleep, specifically ketamine and slow wave sleep, ketamine and risk actigraphy and then specifically suicide ideation markers of response to ketamine.
Before I go any further, I want to be very clear that I am going to be discussing today intravenous ketamine as compared to intranasal esketamine. Esketamine is developed and studied through Janssen and it was recently FDA approved for the treatment of treatment-resistant depression. It has a slightly different formulation than ketamine. I am happy to answer questions about that later although I don’t specifically study that. So, all of what I am going to be discussing right now is intravenous ketamine which has very longstanding use in especially the anesthesia community.
It is a dissociative anesthetic and analgesic. It has been used in surgery for decades. It is something very familiar to anesthesiologists. At lower doses it does have psychotomimetic properties and, specifically, dissociative side effects and usually derealization and other sort of unusual experiences. At higher doses you can have more severe experiences. If you have heard of something called a K-hole, that is related to ketamine and it has led to its abuse on the street as a club drug. It is very rare that somebody only abuses ketamine or is only dependent on ketamine. It is usually done in a polysubstance fashion.
As an interesting note, ketamine was studied in the 1990s as a potential model of psychosis and schizophrenia, so, given to individuals of healthy volunteers to understand the effects of ketamine on cognition. However, the field very much changed in 2000 when it was first described as having antidepressant effects, and so those are much lower than you would see in surgery, on the street, sub-anesthetic doses .5 milligrams per kilogram, often administered intravenously over 40 minutes. That was first described in 2000 by Berman and colleagues.
This is a pretty broad graphic of the potential neurobiological mechanisms of ketamine. I am not going to go into this in great detail; I want to make a few key points. First, this very much has a very distinct mechanism from other antidepressants such as SSRIs. Ketamine is a glutamatergic modulator. We believe that the antidepressant effects are most related to its NMDA receptor antagonist and potentially AMPA throughput, so again, both thinking about the post-synaptic level.
You can see on this graphic that ketamine is what we call a dirty drug, that it works at multiple levels, and so figuring out precisely what ketamine’s effects are on antidepressant effects is still an open topic of debate. You can see specifically its relationship to AMPA and NMDA receptor antagonist. We can also see roles related to astrocytes and GABAergic interneurons.
Why we are here to talk about ketamine is its potential antidepressant and potentially anti-suicidal ideation effects. I work for Dr. Carlo Zarate who was the first person to describe in a randomized clinical trial the effects of ketamine on depression. In contrast to typical antidepressants where we have weeks to months to take effect, ketamine’s effects are rapid, within minutes to hours, and you really see oftentimes the largest effects at one day.
Across the top you have trials with major depressive disorder and bipolar disorder. We have also looked at specific symptoms related to suicidal ideation as well as anhedonia. Ketamine has also been evaluated in many other diagnoses including PTSD, OCD and related to motivational enhancement in substance dependence. Again, for all of these things when you do see effects they happen quite quickly and they are transient, usually up to one week, maybe a significant effect at a week or two.
Part of why I am here talking to you today on the suicide-specific panel are the effects on suicidal ideation. This was a meta-analysis at that point of the published literature. This was a wide range of clinical trials in which people just happened to have suicidal thoughts in a depression or bipolar or PTSD study, as well as a number of studies that have been conducted specifically recruiting for suicidal populations.
What we see when we look at this individual-level meta-analysis is the effect of a single dose of ketamine on suicidal ideation. Again, you see the largest effect size at day one, but you also see potential significant effects out to Day 7. And when we were able to adjust for the effects of depression we still potentially see these effects of ketamine on suicidal thoughts.
That was my general overview of ketamine and now I will talk a little bit about what we have done with sleep. Within the group at NIH Intramural Program, we are really in a fantastic place for research in that we are able to do these trials but then also integrate multi-modal techniques. So, in addition to the sleep studies and the risk actigraphy I am going to present, we do fMRI, MEG, plasma markers, clinical ratings and neurocognitive assessments to really get a full picture of what happens when somebody has one of these rapid-acting interventions and what is going on sort of minute-to-minute at the brain level.
First, I am going to present work from Wally Duncan in our group, who is particularly interested in the relationship between ketamine and sleep. We found that it is associated with changes in brain-derived neurotropic factor, plasma BDNF, at 230 minutes. He was also able to show ketamine was associated with increased total sleep time, increased slow wave sleep, increased REM sleep, decreased Stage 1, Stage 2 REM latency and wakefulness.
We already had a fantastic introduction and discussion of slow wave sleep with Dr. Poe, so I won’t belabor that more specifically, but we know the importance of slow wave sleep and we know oftentimes it is impaired in individuals with depression. What he was able to show was that we have increased slow wave activity in the first non-REM episode after ketamine administration.
Looking at this figure the white boxes represent baseline sleep, the black boxes represent the night after ketamine, and the grey boxes represent two nights after ketamine, all of this done through polysomnography. Again, you see this potential change only in the first non-REM episode on the night after ketamine. Then looking further, connecting the sleep markers with the plasma markers, the change in BDNF was correlated with the change in slow wave activity.
As another type of marker looking at the delta sleep ratio, which is the ratio of slow wave activity between the first two non-REM sleep episodes, he found that this delta sleep ratio predicted change in depression at Day 1, so 24 hours after the ketamine administration. Specifically, when you divide it up into equal to one or over one, individuals with the lower delta sleep ratio demonstrated a greater and sustained clinical improvement.
Obviously, there is a lot to unpack, a lot to explore further, but I think it is an important finding that we not only see these sort of rapid changes in depression with these interventions like ketamine, but we are seeing potentially these rapid changes in sleep processes as well.
Moving on, I am going to talk a little bit about actigraphy. Again, this actigraphy data is probably going to be slightly different than data that has been presented across this panel. Because of the rapid effects of ketamine, we are looking at day-by-day changes. You can’t necessarily aggregate over a week or two as is typical with actigraphy data.
Again, this is Wally Duncan’s work and he took a curve-fitting approach, so, to interpret a lot of this data it is based on metrics such as the central circadian value or mesor, amplitude of the curve, as well as the phase, which way it has shifted, advanced. Overall, you see after ketamine administration a decrease in the circadian value when you compare ketamine to placebo at one day after administration. If you look at baseline we find that responders actually were phase advanced and had a lower central circadian value, so you see potential circadian markers of response to ketamine.
Then when you specifically look at Day 1 or Day 3, at Day 1, responders were phase advanced compared to non-responders. If you remember, they were also phase-advanced at baseline as well so this wasn’t necessarily a change from that. But by Day 3, responders actually now have higher amplitude and mesor as compared to non-responders. So again, a lot to unpack here and understand but really reiterates that you can potentially see these rapid changes even in circadian activity day by day, and they might change over time. It is not constant from Day 1 to Day 3. After this intervention you might still see day-by-day changes.
Lastly, with the time I have left I am going to talk about suicide ideation markers to ketamine. I am not going to belabor this point. We have talked a lot over the last two days about the relationship of suicide and sleep. This is a study that I did looking at sleep studies of depressed individuals who ended up reporting suicidal thoughts the next day. For these figures across the X axis you have time, across the Y axis you have minutes awake.
On the left you have individuals who had suicidal thoughts the next day and on the right you have individuals who did not. If you put this all into the model, significant time awake at the 4:00 a.m. hour predicted suicidal thoughts the next day even when controlling for depression diagnosis, gender and age. But more relevant to this is what happens after ketamine. On the left you have individuals with continued suicidal thoughts after ketamine. Those are the individuals who had suicidal thoughts, received ketamine, went to sleep that night, woke up the next morning and still had suicidal thoughts.
On the right you have patients without suicidal thoughts after ketamine. They had suicidal thoughts at baseline, received ketamine, went to sleep that night, woke up and had absolutely no suicidal thoughts the next morning. What you see is this potential normalization, sort of this relationship with suicidal thoughts and improvement with sleep, really sort of providing this interesting real-time marker of the relationship of suicidal thoughts to nocturnal wakefulness.
Lastly, what we are working on now is digging a little bit deeper into the brain. I have been working with our statistician, Dr. Dede Greenstein, to focus on alpha and beta power in a multilevel functional principal component analysis. Overall, what this means is it is similar to the principal components analysis that you might be used to with sort of an eigenvalue. You have an eigenfunction of power over time.
If you look at what we did in terms of alpha and beta due to their relationship with insomnia, hyperarousal and wakefulness, overall you can see these eigenfunctions. The first eigenfunction really represents a constant shift over time in most of the variants, and the other two represent oscillatory patterns. So you see this for both alpha and beta.
Again, the first PC score is usually most closely associated with traditional polysomnography metrics while these oscillations represent potentially something different. What we were able to do is correlate these with suicidal thoughts and you see potentially the strongest correlation between beta PC2, which I have a little arrow next to, showing again these oscillations in beta power as related to next-day suicidal thoughts, again suggestive of disrupted sleep.
That is it for my talk. I hope I have convinced you that ketamine represents an interesting clinical intervention but also a research tool to understand the relationship between depression, sleep and suicide risk. Right now we are almost to the minute working on evaluating these alpha and beta activities after ketamine administration as a potential marker of antidepressant and suicide response, but another direction also including understanding the interactions between these sleep markers and the other data we have collected including fMRI, MEG plasma markers and genetics.
This work cannot be done alone; it is very much done with a team, so I want to thank everybody involved, particularly my PI, Dr. Carlo Zarate.
I am going to give it over to Dr. Pigeon to take it from here.
Agenda Item: CBT Insomnia Interventions for Depression and Suicidal Behaviors
WILFRED PIGEON: Good day, good evening and thank you, Dr. Ballard. I, too, would like to thank NIMH as well as the Co-Chairs of this really great two-day symposium. And especially I would like to recognize Dr. Bernert, whose early work in this area was really a catalyst for those of us working in this sub-field of sleep-suicide research.
I will be speaking today about cognitive behavioral therapy for insomnia primarily. Here are some acknowledgements and financial disclosures, none of which are related to the content that will be provided today, and my government employee disclaimer.
A quick outline, a few scattershot points that I am going to start with that hopefully coalesce to some extent into a rationale or justification for addressing the following two points, which are whether CBT-I for insomnia has any effects on mood or depression, which we have already heard about some, and whether CBT-I has any effect on suicidal ideation from that.
The first scattershot point is related to whether in fact insomnia is truly a risk factor for suicide. There is a good deal of work on sleep disturbance more broadly. The first meta-analysis was published in 2012 so that means the literature review at that point is now a decade old. At that time, sleep disturbance was indeed found to be a risk factor for ideation, attempt and death. In a subset of the studies, a total of 37 for sleep disturbance and something like 14 or 15 for insomnia, they specifically did show an adjusted relative risk similar to sleep disturbance or insomnia being related to subsequent cross-sectional or subsequent ideation, suicide attempt or suicide.
I wanted to highlight three additional meta-analyses that have subsequently been published in different samples. One in patients with psychiatric diagnoses and only those with depression, and a nice meta-analysis of the sleep-suicide relationship in adolescents that came up with similar findings, but slightly different methodologies, odds ratios and relevant risks.
Most importantly though two very recent meta-analyses, one by Liu, et al and one by Harris, et all, both published just last year, and the important feature here is they included only longitudinal studies. The original meta-analysis included both cross-sectional and longitudinal studies. And even by focusing just on longitudinal studies, the research shows that in the past decade there has been a doubling of studies focused on insomnia, sleep disturbances generally, but also those focused on insomnia and suicide outcome relationships that, again, found similar levels of risk contributed by insomnia specifically.
So the answer is yes, at least for me, insomnia really is a risk factor for suicide outcomes.
Turning to treatment of insomnia, as Dr. Vaughn McCall indicated, certainly clinical practice guidelines suggest that we start with CBT-I. Increasingly ,there are more and more providers providing CBT-I. It is still a constraint issue. Nonetheless, that is what we start with as a recommendation echoed by the European Sleep Society, a couple other organizations, the VA, DOD and the most recent clinical practice guidelines calling for CBT-I to be the first line treatment for insomnia.
What is less perhaps noticed or known about outside of the insomnia field is that there is also a very nice clinical practice guideline by the American Academy of Sleep Medicine that Mike Sateia quarterbacked, and I am highlighting one of the key findings, at least for me, that of course not all sedative, hypnotic or sleep medications are the same. In particular, one of the recommendations was that trazodone not be used for insomnia because of the relative imbalance between harm and benefit. Many of you know that trazadone, depending on the year or who does the study, is either the leading or second leading medication prescribed for insomnia in this country.
Not included in the list of harms was suicide attempt, worked on by my colleague, Jill Lavigne at the VA, who has shown in a sample of veterans, about 150,000 veterans, where we used first-incident prescription of zolpidem as the reference group, that when we compared that group to a trazodone-prescribed group, the relative risk of trazodone for subsequent suicide attempt in the year following incident prescription was 60 percent higher than the reference group with zolpidem.
Next scattershot point: there are a number of barriers to addressing the question does CBT-I serve as a protective factor. You have heard about some of them already. SI as a measure has some problems. Yes, it is a risk factor for attempt but it is not a great proxy for attempt. We heard Dr. Vaughn McCall and maybe in Rebecca’s data also probably some moving around of SI relative to control conditions over time, and of course the low base rate problem.
One way to address that issue is to look at medical records, potentially. This is a study that Tod Bishop, my colleague, engaged in with again the VA electronic health record data. The initial sample was 60,000 veterans. This was a sub-analysis of 20,000 veterans who had a suicide attempt in the last year and a sleep disorder on record. I am going to highlight the finding that I want to talk about ever so briefly, and that is that a sleep medicine visit actually was a protective factor for subsequent suicide attempt. So, an 11 percent reduction in risk; not gigantic but fairly solid and, in fact, in comparison to mental health visits that were also protective but at a smaller degree.
We think, and the data certainly more than suggests, that this sleep medicine visit protective effect on subsequent suicide attempt is really largely driven by sleep apnea, not surprisingly.
Which leads us to another problem. Why can’t we address even in electronic health record data whether CBT-I has an effect? One of the issues is that CBT-I in prior years wasn’t even coded, so it’s hard to find. Secondly, even for individuals who are prescribed sleep medication, or a very clear sleep medication, zolpidem, at 6.25 milligrams, less than 20 percent actually have an insomnia diagnosis, so it is difficult to find them as well, which means we were leaving a lot of people off the table. And it is nearly impossible to find a cohort of comparison patients who have insomnia that is untreated.
So, barriers. Michael Grandner, great title to a commentary, “Insomnia in Primary Care: Misreported, Mishandled and Just Plain Missed.” Certainly true not just in primary care but for us in medical records.
Which leads me to the second main component of my talk today, and that is to answer the question whether CBT-I has an effect on mood and/or depression. I am going to just say yes, it does, and provide data from our lab. It is not a suicide-specific study but it’s from another very complex sample, in this case, survivors of interpersonal violence who met criteria for insomnia, PTSD and MDD, so a pretty complex sample like Dr. McCall’s and I suspect other samples in this session.
We were assessing whether CBT-I had an initial effect on the symptoms of insomnia, PTSD and depression, and then subsequently whether the addition of a trauma treatment cognitive processing therapy was going to be enhanced by CBT-I. So, quickly, the design was randomized CBT-I -- in this case, four 60-minute sessions or so compared to just simple phone checking. And in the second phase of the study, everyone, both treatment arms, received cognitive processing therapy. I hope you will just believe me, a pretty severe, pretty complex sample.
Not surprisingly, large between-group effect size of 1.09, so a dramatic effect on insomnia severity, by delivering CBT-I compared to control conditions. Something we have come to expect. I wanted to focus on the depression findings, a similar finding here. A large effect on depression of CBT-I, and a more moderate but still significant effect on PTSD severity as measured by the CAPS.
Recall that then everyone gets cognitive processing therapy, so, across three time points the effect on depression severity is maintained, as is the effect on PTSD severity. Just from our lab, from a complex sample, 110 participants were randomized in the study supported by the NINR.
I am going to quickly now move to this question: does CBT-I have any effect on insomnia. We began to assess this question because of our University of Rochester affiliation with the Center of Excellence for Suicide Prevention which happens to be housed down the road in Canandaigua, New York, where some folks on our team have been investigators for a decade or more.
We assessed this question initially in combat veterans, another complex sample, with a high degree of depression and PTSD and meeting criteria for both MBD and PTSD. It’s an open label trial, pretty simple first look, eight sessions with CBT-I, 15 subjects, and here we just pre and post in this one arm, and again, large effects. Nice reductions in insomnia severity, depression and PTSD severity. And then the point at the bottom: Of these subjects, three endorsed SI at baseline, none at post-treatment. A tiny signal, but led us to pursue this in earnest. Although published in 2016 the study was done in 2010.
The additional literature is pretty sparse with respect to CBT-I effects on SI. Two clinical case series, one by Rachel Manber and one by Mickey Trockel who is the lead author on a VA training case analysis of effects of CBT-I. A single-item SI measurement, a one-third to 50 percent reduction in those who were SI endorsers at baseline after completion of CBT-I.
One large randomized trial done by Helen Christensen out of the Black Dog Institute in Australia. This was an online delivery of digital CBT-I. In this case the platform was Shuti for a depressed sample, one-item SI measure, and there was a significant reduction compared to the control condition at post-treatment, not at six months. But if you look at the mean, a pretty small drop in mean, so who knows if that is meaningful. I think unfortunately in the study they assessed the entire sample and not only those who endorsed SI.
Dr. Bernert shared some of her work. And I also want to note there is another set of unpublished data that I suspect -- I haven’t seen the data, but there’s a trial that Dr. Sara Nazem has I think completed. It’s a pilot RCT in a military sample of the same digital intervention, Shuti. And finally, our approach that I would like to share with you.
Several groups have shown that we can deliver brief CBT-I, and what we did in a pilot trial was deliver very brief CBT-I both in terms of duration and number of sessions. Two 30-minute sessions and two 20-minute sessions compared to treatment as usual. Cutting right to the data here, again, effect on insomnia severity, depression severity and suicide severity in between-group comparison, a not significant, small effect.
Within-group we show an effect, but again, N equals 50 here, not a significant effect on the Columbia intensity subscale which ranges from about 1 to 25 (inaudible).
At this point, what can we say? We can certainly deliver CBT-I, as other groups have shown, to patients endorsing suicidal ideation. The effects on insomnia and depression are retained. I still think the effect on suicidal ideation is unresolved, awaiting larger trials, but there are plenty of signals that you have heard about in this session to suggest that larger trials make sense at this point.
We did show in our N of 50 trial that in fact a change in insomnia severity did mediate CBT’s effect on SI, and we are following up ourselves with VA R&D funds to essentially replicate this study in a large sample. We are now at 100 or so randomized out of 240. Thanks both to folks involved in that R01 study as well as the good folks in Canandaigua, New York. Thank you for your attention. I will now turn this over to Dr. Melanie Hom.
Agenda Item: Reducing Help-Seeking Stigma for Suicide Prevention
MELANIE HOM: Thanks so much. I am really grateful to be here today. I will be talking about reducing help-seeking stigma for suicide prevention using a brief computerized intervention.
I just want to acknowledge that funding for the projects I will be talking about today is from the American Foundation for Suicide Prevention, American Psychological Foundation and the Military Suicide Research Consortium.
Just some brief background. Connection to care has long been considered a key suicide prevention strategy. We have interventions that have been shown to reduce suicide risk and to therapeutically impact suicidal ideation; yet, data consistently show that less than one-half of individuals reporting serious thoughts of suicide in the past year report receiving mental health services during that same timeframe. And among specific populations, that percentage is particularly low.
In response to this lack of connection to care, a number of approaches have been developed to enhance service use among individuals with elevated suicide risk, the three most common being noted below. A number of folks have sought to enhance screening for suicide risk; for instance, often primary care patients are screened for depression or suicidal thoughts when they come in for appointments and then folks who screen positive are then referred to behavioral health services.
Another common approach to enhance service use is gatekeeper training. For instance, training teachers to notice suicide warning signs amongst students and then referring students to mental health services.
A third common approach to enhancing service use is often psychoeducation, the idea that if we are providing folks with info regarding mental health systems, when to seek treatment, what types of treatment are available, that information alone might help to facilitate help-seeking and connection to care.
Although for these approaches there is some promising data suggesting that they can be helpful in connecting a portion of individuals at risk with mental healthcare services, these approaches often don’t target one of the key barriers to care among individuals with elevated suicide risk, and that is help-seeking stigma. We think about this as any sort of negative stigmatizing thoughts about what it might mean to seek help for mental health problems or to connect with mental healthcare services. Help help-seeking, stigma thoughts might be things like seeking help means that I am weak or I shouldn’t need help at all; I should be able to solve problems on my own. I should be able to manage stress on my own.
We know from the literature that stigma, especially help-seeking stigma, is a very potent barrier to care, and these sort of help-seeking stigma thoughts are often very deeply held and highly engrained, so it can be very difficult for an individual, even one who is profoundly suffering, to overcome the stigmatizing thoughts and to seek help and engage with services.
We also know that there is increased help-seeking stigma and stigma concerns among individuals with elevated suicide risk specifically, and that is understandable. In addition to feeling really concerned about what it might mean to seek care, at-risk individuals often struggle with thoughts like it might be very embarrassing to have to share with someone else that I am experiencing thoughts of suicide. I am really concerned that if I go in and share that I have made a suicide attempt that I will be judged, or that process of disclosure will be really difficult.
Given this, my colleagues and I began to wonder can these sort of help-seeking stigma thoughts actually be targeted and intervened upon such that folks with help-seeking stigma thoughts might be more likely to engage with mental healthcare services. There have been a couple of studies, not within suicide prevention but more broadly, where folks have sought to do exactly that, to try to modify these help-seeking stigma thoughts via CBT approaches. So, having folks with mental health problems who are not currently in treatment touching base with a research study clinician, having a brief conversation when that study clinician is eliciting those help-seeking stigma thoughts, and helping participants to challenge those thoughts and restructure them.
The studies have shown that that type of CBT intervention on help-seeking stigma thoughts has led to reductions in help-seeking stigma and increased intention to seek care, so we have a sense that these help-seeking stigma thoughts certainly can be changed.
You may have already anticipated the problem with using this approach more broadly. First of all, it is quite resource-intensive to have that one-on-one contact between at-risk folks and providers. Secondly, if someone is having a lot of help-seeking stigma they are probably going to be unlikely to even want to have that conversation with someone, whether that is in person or over the phone. There is already this barrier, and so that type of one-on-one approach to targeting help-seeking stigma is likely to not really get to those who have the highest levels of help-seeking stigma.
So that led my colleagues and me to think about might we leverage technology then, drawing from the cognitive bias modification, or CBM, literature. There have been a number of CBM interventions developed specifically to target maladaptive thoughts associated with anxiety and depression. These interventions are typically web-based or can be accessed on a mobile phone, and essentially folks are able to practice challenging negative automatic thoughts just on their own without having to interface with a provider. They are able to just sort of conveniently do that on their own time from wherever.
Our thought was if these types of CBM interventions are helpful in targeting anxious and sort of depressed thoughts, we might be able to similarly design a web-based intervention to target help-seeking stigma thoughts. So we designed a web-based intervention with that precise goal.
With this intervention, folks are shown different sentences or statements, things like seeking help means that I am weak, and they are asked to identify the statement as either being true or false and then receive feedback depending on how they replied. When folks in this instance select true, they get this feedback, incorrect, and then we present a more adaptive helpful thought -- seeking help is a sign of strength. If someone selected false, they got that positive reinforcement, correct, and then also the statement that seeking help is a sign of strength.
We designed this intervention so dozens of different statements like this would be presented to individuals in both directions with the idea that folks practice responding to these prompts.
We designed the intervention so that it is quite brief, it involved three 15-minute sessions, just one per week. Again, very accessible, and designed so folks didn’t have to download an app; they could just select a link and log on to a website, so quite convenient. We very much wanted to design this intervention to be cost-effective and scalable and easily disseminable.
We first pilot-tested this intervention in a lower-risk sample just to see how folks were tolerating the intervention and to get a sense of whether it might even impact help-seeking stigma. We recruited 32 young adults with a current psychiatric disorder who had not sought any sort of mental health treatment in the past year. Folks were randomized either into the computerized intervention I just described or a psychoeducation condition where they read materials about mental health symptoms and treatment and answered some comprehension questions just to make sure folks were attending to the information.
At two months follow-up we saw significant reductions in help-seeking stigma in the full sample, but specifically what I will highlight is that two-thirds of those that achieved clinically significant reliable change in help-seeking stigma were specifically in that computerized intervention group.
We also saw that one-fourth of participants had initiated mental health service use at two-months follow-up. We had a relatively small sample so there was no statistically significant difference between the two conditions in terms of initiation of mental health service use, but we did see that 30 percent of folks in the computerized intervention group as compared to 20 percent in the psycho-ed group initiated mental health service use. We took this as sort of a promising signal in terms of the intervention being able to intervene upon help-seeking stigma and potentially being able to enhance actual connection to care.
We decided to extend the study to a higher-risk sample. We recruited 72 young adults, this time all with current suicidal thoughts at baseline. We specifically recruited the sample with elevated help-seeking stigma concerns and not currently in mental health treatment. So we were really trying to find a high-risk sample specifically that might have quite high concerns about seeking out services.
In terms of clinical severity, 15 percent reported a suicide attempt history, 58 percent a history of non-suicidal self-injury, and at study admission 53 percent reported moderate to severe depression symptoms. Again, all had current suicidal thoughts.
This time we had three conditions -- the computerized intervention I described, a placebo intervention where folks just responded to statements about study skills, and then that same psycho-ed condition where folks were reading materials and answering comprehension questions.
We actually just finished data collection for this study so we are still working to quantitatively look at those data, but I did have a chance to sneak a peek at the qualitative feedback so I want to go ahead and provide some info there because I think it is very illuminating and highlights the potential utility of this intervention.
One participant noted, “The statements opened my mind to different ways to perceive therapy and myself,” so showing how just that brief 15-minute intervention once a week might really increase flexibility in thinking about therapy. All these quotes are from folks in that computerized intervention condition.
Someone also noted, “Last week I sometimes found myself correcting the negative thoughts I usually have about myself,” so, just that practice with restructuring thoughts, sort of generalizing outside of the intervention.
Another person noted, “Even though I agreed with all of the statements, having them reasserted seemed to cement them more in my mind.” We had a number of folks who wrote quotes like this, “I liked the praise they gave when you got the answer correct.”
In general, we received positive feedback about the intervention being enjoyable, tolerable, it didn’t feel like too much of a burden, was quite brief. Folks also commented things along the line of this, “I liked that I was able to complete the task at my own leisure,” so just highlighting the convenience of the intervention.
I want to wrap up by talking a little bit about just the relevance to that intersection between sleep and suicide prevention. We can certainly imagine how this intervention might be used generally within suicide prevention to connect at-risk folks with services. But, as folks have already mentioned in this session, sleep problems often represent an entry point to mental healthcare services. As Dr. Bernert noted, sleep problems are arguably less stigmatized than other mental health problems. As most of the folks have mentioned, there is this overlap between sleep problems and suicide risk.
One thought might then be to figure out whether we can maybe include this type of stigma reduction intervention as an adjunct to sleep treatment. If someone is already seeking out care for sleep problems it might be possible to fold in this type of intervention into work that they are already doing within sleep treatment.
This intervention is particularly ideal because it often complements the CBT approaches that are already used often in behavioral sleep medicine, so folks who are completing CBT-I are already practicing challenging catastrophizing thoughts about not getting enough sleep, and so further practice challenging other negative thoughts fits well within that framework.
As I have noted multiple times, this intervention is quite convenient, accessible and cost-effective, so we can imagine how it could be folded into any sort of online treatment portal. It could be something folks are encouraged to complete on their own between sessions, could be easily completed in a waiting room before or after treatment. The idea here is if folks are already interfacing with some form of healthcare services, it may represent this opportunity to increase openness to seeking out mental health treatment longer term, which might be particularly useful for folks with experiences of chronic suicidality.
That is the future direction we are hoping to pursue, so really interested in looking at results from the efficacy-focused trials we have going on now, but ideally we are hoping to expand and look at this intervention across naturalistic treatment settings to see if we are able to enhance connection to care among at-risk individuals more broadly.
Thanks, everyone. I want to go ahead and wrap up by acknowledging my collaborators on this work and grad students and research assistant who have been invaluable in helping with the collection of these data. I will turn things back over to Dr. Bernert.
Agenda Item: Q&A
REBECCA BERNERT: Thank you so much to everyone. We can open it up for a brief question and answer session. We will be moving a number of questions to the full panel discussion in the interest of time but thought we could start out with just a couple of questions beforehand.
First for Dr. McCall, did the severe side effects and difficulty in ongoing treatment with clozapine limit its use regardless of its effectiveness?
W. VAUGHN MCCALL: The person asking the question is referring to Dr. Meltzer’s InterSePT study, and to my recollection that study did not show any unusual limitations regarding clozapine. Of course, clozapine is always associated with the risk of a granulized cytosis, which is a very bad blood problem, but it was no worse in this group of people at suicide risk than in other people. It did not seem to be a limiting factor.
REBECCA BERNERT: The next question goes to Dr. Ballard. In terms of the use of ketamine as a research tool, what do you think are the next steps for ketamine research around sleep?
ELIZABETH BALLARD: Great question. I think there is a lot that we are doing within our group. We have a recent trial in which we gave ketamine to both depressed individuals and healthy volunteers, and interesting from a mood perspective, we are seeing paradoxical effects in the healthy volunteers that actually they get transiently depressed and you see sort of opposing effects on fMRI and MEG as well with these individuals. It would be really interesting to see what is going on with the sleep and ketamine in healthy volunteers, especially because some people have made connections between the neurobiological effects of ketamine and therapeutic sleep deprivation, which we know can induce negative symptoms, but if done correctly and in a safe environment it actually is also associated with transient improved depressive symptoms in individuals with depression.
That is one thing that we are looking at. But I think the next step for the field is really figuring out if we can integrate ketamine with another potential sleep intervention to be able to capitalize on the effects.
REBECCA BERNERT: Great. Thank you, Dr. Ballard. To Dr. Hom, specific populations that you think this type of intervention might be useful for in terms of future work?
MELANIE HOM: I think at this point definitely targeting populations where there are elevated help-seeking stigma concerns and where we are seeing elevated suicide risk would be most optimal.
And I guess to that question, we are actually also pilot testing this intervention in the military service member sample that is at elevated suicide risk, and it certainly represents an optimal sample to look at this intervention because folks are often connecting with physical healthcare services and have ready access to behavioral healthcare services. But there is such a high degree of stigma and lack of connection to care in that population. I am very curious to see how that goes.
I will also note that we have been able to tweak the intervention for that population and consult military stakeholders to identify specific help-seeking stigma thoughts that come up for service members, so we are excited to see how that goes.
REBECCA BERNERT: Thank you. It looks like there’s a question for me that is focused on what would be the most likely next steps to adapt these treatments to different high-risk populations. Great question.
I think the interventions that we have organized could be easily adapted to high-risk youth populations, for example, and high-risk settings including even within hospitalized settings. I think these interventions could be used so much more heavily or even using computerized techniques to make them more accessible to individuals, and building on other works.
And then of course in late life we see similar changes in youth. We see this perfect storm that Dr. Hasler and others talked about in terms of this disconnect and circadian misalignment that’s going on around school start times. And importantly, that has been tied to some reductions in suicidal ideation, by the way, following changes in school start, which is really interesting.
So youth populations, but then also in late life where we also see age-related changes in sleep that make that very attractive for a sleep focus and to evaluate, especially given the high risk nature of suicide among older adults, to be able to address those sleep problems.
I think we have time for one more question and I am interested to hear from Dr. Pigeon. I absolutely love the work focused on sleep visits and conceptualizing sleep as a protective factor. I look forward to discussing all of this more in the full session but I am curious what your thoughts are around how we could use that, whether it is within accessing different CBT-I treatments and rollouts or in using other methods.
WILFRED PIGEON: I don’t mean to over-plug the VA, but it is such a large system and the data sources are so integrated that it’s a great playground. The nice thing now in the last couple of years is you can find, doing some gymnastics, whether a CBT-I has been delivered. There are templates, and the Insomnia Severity Index is delivered, scores are embedded in the notes so you can pull those, again, with some work.
In the last couple of years, we have seen an uptick in the number of other providers like psychologists code insomnia regularly. Others provide not so much, but we have seen an uptick in insomnia being coded by other providers, so that is going to be helpful.
There was a question about at the federal, state and local levels what’s being done to access. I think that is very regional, so, focusing on large health systems. Honestly, just code insomnia and then we can find it and pull out what intervention and how long people have been doing it using these EHR approaches to determine what is protected and what’s not and what works best for what kinds of folks.
REBECCA BERNERT: Right, absolutely. And maybe developing research agendas and policy targets even around that. I could see that being useful - across different healthcare systems and so forth - to maybe advance screening and other opportunities that hopefully we will have opportunity to talk about more in the full discussion.
SESSION 3: Technology Innovation and Digital Medicine in Suicide Prevention: Future Directions
REBECCA BERNERT: I am going to try to keep us on time. With that, I would like to thank everyone on this panel for the phenomenal talks and turn our attention to our next and final panel, which is focused on technology innovation and digital medicine in suicide prevention. We will focus on future directions in this panel.
I am going to introduce our speakers in this panel including Dr. Gregory Simon from Kaiser Permanente and Dr. David Luxton from the University of Washington in Seattle.
With that, I will turn us over to Dr. Simon.
Agenda Item: Artificial Intelligence and Suicide Prevention in Healthcare Systems
GREG SIMON: Thanks very much. I am Greg Simon from Kaiser Permanente Washington Health Research Institute. I will be talking for the next few minutes about our work on machine learning or artificial intelligence to inform suicide prevention in large healthcare systems.
I want to acknowledge this is work that was done by our broader Mental Health Research Network which is funded under a cooperative agreement with NIMH and we also have some support from the Food and Drug Administration for some of this work. We are a very large research team based at KP Washington, some colleagues at University of Washington, and then the other health systems and research centers participating in this particular work at Kaiser Southern California, Northwest, Colorado and Hawaii regions as well as HealthPartners and Henry Ford. So, a very large team that is responsible and I want to make sure to give them the appropriate credit.
The story I will tell started many years ago when our health system started to use the PHQ9 and other brief standard self-report measures routinely in health systems, and the question came up about would these standard questions help us to identify people who are at high risk of suicidal behavior. So we looked, and these are data that we published almost 10 years ago looking at, when people came in for an outpatient visit and filled out a PHQ9 questionnaire, how did their response to Item 9 of the PHQ 9, which asks about thoughts of death or self-harm, how was that related to their risk of suicide attempt or suicide death over the following year.
What you see here are survival curves or the inverse of survival curves, cumulative prevalence curves. What you see is those who reported having frequent thoughts of death or self-harm nearly every day were eight to ten times as likely to either attempt suicide or die by suicide as those who reported having those thoughts not at all.
So, what this led to in our healthcare systems was the idea that we needed to implement systematic programs for identifying and addressing risk in, first, our mental health specialty clinics and then our primary care clinics. And this led to a sort of standard workflow that has now been implemented across most all of these healthcare systems in our network so that when members or patients come in and complete a PHQ9 questionnaire at a mental health specialty visit or at a primary care visit with a mental health diagnosis or in terms of screening completed annually for all of our members, if they were to score high on Item 9 -- that they have such thoughts more than half the days or nearly every day -- there would be an expectation that the clinician would complete a structured assessment of risk using the Columbia Suicide Severity Rating Scale. Depending on the score on that scale they would be expected to engage in collaborative safety planning and then develop an ongoing treatment plan.
These were data that our health systems considered actionable, but nevertheless, the use of the Item 9 of the PHQ is not that great. So, what you see here is, if you look at the second column score in Item 9, the left hand side is what proportion of outpatient mental health visits had that score. And then we look at suicide attempts that occurred within 90 days. The figures for suicide deaths are actually quite similar.
If you talk about relative risk, if you look at that column that says actual risk, the risk of those who score high is 2.3 percent compared to 0.2 percent for those who score low. That is a relative risk of 10. But the absolute risk is still only 2.3 percent meaning that is a pretty low positive predictive value.
Another way to look at that is, if you look at those people who scored 2 or 3 in Item 9, they account for 6 percent of visits. That’s the 2.5 plus 3.5 percent. And they account for 39 percent of suicide attempts, so that is a six to sevenfold concentration of risk. Useful but not that great. Meaning we still miss 60 percent of suicide attempts and about 60 percent of suicide deaths. And we see a relatively large proportion of those events that occur among people who score low, 35 percent who report having those thoughts not at all, but 35 percent of subsequent suicide attempts and a similar proportion of suicide deaths occur in that group.
So that led to this work saying could we use information in the electronic health records to more accurately identify people who are at high risk for suicide attempts and suicide death. This is work we published at first a couple of years ago.
The first round of work we did on this used data from seven health systems, about 20 million visits by four million of our members including both visits to mental health specialty clinics and visits to general medical or primary care clinics. We linked those data to data regarding suicide attempts and suicide deaths. Suicide deaths data came from the state vital statistics and mortality data in the states these health systems serve.
We harvested from the electronic health records about 150 potential predictors in the categories you would expect, demographic characteristics, mental health diagnoses, mental health medications, use of inpatient emergency services for mental health conditions, past injury or poisoning diagnoses including those that were considered self-harm as well as those that were considered accidental, responses to the PHQ9 questionnaire. There were some other things which I could go into if people have questions, including chronic medical conditions that were recorded in the record. It considered about 200 possible interactions.
I will get a little bit into the sort of nerdy pocket protector territory here but I will go through it pretty quickly. We used some relatively old-fashioned machine learning techniques -- penalized logistic regression, or LASSO models, using a pretty standard approach. We set aside about one-third of the sample. We developed models in two-thirds and then we validated them that other one-third. That is to prevent against over-fitting or sort of fitting to idiosyncratic relationships that are not generalizable.
We did a variety of sensitivity analyses that I can talk about if people have questions. But the bottom line that we have already published is that these models predict relatively well. What you see here are two of the four models we developed. These were to predict suicide death following a mental health specialty visit and suicide death following a primary care visit over the following 90 days.
What we see is an overall classification accuracy or area under the curve or C-statistics. About 86 percent for suicide death following a mental health specialty visit, 83 percent following a primary care visit, and relatively similar numbers for prediction of suicide attempts including non-fatal and fatal suicide attempts.
If we go back to that table we looked at a couple minutes ago, what we see is these machine learning models very clearly do better than just Item 9 of the PHQ. As a comparison, we looked at before and said those 6 percent of mental health specialty visits that have a high score on Item 9 account for about 39 percent of suicide deaths that happened over the next 90 days. Well, if we look at the top 5 percent, so a somewhat smaller group identified by risk scores, then we get that number up to 48 percent. So, considerably better.
In terms of what we missed at the bottom, that lowest category in Item 9, this is 35 percent of events, and if we take the bottom 75 percent of people based on risk score we miss only 19 percent of events. So certainly far from perfect but considerably better than those simple questionnaires.
The work we have done since then was really asking how can we improve on that. We have explored various different avenues including what if we build these models optimizing on different performance metrics. Those of you in this area might know what that means, but that means do we optimize the models for the area under the curve, a C-statistic, which is a fairly standard method, or various other tools? Or what if we looked at much more detailed temporal encoding?
When we did this first round of work, this was relatively crude. We essentially characterized these different predictors -- Ror instance, take the example of a history of suicide attempt. We said in our first round of work did someone have a history of suicide attempt in the prior three months. Did they have such a history in the prior year? Did they have such a history in the prior five years? We said maybe if we look at that in a much more fine-grain way we would be able to predict more accurately.
And then what if we use some of these what I call in quote more “modern” model development methods, moving from these penalized logistic models, which are relatively old-fashioned these days, to what some people would think of as more sophisticated machine learning models?
The answer -- I guess the good news and bad news is none of those really did much better. When we look at alternative performance metrics we really do not see better model performance when we build models based on those. When we look at very detailed temporal encoding -- and what I mean by that is instead of thinking of time in just three different chunks we think of time in 46 different patterns. So what that means is we take each one of those predictors and we represent it 46 different ways, so we’re getting now up into the tens of thousands of potential predictors -- we really don’t do much better.
When we look at what I call more modern model development methods, you could say that steps up a scale from penalized logistic regression to random forest models to neural network models to ensemble models which combines all of those, the gains in performance are either zero or pretty trivial. So we really don’t seem to do much better.
The bad news is I think in terms of getting data from health records to predict suicidal behavior, we may be at the ceiling of that. The good news is that some of these more complex, less transparent and much more computationally demanding methods may not be necessary.
That is sort of where this work is now in terms of these most recent things not yet published but will be published soon. What I would like to do for the next few minutes is just take a step back and talk about some of the more general lessons that I think we have learned from our work in this area.
I am going to point to three lessons. One of those is what I call there is no deep magic in these machine learning or artificial intelligence methods, or at least no deep magic that we were able to find.
I want to talk a little about what I see as a really important difference between the job of prediction and the job of inference, and then one more general talking about the difference between the job and the tool and the importance at least I believe of defining the job we want to do for these methods before we pick the tool we want to use.
When we talk about no deep magic, what I mean there is if we, in the most incredibly detailed and complex way, feed all manner of health records data into the most sophisticated machine learning tools and say find for us what predicts suicidal behavior, what these tools do is they find exactly what a sensible person would predict. We do not identify unexpected predictors or unexpected strange combinations of risk. When we allow the models to fit very complex interactions, we do get more accurate. We don’t discover really new things; what we tend to do is correct sort of over-estimation of risk.
If you want to get logistical about it you could say that in a logistic model, if someone has a history of suicide attempt, a history of psychiatric hospitalization and a history of emergency department visits for mental health reasons, the effect of all three of those is less than multiplicative. So the interaction effect allows us to account for that.
To illustrate that, what we see here in some of our most recent work, over on the left side if we are fitting these models using LASSO or a penalized logistic what we see -- I am not going to ask you to memorize or even go through all of the labels and predictors. But what you see is that the predictors are exactly what you would expect: diagnoses of self-harm, filling prescriptions for antidepressant medicines, score on Item 9 of the PHQ, making lots of mental health outpatient visits, inpatient care for mental health reasons.
If we look on the left side, this is a somewhat different presentation that comes from a random forest model where we actually have thousands of potential predictors, but that ASA stands for any suicide attempt, and you can see that all different representations of that are picked, and that is the highest weighted thing. MHI stands for mental health inpatient care. So, even the most sophisticated models are predicting the thing that a relatively well-informed clinician or even relatively well-informed lay person would know are important.
Now on to the point about prediction and inference. Oversimplified, inference is a question about whether something is true. Inference is all about interpretation. If we are fitting a statistical model for purposes of inference we are trying to interpret what the coefficient basically on that term means.
For prediction, the question is really a utilitarian question. We are asking whether a model predicts. And when we look at what is selected by one of these machine learning methods or if we’re using a parametric method like a logistic model, we’re looking at the coefficient. We are ideally not putting any sort of interpretation on that. We are asking how well does the model predict, and we may look at the predictor selected or the coefficients as a credibility test to say, as we just did, does this model select the things that we might want.
But we want to be very careful in these prediction models about trying to infer anything, and I am going to show that same picture now to make that point. If you read this carefully, what you see is we have all sorts of complicated interactions included in this model. We would say, okay, we have interaction terms between whether someone has a self-harm diagnosis specifically by laceration which is interacted for the days since that, which is interacted with them being in the age group 11 to 17. In a normal statistical world, if we said we will allow a model that selects all sorts of interaction terms without even selecting the main effects, we would say that is an abomination; you can’t do that. In the world of prediction it’s perfectly fine because we are not attempting to interpret the selection of that variable or the coefficient that’s placed on it.
Finally, I want to talk about this distinction between the jobs and the tools. If we are trying to use one of these prediction tools we really need to first define the job. We need to say prediction is to inform a decision. Prediction is to serve a customer. Who is our customer? What question or decision do they face? What information could be brought to bear at the point of decision? What action will they take? And what kind of errors matter most?
If we think about the specific job related to suicide risk prediction, in our case, our customers are outpatient mental health and primary care providers who have a patient sitting in their office, and they are trying to understand is this someone I really need to pay attention to in terms of a risk of self-harm. The question they face is what is this person’s near-term risk. The information they have available would be various things from the electronic health record. The action we are trying to inform is who needs further assessment.
When we talk about errors we need to think about what are the consequences of false positive errors. In this case, it may not be that great because it is not very harmful to do further assessment. False negative errors we may care about more.
I will close with this question of jobs and tools, an old story and a new story, because I think this illustrates what I think is sometimes the sort of unnecessary worship or overuse of machine learning and artificial intelligence tools.
Clever Hans was a famous kind of vaudeville burlesque in the beginning of the 20th Century, the so-called calculating horse. Some of you may know the story. The idea was Hans would go on the stage and someone would read simple math problems or ask Hans to spell something, and Hans would stamp his hoof or pick out letters. What we now know is that what Hans was really able to do was just read the unconscious clues that came from his trainer. It was not a conscious scam, but Hans was able to as he was stamping his hoof or picking out the letters.
But the real point of me bringing up the Clever Hans story is that using a horse to do arithmetic or using a horse to spell things was not a very good use of a horse. Even back in 1902 when Clever Hans was all the rage, there were better ways to spell things and better ways to do arithmetic. It was kind of a parlor game.
Here is a very modern example. This is one, because I’m a person from Seattle, that I use. This is somebody using the most complex data and the most complex statistical methods to conclude at the end that Twitter users in Seattle are more prone to depressive symptoms during winter than Twitter users in Jacksonville or other cities with low weather variability. My answer would be, you didn’t need machine learning for that.
My summary is these are very useful tools when put to the right job, but there are certain jobs they want to do, and if you hope to use these tools you have to first start by defining your job.
I will stop there and turn it over to David.
Agenda Item: Mobile Health Technologies and Suicide Prevention Innovation
DAVID LUXTON: Thank you, Dr. Simon. I appreciate that. In many ways your presentation could have followed mine because mine is a higher level description of the technology and innovation in machine learning, where some of the things I will talk about are really going to what you are already doing, and that is what we need to be doing research on. So that’s excellent.
I am Dave Luxton. My area of research is the development and study of behavioral health technology, so I do a lot with mobile tech and with telemedicine. I do a lot of training and so forth for clinicians on these topics, and so some of my presentation is kind of leaning towards the clinical world but you will find that it is also very relevant for doing research.
I am going to cover what is the need, what is driving the technology innovation in this area and what are some of the trends. I will talk a bit about the state-of-the-art technological capabilities and advancement. I will give some examples of suicide prevention apps and some of their core features. Then I will talk a bit about integration into care. That is more of the practical side of this. I will talk about some of the limitations and considerations. I will also talk about some of the research gaps and my own predictions for the future regarding this technology and its study.
As you are probably well aware, suicide is a major public health problem. It is the 10th leading cause of death overall in the United States. It is the 18th leading cause of death worldwide. It is the second leading cause of death among persons 10 to 34 in the United States, and the fourth leading cause of death among persons 35 to 54 in the US. We have also seen in the last decade or so a very high increase in the rate of suicide among young women and adolescent girls, which is very alarming. There has been some discussion and study about whether or not that is actually tied to technology and social media, et cetera.
Some of the things that are really challenging in suicide prevention are that people who are at risk or high risk are not often in care. Patients are often lost to follow-up after they leave emergency departments maybe for an attempt or after any kind of inpatient care or in any other kind of care setting. And imminent risk, in particular crises states, is very difficult to detect, and really that is because assessment and monitoring can be quite infrequent. High-risk persons can be socially disconnected and may just not have the support that they need. Also I would add stigma associated with mental health and seeking care is another barrier.
Technology can help with that. Technology is really driven by a number of things, but clearly it’s the technology itself. Technology is innovating very quickly. There’s a lot of power in a mobile device that we did not have even just a decade ago. We have a lot of improved capabilities because of our wireless capabilities with 5G and processing power in the devices themselves. Cloud computing is a big one and advancements in AI and machine learning.
Healthcare cost containment is another. Technology can provide a way that we can reduce the costs of healthcare by providing self-care approaches. Also, patient-centered care, which is thinking about what the patient needs and providing tools in their hands when they need them.
Let’s talk a bit about the state-of-the-art. I really want to emphasize the integration of data sources through technology. You think about all the various types of data sources that are available in healthcare such as data repositories and registries, electronic health records, any kind of survey data collected from patients on electronic devices, for example, or kiosks, and mobile devices such as wearables including phones. Implants is another type of mobile device if they are implanted inside a person. This is a really interesting emerging field of data collection. Environmental sensors such as in smart hospital rooms, for example, and more.
A little bit about AI and machine learning. What these bring to this approach in suicide prevention is that with mobile apps or mobile devices they can adapt to user needs and preferences, which provides that customization to what the user is more likely to interact with or want to interact with. It can provide capabilities for image and facial and voice recognition capabilities which is very interesting when you think about using these technologies for assessing affective states.
Virtual reality and augmented reality capabilities, data processing efficiency and automation such as multimodal data analysis for risk detection like in a cloud-based environment, and the improved device operation itself, just improving power consumption, privacy, et cetera.
Automated risk detection is the use of technology to detect patterns of risk in data. One example is linguistic-driven automated risk detection, and this involves using machine learning to scan and analyze and detect patterns in data or text. An example might be use of texting or calls on mobile phones, internet use. Any kind of narrative that the person is recording in some way can be analyzed for potential suicide risk.
Social media is a very interesting one. There have been a number of studies now that have specifically looked at social media use and suicide risk prediction. One example is the Durkheim Project, which was a DOD-funded project a number of years ago that looked at automated data collection from social media among US military veterans.
As I mentioned, facial recognition is an interesting way to assess affective states, and vocal characteristics can detect mood states as well.
I definitely want to emphasize wearables. These are like smart watches. What’s neat about these is they collect data in real time 24/7, and they can capture all kinds of data such as physical location, mood states, quality of sleep, for example. And all of this information can be put together into an analytics environment to run predictions and to create automated alert systems that can alert or notify clinicians or also provide alerts back to the end user.
This one always gets me really excited. This is the use of virtual intelligent agents, or virtual humans. These are built using virtual reality and AI to create highly interactive characters that can be on mobile devices. Some of the advantages of these are that they can be available 24/7, so if they are designed to be coaches for a person who is needing support they can provide that support. They can be used to collect data like survey data by the virtual human doing the interview. They can also be coupled with real-time data collection from other devices such as wearables so that the virtual human could be alerted to when there is some kind of alert that is being detected based on data from mobile devices.
One example might be let’s say a person has made a list of things that are risk factors for them and one might be location, say a location near a bar. The system could use GPS to create an alert that comes up on the phone and maybe the virtual human pops up and says, hey, let’s talk about what’s going on, and the virtual human can help that person cope with the current situation.
One obvious benefit of these is the close physical proximity to the end user. And unlike human healthcare providers, a virtual human can spend unlimited time with the user. Some examples of where they are being tested and have been tested are in places such as hospitals where they can be used for hospital discharge planning, doing surveys, serving as coaches and for support.
I have got to mention augmented reality. This is a really neat technology where virtual reality is projected onto the screen of the camera feed on a mobile device. It has been used I think mostly in the behavioral health world in trauma treatment or exposure therapy where you can overlay something that is a trigger for trauma on the actual environment that the person is in. It can also be used for behavioral change and also for training applications.
Let’s talk about the integration of these into care and how this is relevant for both care-providing and also for research. I want to point out this paper we published five or six years ago, which is very obsolete in the technology world. Five years is a century in technology, but I think it is still highly relevant. And this was the first, or one of the first, papers that specifically reviewed mobile health techs for suicide prevention and made some recommendations for clinicians.
A number of people have reported what are the features of mobile apps for behavioral health and for suicide prevention. NIMH created a list of six primary functional categories of mobile health apps.
But back in 2015, in that paper we specifically asked that question about mobile apps for suicide prevention and we identified six things back then as really the common things that these apps usually contain. Those are: information, education for training purposes like psychoeducation; resource locators, helping a person find help if they need help. Emergency buttons, so these are basically buttons where a person who feels like they are in crisis can hit that button and it calls somebody, whether that’s a clinician or a friend or potentially even 911.
Also safety planning and coping tools -- The apps can provide and help the person create the safety plan directly in the app and remind them of the steps of their safety plan. Assessment and automated intervention, and, also, provide social connection and support.
Here is just a quick example, a sampling of three different mobile apps that really encompass these features. Suicide Safer Home is one that just provides some information about making home environments safe and focuses on mean restriction and some other recommendations for suicide prevention.
The Relief Link app is one that combines features and includes a resource locator and some psychoeducation, and also it has an emergency button feature as well.
And the Virtual Hope Box. There are several of these now. What I’m showing here was developed by the DOD and the VA, but it falls from the idea of a virtual help box that a person who feels they are in crisis or can become in crisis can rely on this tool to help distract them from the disturbing thoughts that they are having, provide them some coping tools and help them to relax.
I am not going to go through this entire list but this is from that same paper from 2015. We put together really five steps that clinicians and/or researchers who are researching this technology in the clinical setting would want to follow. The first is identify clinical needs. You want to match the technology with what the clinical need is. Part of that is it’s important to know what the client or your patient is comfortable with. If they don’t want to use a technology or they are not comfortable with it, you are going to have to work with that. Otherwise, it is not going to work.
Determine what apps are available and assess the quality of the tool in application. This is a huge thing because there are so many mobile apps out there and so many now specifically for suicide prevention that you really want to select from the ones that are the most reputable.
Review regulatory and data security requirements. This is a big one. Whether or not you need HIPAA compliance, for example; data security and privacy. Very important.
You want to evaluate and test it yourself. If you are going to be using it with patients or clients or in a study make sure you know how to run it.
Also, plan use and develop protocols. Here I’m talking about things like a safety protocol, what happens if the person is in crisis and they are on their app and they have a texting thing and they text you after hours. You are going to want to develop your own protocols for addressing things such as emergencies.
Just a couple of quick limitations to the technology. Its access. If people don’t have access to the technology, whether it’s the tools themselves such as mobile phones or even just networks, if they are not on good internet connection that is obviously a barrier. And I always ask this one, too: will deployment and its availability be fair and equitable. This is a major ethical issue to think about as well.
Lack of evidence base -- but this is growing. We are seeing a lot more research in this topic specifically testing mobile apps for suicide prevention.
And then limited adoption, and this can be limited adoption by clinicians because they are unaware of the capability or they have concerns about data and privacy, also, cost or the lack of supporting infrastructure to use the technology.
Here is just a handful of what I think are some increasingly important research focus areas and also gaps. One of course is in the technology itself. The innovation of applications such as machine learning applications and facial recognition is another great example. This will always be a need because the technology is constantly evolving and improving.
Behavioral data science -- This really gets back to Dr. Simon’s work in his presentation with that focus on identifying what are the best models for using the data and the technology to either infer things or do prediction models. I certainly agree that having more data and more technology isn’t necessarily the answer; parsimony and also clinical applicability and feasibility. If you can’t really make sense of the data from multiple sources and an old-fashioned clinical interview is the most parsimonious, most easily adaptable, most effective, then that is probably what we should be using.
But I do have high expectations that the integration of technology is going to be revolutionized and increasingly so over the next decade and we are going to see a lot more integration of data towards prediction and other clinical applications.
I would just add clinical behavioral outcomes. We need to always be doing trials on the technology itself to see how effective it really is, and that is specifically both in augmenting treatment and when it is the treatment itself.
Just a couple resources that are available. One of the things that often comes up is whether or not the app itself or the mobile application, the technology, is a medical device. The AMA publishes some great information on this. I also recommend checking out mHealth Evidence as well. It is a great resource.
And quickly, the future. Bigger databases, more cloud computing, improved sensing and affect detection technologies, better integration with health records. AI and machine learning will continue to improve. Virtual intelligent agent integration and augmented reality is only going to increase.
Final thoughts. You can do this. You can integrate this into your research and into your care. You should. If you are not, your clients and clinicians already are. And always be aware of ethical and safety issues regarding the technology. Be competent in their use, and keep on research this and seek additional and ongoing training. Thank you.
Agenda Item: Q&A
REBECCA BERNERT: Thank you so much, Dr. Luxton and Dr. Simon. I am going to take this opportunity to invite all of our panelists to rejoin our panel here for a full panel discussion. I will give everyone a few moments to come back and join us. In the meantime, I would like to ask a few questions to our most recent panelists.
Dr. Luxton, one question that is coming up and came up also in the last session is regarding some of the availability of these techniques or these apps in the public domain. I am curious what your thoughts are regarding in particular the virtual human technologies that are available. A question about how sophisticated those are and whether or not you think that could be novel in terms of the delivery of treatment resources.
For example, we have heard a lot about how efficacious some of these wonderful treatments are and they are very quick acting, and we already know that those are published out there in the literature. Could that be an opportunity to perhaps have a listing of resources that could be customized or that could actually promote access to care?
DAVID LUXTON: I am super-hopeful about this technology. The value is it doesn’t need to be any more realistic than the basic kind of chat bot approaches. We are finding those to be fairly effective. The more realistic they are, that’s great, it looks great, but it doesn’t really need to be that sophisticated.
We are seeing their application already, as I mentioned, in some settings such as in hospitals and also as augmenting suicide prevention therapeutic techniques. The thing to watch out for is there are a lot of companies developing this stuff now and you have to think about all those issues. Are they really aligning it to actual evidence-based approaches, and data security, those kinds of issues that I think are really important.
I would love to see more resources on the availability of using that type of technology specifically in clinical care with regards to suicide risk management.
REBECCA BERNERT: Thank you so much, Dr. Luxton. Dr. Simon, a question for you regarding the utility or your thoughts regarding use of machine learning for detection of sleep disturbances potentially. I know you do some work on healthcare learning models. Just your thoughts regarding possibilities there based on your work and expertise.
GREG SIMON: I guess one way I would comment on that in terms of bringing these two things together is that I think the healthcare data have proven themselves to be relatively useful in what I would call a risk stratification task. Our ability to stratify people into these people in the top 1 percent, these people are in the top 5 percent, and especially for relatively stable risk.
It was interesting to us that even though, as I said, when we use these much more complicated time patterns, most of the things these models predicted were relatively long term. Has someone made a suicide attempt in the last year. And whether that was in the last month didn’t seem to be that much different from the last year.
The healthcare data seem to be relatively good at what I call the who prediction. By that I mean who is at risk. So far, not that useful about the when prediction. If we are going to get any better at the when prediction then we probably need to think about things that happen outside the healthcare system and are not visible in medical records data. So those might include some of the things that Dave Luxton was talking about in terms of things that people’s mobile devices might know.
It also might include -- Healthcare data are very poor at registering important things that happen in people’s lives. The healthcare data can tell us exactly when somebody filled a prescription for which drug in what quantity, but they don’t tell us when somebody’s spouse died, by and large. There is not a diagnosis for that. Or they don’t tell us when somebody was evicted.
So we are interested in asking questions about what could you do with healthcare data and these other kinds of what people would call real-world data, data that come from the mobile devices on what happens when people are out living their lives, and data about other things that happen to people in the world.
So in terms of things we could get from healthcare data, I think we have pretty good tools for the who prediction but I suspect we are about at the ceiling on that.
REBECCA BERNERT: Thanks, Dr. Simon. One of the topics that was brought up in the State-of-the-Science overview was this issue with risk assessment and variability in different training practices. I know that Dr. Luxton, for example, you have done some work with risk assessment training, and we are seeing some work at the policy level and state levels, some changes, and also within healthcare systems -- so I am interested to field this question to both of you -- in mandated training for risk assessment in all hospital settings. And it is something that your state uniquely has begun, and many people don’t know that the State of Washington has now required mandated training across hospital settings.
And I worked with the State of California who recently has prioritized a similar initiative that is being approved for state-level training among clinicians in these areas.
So I am curious. Applied to sleep, your thoughts regarding whether or not that could be something that could be beneficial and then specifically in the study of sleep as a target for risk assessment screening whether it’s using a technology in an application or within a healthcare setting.
DAVID LUXTON: That is a great question. We have seen it to be very effective here in Washington, the training for suicide prevention for all healthcare providers. When I teach it I always have a little module about sleep because I think it is obviously highly relevant for suicide risk management and prediction, et cetera, so I always teach that.
I think there probably could be some benefit in mandating some training on the sleep management piece because we know how it’s tied with so many other health conditions as well. What’s interesting is how do you go far enough with it to make it really useful in a limited amount of training and in different fields. In different fields physicians might have more training in that already versus a dentist. So it’s something that you have to consider what is the level that is really effective to train on.
GREG SIMON: I might add a couple of things. Washington State has put in place a sort of requirement that healthcare professionals get training in suicide risk assessment, which is I think in general a good thing.
There are some interesting questions when you think about our understanding of what I would consider sort of the data question. What data elements, if you want to put it in that sense, are relevant in terms of actually predicting risk? Then there is the human side about what is the right way to ask about that, in what context, in what way.
One of my colleagues has actually done some interesting work. One of the things we are very interested in is those people who fill out one of these standard questionnaires in one of our clinics and on Item 9 of the PHQ or even on the Columbia Suicide Severity Rating scale they score zero. They say no, I don’t have thoughts like that. And then some portion of those people will actually show up in the emergency room after a suicide attempt in the next 30 days.
We have talked to some of those people. This is the other end of research. This is the N of 20 rather than the N of 20 million. You talk to those people and say could you tell us about that? What happened? Because obviously, a risk assessment can fail. And we heard various things. There certainly were people who were concerned about involuntary treatment and said I was having those thoughts and I didn’t want to tell you. There were people who really did say I was not having those thoughts at the time; it really did come over all of a sudden. There were people who talked about substance use. I only have those thoughts when I’m intoxicated and of course I was not intoxicated on the day I came to my clinic visit.
One of the ones that was most interesting and in some ways sort of heartbreaking to me were the people who said I am really trying to be the kind of person who doesn’t have those thoughts anymore, so on that day when I came to see the doctor -- Essentially what they said is I answered like the kind of person I am trying to be. It’s hard not to sympathize with that even though it messes with our risk assessment.
So when we think about training, the components of that need to be what are the right questions to ask, but what is the right way to ask them, and what is the sort of social context to set up so that we would expect people to really engage with us and tell us the truth about that.
One of the things we have really struggled with is talking about training in collaborative safety planning. How do you train about that and how do you measure that, because safety planning is not about telling people to do something; safety planning is talking to people about what will work for them. We try to generate data about things. The data we have tend to be from looking in the electronic health record and saying what does the safety plan look like, but I don’t know if that tells us much about the process of collaboration that generated it or not.
REBECCA BERNERT: Thank you so much, Dr. Simon. I think that is a perfect introduction to kick off questions for the entire panel, and thank you so much for all of the spectacular presentations over the course of the last few days -- to ask the panel and invite discussion on measurement issues, which of course we know have come up consistently across all talks, whether you’re talking about sleep disturbances and different types of sleep disturbances and really wanting to maximize study of different symptom components, symptom dimensions and so forth. Some work that was noted by Dan Buysse and Tom Neylan and Dr. Hasler’s talk specifically around this as well.
And then of course in suicide prevention where we see a great deal of variability in suicide risk assessments and a lack of adherence to some of the uniform guidelines that we have regarding the measurement of self-directed violence and specifically suicidal behaviors.
I am curious to field to the entire panel ideas regarding each person’s individual area of specialty and how we might be able to advance understanding of mechanisms and also enhance the use of consensus around certain types of sleep assessments or the assessment of specific sleep variables or suicidal symptoms and so forth to guide precision in this particular area.
DAVID LEITMAN: Rebecca, before people answer I would just like to -- This is actually one of the things that I was going to ask. I wanted to ask the same question that Rebecca is asking but slightly differently. One of the things that NIMH can do is it can promote certain guidelines for research for how to measure aspects of sleep or aspects of suicidal thoughts or behaviors or ideation, guidelines that we think people would accept and that might elevate the level of the current research that is done.
So, when you are answering Rebecca’s question could you also think about it from the perspective of what are things that NIMH could do in terms of the guidelines that it can promote?
REBECCA BERNERT: Great point, thank you so much.
W. VAUGHN MCCALL: I will take a stab at it first. I want to reflect on the condition of -- We’re talking about clinical trials. When thinking about the condition of clinical trials in suicide versus, say, clinical trials in depression, we are as suicidologists woefully behind in comparison. If you were doing a clinical trial in depression you are almost certainly going to have an observer-rated scale and it is either going to be the MADRS or the Hamilton, one or the other. You’re probably not going to go too far off the grid from that. And you may have one patient-rated scale of depression severity just to balance out the observer scale.
If we are going to get ahead in thinking about interventions in suicide, we are going to need to eventually come to the same terms that the depression researchers have; that is, we need an agreed-upon scale which is sensitive to change and reflects the things that are important in the prevention of suicide.
The group that I work with, including Michael Neydorf who is not on the call today, just taking some of the clinical trials work we have done in suicidal patients, we had three measurements. We did the CSSRS, we did the Beck Scale for Suicide Ideation and we looked at the suicide item on the Hamilton, so we measured suicidal ideation three different ways. Then we did a post hoc analysis subjected to item response theory. And the question is which of these approaches, observer-rated, patient-rated, 21-item, one-item -- what do we have to do, and can we come to some sort of consensus.
I would say one of the things, David, and what you’re saying regarding what can NIMH ask for is multi-modal approaches. But for now, until we can make a decision, simply saying we are going to measure suicidal ideation one way, I don’t know that we know enough to say which one way that should be. And so we probably should insist upon multiple simultaneous measurements until we get it figured out.
DAVID LEITMAN: Thank you. I would note that you have a little bit of feedback on your mic, just FYI.
ANDREA GOLDSTEIN-PIEKARSKI: I would just say that one thing I think is really important is helping to have these measures be open source and not behind paywalls, because that certainly is a limitation especially for early career investigators, who may not have a lot of funding to go into all of the different measures.
GREG SIMON: Yes. I was going to add the one thing that makes this task especially challenging but especially interesting is, unlike -- To use that analogy about depression, in depression, that thing we are hoping ultimately to affect is, to put it in nerdy terms, an area under the symptom curve. What we’re trying to say is, okay, can we reduce the overall severity of depression for this person integrated over time.
We talk about suicide prevention. It is an act, an act that happened at a specific point in time and occurred in minutes. As pointed out earlier in this program, suicidal ideation may be a risk factor for that, and I don’t want -- You know, reducing suicidal ideation in and of itself is good. That is an unpleasant and miserable experience for people to live with and we would like to reduce that.
But ultimately, we are interested in the prevention of that act, and we really actually don’t know that much about how time patterns of inner experience relate to probability of that act. That is likely to come from some of the stuff that Dave Luxton was talking about in terms of more ecological momentary assessment and sort of in-the-moment assessment, and there is some interesting stuff starting to come out about that.
This is really challenging. Because suicidal behavior is relatively rare, we need data about a lot of people to try to understand it, which certainly pushes you toward let’s try to have everybody look at this the same way and use standard things, but premature standardization on an approach is sometimes not good because you standardize on something where you don’t know enough to decide. So that is a longwinded way of saying this is kind of complicated.
I think, going back to the question about NIMH to say maybe we need some balance between here are some standard things that we think are important to measure in all studies about this, but to encourage some creativity and to encourage -- because that is how we’re going to discover that those standard things -- And you know it is not that dissimilar to other clinical areas in mental health where we might say -- Well, I will betray some of my prejudices here. We might say if you’re doing a study about depression, probably you should use some standard measure like the Hamilton Depression Rating Scale. But the Hamilton Depression Rating Scale is something somebody just made up in 1967. It’s not like it was developed through a data-intensive, empirical process.
And understanding between-person differences in depression, I don’t think the Hamilton Depression Rating scale is going to get us there because if it was going to get us there it would have gotten us there already. We clearly need very different and new measures to understand that kind of heterogeneity.
So this is my longwinded way of saying there has got to be some balance of here’s something so that we can have some comparability that everybody should do, while clearly acknowledging the measures we have are just not getting us there and something different is going to have to come along.
REBECCA BERNERT: Right. A few thoughts there. Dr. Hom did some work previously regarding the way that you ask - essentially the questions surrounding suicidal symptoms -matters - and the method matters. I thought maybe you could talk a little bit about that.
And then, Dr. Zuromski, there was a great tie-in to some of the work that you are doing that’s so wonderful on EMA work, and then tie in with work by both Dr. Simon and Dr. Luxton and others. I thought maybe you could start us off, Melanie.
MELANIE HOM: I think in assessing suicidal behaviors specifically we have seen a lot of challenges in, first of all, disagreement among researchers and clinicians around how to even define various forms of suicidal thoughts and behaviors and disparity through the existence of uniform definitions. And then there’s this added layer of even once we can sort of agree as researchers how we are conceptualizing a suicidal behavior like a suicide attempt, trying to actually explain that definition to an actual research participant or patient becomes incredibly difficult. We sort of take for granted that we understand what certain behaviors look like, but that doesn’t always translate to patients and participants being able to answer questions.
So we definitely have found across studies that these sort of single-item assessments and self-report assessments of suicide attempt history often yield inaccurate or inconsistent responses from folks, just because folks often don’t know did their behavior actually constitute a suicide attempt. Oftentimes clear definitions aren’t really provided.
At least in my work across a couple of settings we have definitely seen that interview-based measures where folks have the opportunity to clarify what behavior they’re thinking of and for clinicians to ask follow-up questions seem to be quite important. We see quite a bit of over-reporting of suicide attempt history without a more nuanced follow-up.
In particular we have seen that the Columbia appears to be particularly good at parsing apart different suicidal behaviors. But it is definitely a challenging area of research and sort of calls into question what we know about suicide attempts from prior studies, wondering did prior studies capture a higher rate of suicide attempts than might have actually been present. We see quite a disparity, so definitely an area where there needs to be more work.
KELLY ZUROMSKI: I was going to comment as well when Dr. Simon was talking about the importance of thinking about in what kind of study you are assessing suicide. Everything that Melanie just talked about is great, and the Columbia is great, but you cannot give a Columbia every single time when you are doing EMA surveys. If you’re asking someone six times a day if they are thinking about killing themselves you can’t have them doing a Columbia measure.
Just the importance of establishing consensus and exploring further how to even ask about these things and different time intervals when you’re asking someone momentary items versus trying to get a history for their entire life of suicidal thoughts and behaviors.
We are actually doing some work in the lab trying to explore different ways of asking about it and trying to get at what is intent, what are people meaning when they say intent, especially when they’re thinking about it in the moment.
I agree with Dr. Simon as well that it is definitely an area where consensus can be helpful in terms of open science and collaborating and getting big enough datasets to investigate things that we are interested in but also still very much being at a place, especially for a lot of the EMA work, where we are not exactly sure what we ought to be asking.
GREG SIMON: It is also interesting, even when we think about relatively simple self-report questionnaires, to think about what other data are hidden in there. One question we are very interested in starting to look at, although I know this creeps some people out so maybe I should hold my tongue, but is when people fill out these standard questionnaires we are interested not just in what the response is but the time it took to record the response in terms of how many milliseconds of delay between the presentation of a question regarding suicidal ideation and the recording of the response and how that relates to subsequent suicidal behavior. I have to think there is going to be something there.
REBECCA BERNERT: The other thought I had is just regarding the difficulties in acute measurements. And as Dr. Zuromski mentioned and with your work, Dr. Ballard, and in ketamine treatments in particular, really looking at rapid reductions can be very difficult.
I know that myself and everyone else during the entire treatment panel session, we all are using multi-method assessment approaches, appreciating that each assessment has limitations. But of course we do want to have comparability and so we want to use the Columbia, but alongside a lot of other measurements that might be able to be assessed in real-time.
The other question I have for folks is really trying to get at the other end of the symptom dimension if we’re talking about suicidal symptoms, and then I would like to turn us over to the sleep focus. You know, suicidal symptoms – they are a symptom of depression, and the milder end of the spectrum as we talk about -- when we’re explaining basically best practices in risk assessment and providing feedback to patients, for example -- the focus on the full symptom dimension. If you are thinking about desire, sometimes individuals aren’t even having necessarily desire; they are just having thoughts about death. And we are not necessarily talking about that.
The other question that I love, too, is items on the Beck Suicide Scale looking at, look - maybe you don’t have suicidal desire, but if you are put into a life-threatening situation would you fight? Would you fight for survival? And we know that the will to live is reduced and is not reflected, unfortunately, in measures like the Columbia.
So, really wanting to maximize understanding, and taking this dimensional RDoC approach, would involve really trying to get at the full end of the spectrum, in my view, and I am interested to hear what others think.
DAVID LEITMAN: I also want to piggyback on what Rebecca is saying. One of the people who is listening in today has pointed out that data show that, at least according to his claim, 50 percent of people who die from suicide have no recent history of mental health problems. He suggests that we should look upstream at financial factors, job-related factors, relationship factors as well and not just focus on mental health issues.
GREG SIMON: I immediately translate your last question into my world, which is health records data, but it does lead to some interesting questions that we are hoping to start looking at about various health behaviors that might be related to will to live. Things like, do people who die by suicide get their colonoscopies? Do people who die by suicide complete their cancer screenings or get flu shots? It gets to this idea about future orientation and what healthcare behaviors might indicate future orientation or lack of future orientation.
I guess I am a behaviorist at heart because I believe if you want to know what people think that you want them to say, you ask them. If you want to know what people are actually thinking you observe their behavior. So I am interested in what are some of these healthcare behaviors and how they might relate to some of these underlying constructs that we would hope to measure.
Of course, those behaviors, there are probably other, sort of consumer type behaviors, other behaviors that are out in the free living human world that would be much more informative, but just the healthcare data are what we often have access to.
REBECCA BERNET: Absolutely. And Dr. Pigeon I think had a question on follow-up?
WILFRED PIGEON: I have three quick points. With respect to David’s most recent comment and Dr. Simon’s contributions, also not to forget social determinants of health. Hong Hu at UMass is currently engaged in using NLP approaches to pull those out of medical records.
Secondly, let’s not forget computer-adaptive testing and push and association tests in this potential list of where we should go. Certainly, CAT can be applied to both suicide and the sleep side.
Finally, David, if I could encourage you to encourage us to provide some answers for you so that NIMH and other institutes can then disseminate that to review groups and committees and panelists who diverge very significantly in terms of their feedback. Don’t use this instrument, please use that one, this one stinks, this one doesn’t. So guidance would be helpful.
DAVID LEITMAN: Yes, and I think we are going to get to some of those with what Rebecca is going to talk about when we get to sleep. I listen in on many study sections, and when it comes to measuring sleep or measuring actigraphy, everybody has their favorite device, their favorite method. This one’s bad, this one’s good, this one’s horrible, this one’s better. Maybe it is unsettled right now and maybe we just have to let everything hit the wall and see what sticks. But it would be nice to be able, from our perspective, to have some idea if some form of consensus does exist.
REBECCA BERNERT: Let’s open it up to the sleep specialists on the call and ask for additional thoughts about sleep disturbances, sleep measurement - behavioral, using more objective assessments and different ideas there, as well as circadian measurements or measurements that get at that circadian component or even that misalignment component that we have heard about is so important.
THOMAS C. NEYLAN: Maybe I will jump in here. I think it is safe to say that it’s unsettled in our field. If you asked the sleep expert on this call or the expanded attendees, if held at gunpoint what would you say is the most important sleep measure, you would not get agreement from everybody across the panel. Partly because I think you can see from this conference that we know that there are a lot of things that are important. We know that circadian alignment is important, we know that slow wave sleep is important, we know that REM sleep is important, we know that sleep consolidation is important, we know that sleep disorders are important. The list goes on and on.
Especially when you’re tasked with looking at suicide as an outcome and you have this rare outcome with a multivariate predictor, you’re talking about a power nightmare. It’s almost like we need to think about trials that try to select more homogeneous populations based on a particular problem in the sleep world, whether it’s frequent nightmares or if it’s a sleep disorder or if it’s short sleep duration or something like that, as opposed to taking all comers. I think the scale of the trials needed to answer some of these questions raised in these two days is just not pragmatic.
DAVID LEITMAN: I think we definitely need to follow it up. and I am thinking about what Vaughn McCall said and his calculation that we might need 10,000 subjects if we want to actually study suicide attempts.
W. VAUGHN MCCALL: That was suicide deaths.
DAVID LEITMAN: Deaths. And that was similar to the calculation I made.
My final question was actually what would a large-scale study look like? When I did the research preparing for this conference, I found that many different aspects of sleep, as we went over here at this meeting, are predictive of suicide attempts, from attitudes towards sleep, towards abnormal polysomnography, to nightmares. And it seems like there might be multiple mechanisms, not a singular mechanism, that likely explains all of these things.
If you couple that with the disjunction between suicide ideation and actual suicide attempts and suicide deaths, you need a large-scale study to do that. So, one of the things I hope that we can discuss or start the discussion at least a bit from this meeting, would be what a large-scale study would look like.
REBECCA BERNERT: The important point that brings up, and I am glad you mentioned it, is just getting ahead of the curve because the other option is going into high-risk settings. We know emergency department settings are prioritized for a reason. Really, all transitional settings are associated with higher risk for suicide.
But if we look at ED settings and we look at post-hospitalization and carrying over, evaluating and maybe oversampling sleep in that environment, we could look at the prevention of repeat suicidal behaviors and then also look at access to care and all of these other things.
I wanted to bring that up because it is relevant to a lifespan approach, particularly in youth - where we know, according to the Institute of Medicine, it isn’t 25 suicide attempts for every death by suicide; we have 200 attempts estimated for every suicide death in youth. This is an incredible opportunity to get ahead of the curve and actually prevent loss of life in our youngest population, who are otherwise incredibly healthy but are dying by suicide at staggering rates.
I would love to hear thoughts about that. And what kinds of assessments could we do in sleep environments to get at biomarkers and other medical issues that are going on, and maybe separate and tease apart the two, that bring up measurement.
PETER L. FRANZEN: I think some of these issues that are being talked about are really the design -- or the question fits the design. If you really want to be looking at -- Even people who have active suicidal ideation don’t necessarily have it every day. So, if you want to extend people -- You know, we made the decision to try to watch people over a longer period of time to have more events, not necessarily even attempts but just ideation, and so we chose not to ask questions multiple times a day because how long can you actually get people to do that.
But this idea that we need to look at sleep more holistically and different aspects of sleep I think is really critical because the literature is all over the place about what kind of sleep factors are related, because it really hasn’t been looked at all that systematically or in a sleep health kind of framework.
GINA POE: I just want to reiterate what Dr. Franzen is saying in that there are all kinds of things that go wrong, but one of the themes that seemed to be constant and consistent throughout all these studies is sympathetic nervous system activation. And so I would suggest that, if we want to measure something, it would be to develop a good way to measure sympathetic nervous system activation reliably - sympathetic system activation, reliably across sleep and during sleep.
If I were to pinpoint one thing, that is the thing that I would develop.
GREG SIMON: I was going to say that when you talk about huge samples, I think some of the work that has been done about risk stratification can be helpful there. We are able to identify people who are at relatively high risk.
For instance, we can accurately identify people who have a 90-day risk of suicide attempt at 10 percent or more. There your event rate in that group is pretty high. We don’t have a very good way of saying exactly when are those people at risk. But if we are interested in more intensive study of sleep and other risk factors, we certainly could focus on high-risk people and be much more efficient.
One of the big challenges, and this is a hard one, is that so much of suicidal behavior, suicide attempts and suicide deaths, a substantial proportion occur among people who we don’t know they are at risk, and that is very difficult to study. And it’s sort of that prevention paradox just like with cholesterol or high blood pressure that, yes, you can really identify the people at risk but there are a still a lot of events that occur in the lower one-half of the risk distribution, and studying that is a real challenge.
REBECCA BERNERT: I believe Dr. Hasler has a question.
BRANT P. HASLER: I was thinking about this already and then Dr. Poe’s comments about sympathetic nervous system. Something I have been struck by is how the consumer wearables are now surpassing our research-created actigraphs in terms of being able to measure sleep, especially when they have the heart rate measure being better at detecting sleep staging. So we have an effective, cheaper device out there - that is being widely disseminated in the population, including things like the ABCD study where we have 12,000 adolescents.
I guess I am not positive if they are trying to do that in the entire sample. Peter, do you know? Does anybody know? But I am wondering, are our consumer devices going to provide a platform for large-scale studies of this including things like ABCD that are already ongoing, or where perhaps even if the relative rate of suicide ideation is small you still have a big enough sample that maybe you get a signal.
One other thing I was going to say, and I don’t know if Dr. Burgess wants to weigh in on this. I am also struck that at the same time we are seeing better and better prediction using actigraphy, sometimes with light data sometimes with not light data, and predicting circadian metrics using these mathematical models. And there is a paper that just came out that Helen is on talking about how good this is getting, even including misaligned populations with variable sleep schedules where it has struggled in the past. So I think there is some potential for trying to get more at these circadian metrics as well.
HELEN J. BURGESS: I definitely think we are making progress, and it’s my dream that we will have wearables out there that will eventually give us some pretty good proxy markers for circadian timing. The problem is I just think we are not quite there yet.
I think if you take healthy young people, who are on fairly regular schedules and you use light data and activity data from actigraphy, you can predict the dim-light melatonin onset usually within about an hour of where it is. But if you take people who are on more variable schedules -- and of course we know people who are at risk for suicide have more variable sleep -- or for example you look at shift workers, then the prediction really starts to fall apart.
I think we are getting there. I would agree with you that anything that includes heart rate is definitely going to help us, and I think the heart rate variability measure is great. I think we just need to keep in mind that we don’t quite know what is probably a circular relationship where, to some extent, sleep is driving heart rate variability because we know it increases with slow wave sleep, for example, and decreases during REM. But on the other hand, if someone is stressed and the sympathetic activity is high, then of course their sleep will be disrupted.
But certainly, as a proxy marker of sleep, I think heart rate variability would be useful.
REBECCA BERNERT: That brings up another great question that takes us into the treatment world and realm regarding just the effects of CBT-I on the circadian system and on that sleep-wake regulation and stabilizing the clock and using that low-risk approach for prevention of higher risk outcomes.
I know we are interested in our studies to look at some of the actigraphy data and the light data and so forth, and we have DLMO assessments for a sub-sample pre- and post, and are interested to do dismantling studies. But, for those that are specialized in treatments in particular, the question comes up regarding what is going on with this effect, with the sleep variability - and it’s sort of this driving factor that makes absolute sense conceptually in terms of insomnia and nightmares and other sleep factors.
But that is largely targeted as well within a CBT-I framework. We theorize that we are going to see stabilization of the sleep-wake system independent of social rhythms therapy, for example, just as a product of the sleep consolidation protocol itself, which I don’t know how familiar people are. But if others could chime in regarding that with thoughts about its ties to suicide, I think that would be wonderful.
PETER L. FRANZEN: This doesn’t really answer that except to say that when you look at ultra-high risk populations, so, people who have usually affective psychopathology and ongoing suicidality, insomnia isn’t always the main thing, and it’s all over the place.
The suggestion earlier might be smart, which is maybe you need to have a more homogenous population and see how that works. But I think also, at the same time, we need to think about going broader so that we can target the trans-diagnostic kind of sleep problems that we see.
REBECCA BERNERT: Other thoughts regarding the sleep variability component and some of the circadian aspects within treatment?
HELEN J. BURGESS: I was going to mention that I would definitely think that CBT-I should stabilize sleep. But one thing that I have wondered about as to the circadian question is CBT-I includes the instruction to get out of bed if you can’t sleep, and we just wonder a little bit about light exposure during the night.
Maybe there are studies out there. I don’t know if I have seen anything where people were looking at the effect -- Rebecca, maybe your study will be the first to kind of look at circadian timing with these gold standard measures such as the dim-light melatonin onset before and after CBT-I.
Somewhat of an easy solution to this of course would be, if people are comfortable doing it, to give the advice that obviously it should be dim light. But we know, for example, Brandt has been working with blue blockers which just block out the blue wavelengths of light so the light becomes a lot less potent in terms of shifting circadian timing. So there would be some small add-ons potentially to CBT-I to try to reduce possible circadian disruption. But I would agree that certainly a big benefit is just stabilizing the nighttime sleep episodes.
THOMAS C. NEYLAN: The interesting thing about the circadian aspect of CBT-I as an outcome is that in some ways it is almost an adherence check, as clearly stabilizing the timing is part of CBT-I. That is what the intervention does. So, seeing a stabilization of the sleep wave timing would be almost more of a marker that your subject has adhered to your advice that you gave them.
I think it remains a really interesting question about why CBT-I is so good for so many outcomes, especially since it has such negligible effects on sleep duration. Why is it so helpful? It is really an interesting question. Testing efficacy of CBT-I for -- name your outcome -- is almost not even interesting anymore. You can hardly get people interested in that question.
The real issue is why, if you force people to spend less time in bed and you stabilize the timing, do they feel so much better? I think that is such an interesting question that has not been answered.
REBECCA BERNERT: One idea that came up around this DARPA workshop that I was part of a number of years back was the idea of agency, and I think Dr. Woodward had brought this up as well. Certainly in the treatment of nightmares and insomnia, you are empowering the patient with all kinds of scientific information that normalizes understanding of how insomnia develops, of course, very naturally, and is learned over time, and can be easily unlearned, and puts you in the driver’s seat.
So I would love to hear other thoughts on that from the panel.
ANDREA GOLDSTEIN-PIEKARSKI: I was just going to add that one interesting thing about the CBT-I mechanism is there are dismantling studies which also show that even individual parts seem to work pretty well regardless of which piece, which is really interesting because it argues that they potentially are doing it through different mechanisms while still having the same outcome. I just wanted to add that.
REBECCA BERNERT: That is a great point, especially if we are really trying to get at some of the mechanistic aspects of it and then looking at some of the dismantling components to figure out what is driving what.
ETI BEN SIMON: I also wanted to add that we see a lot of important (inaudible) quality when we add a lot of the affective measures that we look at. I think anything that consolidates the continuity of sleep really has better or more impressive effect on affective measures than duration alone.
If I can also just point out something you mentioned before about teenagers and suicide risk. I am thinking that a lot of the teenagers and young adults that we come into contact with are in institutions, either colleges or schools, and there are these really effective brief interventions we can do regarding sleep hygiene. I just wonder out loud why are we not just giving them up? Why are we saving them for after people show up with distressing signs? This is very helpful to everyone, especially at that age.
REBECCA BERNERT: Right. That is such a fantastic point. And there was work two decades ago -- and we include this as part of our treatment -- going over just some of -- I think Rachel Manber conducted a study simply looking at normalizing the sleep-wake schedule, which is also wildly dysregulated for all of us, as you can remember when we enter young adulthood and certainly in college, and everything improves. And it’s such a low risk. And that is where individuals were asked to create any schedule they want, but just stick to it. Make it straightforward, make it stable.
That is something we can do at entry points, right - from more of a prevention standpoint. And I think this speaks to some interesting discussions about looking more upfront of risk - and really looking at the novelty of sleep being this universal, biological need that is non-stigmatizing - that we all need help with - and that we all can relate to, to actually design interventions that can prevent its occurrence in developing. That is something that, doing these trials, we became very interested in. I’m sitting here thinking, wait, we know exactly how it develops. Why not just create a component of this and basically offer it to veterans pre-deployment? Why are we waiting?
Same thing with even nightmare development. I would be interested to hear thoughts regarding the prevention of nightmare development.
KELLY ZUROMSKI: I would just make a point on the technology piece, not to beat a dead horse on that, but thinking about dissemination of sleep treatments and suicide treatments and thinking about some other work that we are doing right now trying to -- as Dr. Simon was saying, if you want to see what people are doing, observe them -- observe people possibly with technology, with wearable devices and getting some active input from them with surveys. But finding a way to maybe provide especially sleep hygiene interventions in the moment when people are needing that reminder. If they are on their phone and they’re getting a blast of light in their face late at night, having something pop up so that you can engage with the individual in that moment to try to provide them a piece of that psychoeducational information that we are all talking about would be great if it was just available to everyone.
So just another pitch for the role that technology can play here if we are rolling these interventions out at scale and in moments when they are really needed.
REBECCA BERNERT: Thank you. I see the question from you, Wil (Pigeon), just drawing our attention to an excellent question asked by the audience.
Given the strong literature about shifted sleep-wake schedules in adolescents and the literature on sleep and suicide risk, I would hope that one of the objectives of NIMH would be to try to implement policy change around school start times, something that the entire sleep world has had consensus around for over 20 years but is still not changing. Thoughts from the panel?
HELEN J. BURGESS: I was going to say the school start time, yes, that issue is extremely difficult. I was just going to add that I know some folks that study substance abuse in adolescents, for example, and they work on getting messaging out via TikTok and SnapChat and Instagram. So I think the schools would be a great place to try and introduce some of this information, but also, obviously, a lot of the social media platforms that adolescents are on as well would make sense.
REBECCA BERNERT: The one topic that came up throughout and is really thematic, I think, throughout this workshop has been focused on resiliency and self-healing. Although we didn’t get to talk about it, I think it is something that’s being drawn out of all of this work, and the idea can be also viewed as a protective factor, which always gets less attention than risk research.
So I just want to end, before we switch to closing remarks, to invite discussion and collaboration regarding that in particular - seeing what we’re seeing in terms of its benefits, protective effects, and perhaps looking at different neurobiological substrates and mechanisms that might get at that as well – that are more and more focused on self-resiliency and other models.
Agenda Item: Challenges, Opportunities and Future
Directions - Closing Remarks
REBECCA BERNERT: Thank you, everyone, for your phenomenal presentations. On behalf of myself, NIMH and my Co-Chair, Dr. David Leitman, I would like to thank everyone for joining us these past two days for this incredible workshop. We couldn’t be more grateful for our speakers and participants and, in particular, during such an extraordinary time. For everyone on this panel and in the audience - it is such a difficult time, and we are so grateful for your participation and collaboration. We hope that we can get this video out not long after and far beyond this workshop, as well, so that others can be able to watch it if circumstances didn’t allow them to be part of it today.
As part of the workshop, we have addressed critical topics at the intersection of sleep and suicidology fields, as well as a lot of specialties and sub-fields within each of these different areas of scientific investigation - and have discussed a broad array of findings and cutting-edge research underway pointing to shared mechanisms underlying risk, highlighting exciting areas of promise to advance the study of sleep as a warning sign, screening tool, and a novel therapeutic target - all looking very promising for next steps in suicide prevention.
I hope this will invite future investigation and multidisciplinary collaborations to not only attract new work but also investigators to this promising new area, as well as different guidelines from NIMH to advance understanding and prioritization in this area.
One thing that came up that was very consistent, and I think was felt throughout the workshop, is the need for more discussion and more time. Each person was tasked with an incredibly tall order to be able to put their deep expertise in multiple fields within a 15-minute talk, which is not nearly enough, but even the discussion itself - I think we could speak so much more about it, and we are just so grateful for everyone to at least give us all a primer on this work.
Our goal is to synthesize the findings from the workshop, as I mentioned before, with discussion questions that we have answered today or that may not have been able to be answered or sent to us, so that we can help delineate mechanisms and underlying risk factors in this area and then address critical gaps to guide innovation and maybe next steps in the future - perhaps as part of a white paper or a research agenda.
Last, I would like to end by just saying what an honor it has been to Co-Chair this with you and my Co-Chair, Dr. David Leitman. I really take special pride knowing the history of his work and my own experiences in the area.
Just to end with an anecdote, when I first began work in this area, it was met with a lot of resistance throughout. I got all kinds of different questions. It was either resistance, skepticism or confusion - about how could sleep possibly be related to suicidal behaviors? Also, one of my personal favorites: well, that would be too simple if that were true. And I thought, wow, what an incredibly good problem to have, though, if that were the case. And then also that these trials couldn’t be conducted safely and across these incredibly high-risk populations, which have historically, of course, been excluded from clinical trials, limiting our development of different advancements in suicide prevention.
So with that in mind, it gives me a unique sense of pride, and I just couldn’t be more thankful for the spectacular presentations today, as well as the leadership of those both within the conference and those not able to be a part of it. With that, I am excited to hand it off, and again, thank you for this work and our shared mission to impact change in this area. I want to end by thanking everyone, and handing it off to David for closing remarks.
DAVID LEITMAN: Thanks, Rebecca, and thank you to all the panelists for your contributions to what has been a lively and thoughtful workshop. As one of our speakers put it yesterday, the importance of sleep as a risk factor for suicide is that it is proximal, dynamic and modifiable.
From the discussion of this workshop as a whole it’s clear that understanding of the psychological and biological mechanisms by which sleep mediates suicide and its role in sleep intervention presents both challenges and opportunities. Examining the role of sleep as a risk factor as well as a therapeutic target for suicide will continue to be an area of focus for NIMH. I would therefore encourage those of you with ideas or proposals in this area to reach out to us here and share your ideas with us so that we may help you guide these proposals further.
Finally, I would like to thank you all on behalf of the Division of Translational Research. I would like to thank my Co-Chair, Rebecca, the Planning Committee members, and I would also particularly like to thank the coordination production team of Jonelle, Kayla and Andrew from the Bizzell Corporation. Thank you all for joining us here today, and good-bye.
(Whereupon, at 5:01 p.m., the workshop was adjourned.)