Skip to content

Post-Traumatic Stress Disorder (PTSD) Risk Prediction


Location: Baltimore, Maryland

Sponsored by:
National Institute of Mental Health

On November 2, 2011, NIMH convened a group of experts in post-traumatic stress disorder (PTSD) and various domains of risk and resilience to assess existing and new data on risk assessment, to identify significant gap areas where additional focused research efforts might translate into major improvements, and to assess the readiness of the field for validation studies to refine practical tools that can be used in high risk prevention trials.

Over a lifetime, exposure to potentially traumatic events occurs frequently, and research has identified significant psychiatric consequences of trauma exposure. Although PTSD is among the most common and studied conditions associated with trauma exposure, the majority of exposed and acutely symptomatic persons naturally recover functioning. For the significant minority that does not recover, however, PTSD and related co-morbidity can be quite debilitating. The importance of finding ways to prevent PTSD is underscored by recent reports of high prevalence among veterans of ongoing and past military conflicts, survivors of disasters, terrorist attacks, and shooting incidents, and common interpersonal violence and accidents. While research into the biological dysfunction underlying PTSD has opened avenues for pre-emptive intervention, including exploratory off-label and novel pharmacological intervention, as well as psychotherapy interventions, it is currently not possible to differentiate trauma survivors (early on) who will recover naturally from those who will develop enduring symptoms.

While work on early and pre-emptive intervention must continue, there are barriers to making it efficient. Noteworthy among these barriers is the challenge of needing to recruit generally very large symptomatic samples to know whether any given intervention is facilitating adjustment beyond what might occur without intervention. Moreover, if the field is successful in developing effective early interventions which can become part of routine emergency triage and follow-up care, clinicians and trauma patients will need tools to guide their decisions on if, when, and how aggressively to intervene. Over the course of the one-day meeting, researchers reviewed current scientific knowledge, highlighted critical challenges, and identified research gaps in our scientific understanding of PTSD.

Late-breaking findings on optimizing prediction

In 2009, NIMH and the U.S. Department of Veterans Affairs funded a number of new exploratory risk prediction projects to enhance related work already underway on trauma risk and resilience. In order to harness and integrate the output from these efforts, investigators were invited to share late breaking findings at the meeting to create shared/updated knowledge in various domains of potential risk predictors, and set the stage for identifying advances and gaps. The first presenter, Martha Shumway, Ph.D., discussed the statistical, conceptual, and practical insights learned from previous studies of risk prediction. Noting that ten cases are needed for every predictor examined, Dr. Shumway suggested that there is potential desirability of assessing risk in populations known to be high-risk (e.g. injured crime victims), particularly where some data may already be available to optimize power. Next, Ming Tsuang, M.D., Ph.D., D.Sc. and Stephen Glatt, Ph.D. presented on heritable risk as conveyed in the amount and types of genetic transcripts (messenger RNA) expressed in peripheral blood. Researchers assessed blood samples of U.S. Marines studied both before and after combat deployments. Preliminary findings suggest that although no individual gene-expression level prior to deployment is sufficient to predict PTSD, collectively a larger number of genes may aid in prediction.

The balance of effect size and reach (i.e., how transmittable or useful a risk calculator will be across settings) was the focus of a presentation from Douglas Zatzick, M.D. He emphasized the desirability for easy-to-administer and interpret measures: for example, compare a 0-10 scale of distress patients might rate themselves on upon entering and preparing for discharge from care, with a biomarker test that is costly, may take weeks to run, and that may be complicated to interpret. Nancy Kassam-Adams, Ph.D., is conducting secondary analysis of existing data from many smaller prospective studies of acute trauma in children. Variables under investigation span demographic and time-of-trauma variables—those assessed within one day post-trauma, and those assessed two days to four weeks post-trauma. Dr. Kassam-Adams discussed the challenge of conducting such cost-effective secondary analyses without the benefit of common data elements or measurement strategies across studies. Glenn Saxe, M.D. reviewed his work utilizing network science as another approach to exploring relationships between large sets of complex data (such as PTSD risk factors). Dr. Saxe demonstrated how the principles of network science may be helpful in identifying key clusters of risk factors, and how this, in turn, may aid in determining both markers of risk as well as promising targets for the development of new treatments. Finally, the group heard from Joseph Boscarino, Ph.D., M.P.H. about his work to develop and validate a more robust PTSD risk scale from several large cohort studies and for use in clinical settings to detect PTSD.

'Straw-man' matrix of risk domains and measures

Prior to considering domains and measures of risk for which there is agreement and/or disagreement and identifying areas where additional data are needed, the group reflected on experiences in other areas of medicine that may be instructive for developing robust PTSD research and clinical assessment tools. Discussion focused on how, in contemporary care for cardiovascular disease, persons with symptoms are systematically evaluated on broad dimensions of risk with progressively involved tests. While no single measure can reliably predict disease onset and course, clinicians combine results from multiple tests to maximize positive predictive power and stage interventions according to accumulating risk factors. Cardiovascular researchers and psychometricians have systematically evaluated numerous demographic, biological and behavioral factors, yielding models that assign point values to certain variables based on their relationship to the outcome of interest (e.g., therapy failure, cardiovascular event, death). The process of adjusting variables in and out of prediction models (and the weights assigned to them) allows for the manipulation of specificity and sensitivity of the tool. This approach has led to establishing risk-score cut-points that are used by clinicians and patients in deciding on how aggressively to manage their symptoms. Other methods of classification (e.g., signal detection regression) that emphasize clinical significance of tests/measures in relation to the outcome of interest are increasingly being used to develop decision algorithms.

Building on input received in advance of the meeting and new insights shared in the prior presentations, the group reviewed a straw-man matrix of risk predictors ranging from those easily and routinely collected (e.g., demographic factors) to those that may be challenging to obtain or implement in many healthcare settings (e.g., biomarkers). There was some agreement around data elements that could be used in PTSD risk prediction. Although many common elements are of limited utility when examined individually, and though the field lacks an agreed upon approach to weighting measures, some combination of this collection of measures may approximate an initial crude screen for those most at risk. Participants proposed that a number of existing measures, if optimized, may be sufficiently sensitive to identify eight out of ten trauma-exposed persons likely to develop a clinically meaningful adjustment problem (e.g., PTSD, major depressive disorder). Variables identified for consideration in an initial screen following trauma exposure included:

  • Sleep quality before and after trauma
  • Trauma type (intentional or not)
  • Peri-traumatic tonic immobility
  • Perceived sense of threat
  • Perceived self efficacy
  • Cognitive flexibility
  • Perceived/anticipated support/help
  • Social support before during and after
  • Family/social unit cohesion
  • Initial PTSD symptoms: psychological and physiological distress
  • Post trauma cognitive factors: how one remembers and thinks about it
  • Heritable risk: personal and family history of psychiatric, alcohol and drug problems
  • Childhood adversity/trauma
  • Predictors of future trauma: personality, alcohol, drug use, patterns of services use
  • Meaning attached to trauma experience
  • Parental stress related to child wellbeing
  • Prior perceived trauma (not exposure to potentially traumatic events)
  • Sense of control: over especially grotesque events and perceived threat
  • Negative emotionality
  • Cognitive appraisal style
  • Exposure characteristics
  • Personality factors
  • Difference score on a 0-10 scale of distress at emergency department (ED) admission vs. leaving the ED/discharge
  • Difference score on a measure of how upset a patient was in the ED assessed 24 hours later, and how they feel several days or weeks later

The group embraced an approach modeled on how heart health/disease risk calculators have been developed by using data from the Framingham Heart Study, noting that there are many useful measures of risk currently available. They also noted the need to develop a parsimonious set of measures and weighting strategies from among the highly correlated and overlapping list of identified measures. Discussions touched on optimal sensitivity and specificity, as well as clinical significance. The group discussed the likely need for more than one approach or even two staged approaches based on timing (e.g., within hours versus days/weeks of trauma), setting/context (e.g., battlefield, state-side military hospital, civilian ED, post-disaster community outreach), and purpose (e.g., initial low cost, easy to administer screen of many persons to ascertain who should be monitored further requires lower specificity and higher sensitivity, whereas next level care may require more intensive batteries of measures to guide how aggressively to intervene with persons in need would require greater specificity).

Identification of gaps

Earlier discussions afforded participants the opportunity to identify risk measures that they believed should be considered for inclusion in a PTSD risk calculator. This session focused on revisiting topics mentioned during the day where there is interest and promise, but where additional research is needed to improve prediction. Participants observed the following:

  • Current data typically identify psychological/behavioral measures, and not biological measures, as useful in risk prediction; several biologically plausible measures have not been adequately examined.
  • Many studies have evaluated risk predictors largely in isolation from other potential predictors, seeking to establish the predictive utility of one measure or construct; few studies have attempted to compare combinations of factors to optimize prediction.
  • Researchers have employed slightly different measures of the same factor (i.e., numerous symptom scales have been used to assess acute symptoms); little effort has gone into identifying a common/core approach.
  • Some studies have assessed risk within hours of trauma exposure and others have assessed changes in various indices of risk over time; few have attempted two-stage screening and triage starting in acute care settings and transitioning to home/other locations in the same samples.
  • There appears to be adequate existing data on psychometric/psychological and behavioral measures from many different studies involving diverse populations and contexts; a major contribution to the field would be to conduct cross-study analyses to reduce multiple overlapping/redundant measures to a core set that explain the most variance in outcome most consistently and to specify the optimal time since trauma for administration. Beyond providing a much-needed parsimonious tool for subsequent research, this type of effort could sharpen hypotheses regarding key biological indices that could in turn be developed into risk predictors and/or as targets in studies focused on causes, biological dysfunction, and pre-emptive interventions for PTSD.

Participants also noted several gaps in knowledge and identified numerous putative risk constructs and measures worthy of additional focused research. These included constructs, measures, and approaches for capturing data on:

  • Moral injury—a feeling that all that is good is gone; an existential crisis
  • Non-ideographic “objective” characterization of trauma—i.e., not based on unique elements of the individual's personal experience
  • Aloneness—who is present at the time of trauma and when trauma/injury is treated – particularly when children are involved.
  • 'Embedded-ness'—the degree to which individuals are enmeshed in a social network
  • Biology of change—e.g., heart rate; blood pressure; cortisol; Norepinephrine; gene expression; DNA methylation; central nervous system changes, gene x environment profiles
  • Biomarkers of unfolding neurobiology of PTSD (e.g., the use of EKG and blood tests to identify heart damage) to detect changes heading to PTSD
  • Stress tolerance or ‘Total stress'—the ability, post-trauma, to characterize how well an individual's biological stress response system was functioning prior to the trauma
  • Ability to attend to external and internal needs—frontal lobe/executive control function

Next steps/Wrap-up

Participants discussed their views on advancing the state of the science on predicting PTSD. Participants emphasized the practical need to engage in secondary analyses of existing data to generate a robust, parsimonious, and relatively uncomplicated core measurement approach. It was also noted that further development of multiple putative biological markers of risk would yield valuable data in the service of improving post-traumatic psychopathology classification and prediction, as well as understanding the biological dysfunction underlying PTSD and new treatment development.

For more information, please contact Farris Tuma, Sc.D., or Susan Borja, Ph.D.