ArticlePublisher preview available

Validated Predictive Algorithm of Posttraumatic Stress Course following Emergency Department Admission after a Traumatic Stressor

Authors:
Katharina Schultebraucks at Columbia University
Katharina Schultebraucks
Arieh Y Shalev at NYU Langone Medical Center
Arieh Y Shalev
Vasiliki Michopoulos at Emory University
Vasiliki Michopoulos
Corita R. Grudzen
Corita R. Grudzen
Corita R. Grudzen
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Show all 14 authorsHide
Read publisher preview
Download citation
To read the full-text of this research, you can request a copy directly from the authors.

Abstract and Figures

Annually, approximately 30 million patients are discharged from the emergency department (ED) after a traumatic event1. These patients are at substantial psychiatric risk, with approximately 10–20% developing one or more disorders, including anxiety, depression or post-traumatic stress disorder (PTSD)2–4. At present, no accurate method exists to predict the development of PTSD symptoms upon ED admission after trauma5. Accurate risk identification at the point of treatment by ED services is necessary to inform the targeted deployment of existing treatment6–9 to mitigate subsequent psychopathology in high-risk populations10,11. This work reports the development and validation of an algorithm for prediction of post-traumatic stress course over 12 months using two independently collected prospective cohorts of trauma survivors from two level 1 emergency trauma centers, which uses routinely collectible data from electronic medical records, along with brief clinical assessments of the patient’s immediate stress reaction. Results demonstrate externally validated accuracy to discriminate PTSD risk with high precision. While the predictive algorithm yields useful reproducible results on two independent prospective cohorts of ED patients, future research should extend the generalizability to the broad, clinically heterogeneous ED population under conditions of routine medical care.
Schematic overview of the study design Displayed are the basic steps of the model development and model validation.
Schematic overview of the study design Displayed are the basic steps of the model development and model validation.
… 
Predictive performance in terms of discrimination and calibration In panel a, the ROC curve shows the specificity and the sensitivity of the predictions on the training set (blue line) and the external validation set (orange line) and is accompanied by a calibration plot for the predicted probabilities on the training set (orange line and blue bars) and the external validation set (red line and green bars) in panel b. The bars in the calibration plot in panel (b) displays the predicted probabilities in 10 bins [0, 10%], (10%, 20%],…, (90%, 100%], whereas the lines visualize the predicted probabilities in two bins (low vs. high probability).
Predictive performance in terms of discrimination and calibration In panel a, the ROC curve shows the specificity and the sensitivity of the predictions on the training set (blue line) and the external validation set (orange line) and is accompanied by a calibration plot for the predicted probabilities on the training set (orange line and blue bars) and the external validation set (red line and green bars) in panel b. The bars in the calibration plot in panel (b) displays the predicted probabilities in 10 bins [0, 10%], (10%, 20%],…, (90%, 100%], whereas the lines visualize the predicted probabilities in two bins (low vs. high probability).
… 
Variable importance for the training set a, Variable importance using SHAP for the training set based on EMR data plus ISRC and PDEQ. The SHAP value is calculated for each feature by comparing what the model’s prediction would be without the feature and with the feature in every possible order of adding the feature to the model. The bar plot shows the mean absolute SHAP value per feature. The larger the SHAP value, the more important the feature is to discriminate between the non-remitting and resilient trajectory. b, SHAP summary dot plot (for the same analysis as in a) displaying features that influence model predictions of positive outcome (non-remitting class) the most. The higher the SHAP value of a feature, the higher the log odds of a non-remitting PTSD trajectory. Features are first sorted by their global impact (y axis). For every individual in the sample, a dot represents the attribution value for each feature from low (blue) to high (red). The density of the plot shows that the ISRC is the most important predictor and that higher ISRC levels give rise to higher SHAP values (higher probability to be in the non-remitting class) displayed on the x axis. Chloride (which is the most important predictor in the external test set; Supplementary Fig. 7) shows that a higher score increases the likelihood (a log odds ratio) of being assigned to the non-remitting trajectory by the model (higher SHAP value). c, SHAP variable importance for the training set using EMR data plus the four most predictive ISRC items. d, SHAP summary dot plot for the respective analysis from c. Self-report measures are colored in dark blue in a–c.
Variable importance for the training set a, Variable importance using SHAP for the training set based on EMR data plus ISRC and PDEQ. The SHAP value is calculated for each feature by comparing what the model’s prediction would be without the feature and with the feature in every possible order of adding the feature to the model. The bar plot shows the mean absolute SHAP value per feature. The larger the SHAP value, the more important the feature is to discriminate between the non-remitting and resilient trajectory. b, SHAP summary dot plot (for the same analysis as in a) displaying features that influence model predictions of positive outcome (non-remitting class) the most. The higher the SHAP value of a feature, the higher the log odds of a non-remitting PTSD trajectory. Features are first sorted by their global impact (y axis). For every individual in the sample, a dot represents the attribution value for each feature from low (blue) to high (red). The density of the plot shows that the ISRC is the most important predictor and that higher ISRC levels give rise to higher SHAP values (higher probability to be in the non-remitting class) displayed on the x axis. Chloride (which is the most important predictor in the external test set; Supplementary Fig. 7) shows that a higher score increases the likelihood (a log odds ratio) of being assigned to the non-remitting trajectory by the model (higher SHAP value). c, SHAP variable importance for the training set using EMR data plus the four most predictive ISRC items. d, SHAP summary dot plot for the respective analysis from c. Self-report measures are colored in dark blue in a–c.
… 
Figures - available from: Nature Medicine
This content is subject to copyright. Terms and conditions apply.
A preview of this full-text is provided by Springer Nature.
Learn more
Content available from Nature Medicine
This content is subject to copyright. Terms and conditions apply.
Letters
https://doi.org/10.1038/s41591-020-0951-z
1Department of Psychiatry, New York University Grossman School of Medicine, New York, NY, USA. 2Vagelos School of Physicians and Surgeons,
Department of Emergency Medicine, Columbia University Irving Medical Center, New York, NY, USA. 3Data Science Institute, Columbia University, New
York, NY, USA. 4Department of Psychiatry and Behavioral Sciences, Emory University School of Medicine, Atlanta, GA, USA. 5Yerkes National Primate
Research Center, Atlanta, GA, USA. 6Ronald O. Perelman Department of Emergency Medicine, New York University Grossman School of Medicine,
New York, NY, USA. 7Department of Psychiatry and Behavioral Neurosciences, Wayne State University, Detroit, MI, USA. 8Department of Counseling
and Clinical Psychology, Teachers College, Columbia University, New York, NY, USA. 9Center for Alcohol Use Disorder and PTSD, New York University
Grossman School of Medicine, New York, NY, USA. 10Dell Medical School, Department of Psychiatry, University of Texas at Austin, Austin, TX, USA.
11Institute for Early Life Adversity Research, University of Texas at Austin, Austin, TX, USA. 12McLean Hospital, Harvard Medical School, Boston, MA, USA.
13AiCure LLC, New York, NY, USA. ✉e-mail: ks3796@cumc.columbia.edu
Annually, approximately 30 million patients are discharged
from the emergency department (ED) after a traumatic
event1. These patients are at substantial psychiatric risk,
with approximately 10–20% developing one or more disor-
ders, including anxiety, depression or post-traumatic stress
disorder (PTSD)2–4. At present, no accurate method exists to
predict the development of PTSD symptoms upon ED admis-
sion after trauma5. Accurate risk identification at the point
of treatment by ED services is necessary to inform the tar-
geted deployment of existing treatment6–9 to mitigate sub-
sequent psychopathology in high-risk populations10,11. This
work reports the development and validation of an algorithm
for prediction of post-traumatic stress course over 12 months
using two independently collected prospective cohorts of
trauma survivors from two level 1 emergency trauma centers,
which uses routinely collectible data from electronic medical
records, along with brief clinical assessments of the patient’s
immediate stress reaction. Results demonstrate externally
validated accuracy to discriminate PTSD risk with high preci-
sion. While the predictive algorithm yields useful reproduc-
ible results on two independent prospective cohorts of ED
patients, future research should extend the generalizability to
the broad, clinically heterogeneous ED population under con-
ditions of routine medical care.
Previous studies identified multiple trauma-related predic-
tive signals of PTSD risk7,12–17, including aspects of the biologi-
cal stress response18–23, immune response24–26, threat perception,
psychophysiological arousal15,19,27 and psychosocial determinants
of clinical risk28. Many indicators related to these biological sys-
tems and psychosocial indicators are routinely collected in the
ED and logged in the electronic medical records (EMRs), mak-
ing them viable as candidate predictors of risk. Some factors, such
as self-reported psychological stress, are not yet part of the medi-
cal routine and only about 7% of level 1 trauma centers routinely
screen for PTSD29.
Notably, PTSD comes with long-term clinical and pecuniary costs
to both the individual and the healthcare system. While empirically
validated treatments are effective in reducing the risk for PTSD6,8,9,
early prevention strategies are typically not implemented due to the
lack of established methods for timely and reliable risk identifica-
tion11. The ED visit is often the sole contact of trauma survivors
with the healthcare system and the time immediately after trauma
opens a critical window to prevent the development of PTSD11,30.
Accurate identification of risk for PTSD during ED evaluation using
algorithms running on accessible data sources would provide new
opportunities for cost-effective and scalable methods of risk assess-
ment and intervention to reduce the prevalence of PTSD without
posing high additional burden for ED personnel11.
The use of predictive models to integrate multiple post-traumatic
stress (PTS) risk indicators has demonstrated moderate to strong
predictive accuracy on a proof-of-concept level12–14. However, the
frequent lack of external validation in the literature obscures the
generalizability of model performance31,32, ultimately hampering the
implementation of such algorithms in clinical practice. Recognizing
the high clinical need, the US National Institute of Mental Health
has funded a large multi-site consortium that will start to collect
data from independent sites33 suitable to evaluate the reliability
of emerging predictive models of PTS course. Independently of
this effort, large hospital systems are actively working to identify
novel methods that can be integrated into the standard of care to
improve patient outcomes and decrease long-term costs to the hos-
pital system34. Together, there is an indication of both the necessary
research and clinical interest in the development and deployment of
data-driven approaches to predict the clinical risk of psychopathol-
ogy in the context of ED healthcare.
We set out to develop and test the prediction of PTSD symptom
development in a reproducible way across independent samples. At
two independent sites (Supplementary Figs. 1 and 2), ED patients
who reported experience of a traumatic event according to trauma
criterion A35 were enrolled in a prospective longitudinal cohort.
A validated predictive algorithm of post-traumatic
stress course following emergency department
admission after a traumatic stressor
KatharinaSchultebraucks 1,2,3 ✉ , AriehY.Shalev1, VasilikiMichopoulos4,5, CoritaR.Grudzen6,
Soo-MinShin6, JenniferS.Stevens 4, JessicaL.Maples-Keller4, TanjaJovanovic7, GeorgeA.Bonanno8,
BarbaraO.Rothbaum4, CharlesR.Marmar1,9, CharlesB.Nemeroff10,11, KerryJ.Ressler4,12 and
IsaacR.Galatzer-Levy1,13
NATURE MEDICINE | VOL 26 | JULY 2020 | 1084–1088 | www.nature.com/naturemedicine
1084
Content courtesy of Springer Nature, terms of use apply. Rights reserved
... A total of 30 studies were included in this Analysis (Table 1): 12 crosssectional studies 26-37 and 18 longitudinal studies [38][39][40][41][42][43][44][45][46][47][48][49][50][51][52][53][54][55] . Most of the studies focused on post-trauma variables only (n = 26) [26][27][28][29][30][31][32][33][34][35][36][38][39][40][41][42][43][44][45][46][47][48][49][50][51] , whereas one study assessed both pre-trauma and post-trauma variables 53 and three studies focused on pre-trauma variables 52,54,55 . ...
... A total of 30 studies were included in this Analysis (Table 1): 12 crosssectional studies 26-37 and 18 longitudinal studies [38][39][40][41][42][43][44][45][46][47][48][49][50][51][52][53][54][55] . Most of the studies focused on post-trauma variables only (n = 26) [26][27][28][29][30][31][32][33][34][35][36][38][39][40][41][42][43][44][45][46][47][48][49][50][51] , whereas one study assessed both pre-trauma and post-trauma variables 53 and three studies focused on pre-trauma variables 52,54,55 . Twenty-one studies used a classification approach to predict probable PTSD diagnosis (n = 12) 27,28,30,31,34,36,37,44,47,50,51,55 , PTSD trajectory membership from latent growth mixture models (n = 6) 39-42,46,53 , both (n = 2) 45,52 or symptom profile from latent profile analysis (n = 1) 54 . ...
... Only four studies used the gold standard and validated the generalizability of the results using an external sample 43,46,51,55 , and eight studies validated their model in a holdout set 26,30,38,45,47,[50][51][52] . In addition, 13 studies used resampling techniques 26,30,[38][39][40]43,[45][46][47][50][51][52]55 such as crossvalidation (k-folds 30,43,46,51,55 and leave-one-out 26,38,45,50,52 ) and bootstrapping 39,40,47 techniques, 3 used nested cross-validation 28,35,42 and 4 studies did not specify their validation method 29,33,34,53 . ...
View
... Previous research reported a broad range of risk and protective factors for PTSD, including demographic; socio-economic; psychiatric; psychosocial; biological; trauma history; and environmental domains [19][20][21][22]. Recent studies with a computational approach strongly indicate that these risk and protective factors for PTSD interact in dynamic non-linear ways and seem to differ between PTSD symptom trajectories (e.g., [23,24]). Increasing empirical evidence supports the existence four common distinct courses or trajectories of PTSD symptoms following trauma exposure [25]. ...
... A growing number of studies support the potential of machine learning in prognostic risk classification. These studies achieved good accuracy in classifying recently traumatized individuals into their subsequent PTSD diagnostic status and/or symptom trajectory (e.g., [23,24,[26][27][28][29][30]). However, most of these existing prognostic risk classification models cannot be applied beyond acute medical care settings, as they mainly include acute biomedical and hospital patient record information. ...
... The test sets require a sample size of n = 75 (n = 6 cases; n = 69 non-cases) for men, and n = 82 (n = 6 cases; n = 74 non-cases) for women. In order to meet the proposed rule of thumb of at least 25 cases available in the training set, this requires to include a minimum of 31 cases in both samples [24]. Expecting minimally the same allocation ratio of the chronic trajectory relative to the resilient and recovery trajectory as in the TraumaTIPS cohort, this translates into a total sample size of n = (74.964*5) ...
Towards accurate screening and prevention for PTSD (2-ASAP): protocol of a longitudinal prospective cohort study
Article
Full-text available
  • Oct 2024
  • BMC PSYCHIATRY
Background Effective preventive interventions for PTSD rely on early identification of individuals at risk for developing PTSD. To establish early post-trauma who are at risk, there is a need for accurate prognostic risk screening instruments for PTSD that can be widely implemented in recently trauma-exposed adults. Achieving such accuracy and generalizability requires external validation of machine learning classification models. The current 2-ASAP cohort study will perform external validation on both full and minimal feature sets of supervised machine learning classification models assessing individual risk to follow an adverse PTSD symptom trajectory over the course of 1 year. We will derive these models from the TraumaTIPS cohort, separately for men and women. Method The 2-ASAP longitudinal cohort will include N = 863 adults (N = 436 females, N = 427 males) who were recently exposed to acute civilian trauma. We will include civilian victims of accidents, crime and calamities at Victim Support Netherlands; and who were presented for medical evaluation of (suspected) traumatic injuries by emergency transportation to the emergency department. The baseline assessment within 2 months post-trauma will include self-report questionnaires on demographic, medical and traumatic event characteristics; potential risk and protective factors for PTSD; PTSD symptom severity and other adverse outcomes; and current best-practice PTSD screening instruments. Participants will be followed at 3, 6, 9, and 12 months post-trauma, assessing PTSD symptom severity and other adverse outcomes via self-report questionnaires. Discussion The ultimate goal of our study is to improve accurate screening and prevention for PTSD in recently trauma-exposed civilians. To enable future large-scale implementation, we will use self-report data to inform the prognostic models; and we will derive a minimal feature set of the classification models. This can be transformed into a short online screening instrument that is user-friendly for recently trauma-exposed adults to fill in. The eventual short online screening instrument will classify early post-trauma which adults are at risk for developing PTSD. Those at risk can be targeted and may subsequently benefit from preventive interventions, aiming to reduce PTSD and relatedly improve psychological, functional and economic outcomes.
View
... This should be considered when interpreting our findings and represents a limitation of our study where future research would benefit from including questions about treatment history to enhance the understanding of its potential impact on PTSD outcomes. Gains have been made in recent years in integrating various acute variables into predictive models, which can potentially improve early identification of people who will develop PTSD [30,31]. Fifth, we did not use a clinical face-to-face diagnosis for ASD and PTSD. ...
Unveiling the neglected role of the intensity of acute stress disorder in the prediction of full- and sub-threshold posttraumatic stress disorder: looking beyond the diagnosis
Article
Full-text available
  • Dec 2024
  • SOC PSYCH PSYCH EPID
Purpose Exposure to traumatic events may lead to the development of Acute Stress Disorder (ASD) within the first month post-trauma in some individuals, while others may not exhibit ASD symptoms. ASD was introduced as a potential early indicator to identify those at higher risk of developing Posttraumatic Stress Disorder (PTSD), however, PTSD can occur in some individuals even without prior ASD. Assessing ASD post-trauma can assist in identifying those who would most benefit from intervention to prevent later PTSD, yet the predictive power of ASD varies across studies, with intensity of ASD symptoms and subthreshold PTSD often less considered. Methods A prospective cohort study on 426 health workers exposed to the Beirut Port Blast assessed DSM-5 ASD and symptom intensity using self-report questionnaire at two distinct time points: 9–15 and 21–27 days post blast. DSM-5 PTSD was assessed afterwards at 6–7 months via self-report questionnaire post-exposure. Probit models predicted full and subthreshold PTSD. Results Using ASD diagnosis alone, the sensitivity 9–15 days after trauma was better than 21–27 days after trauma (75.68% vs. 58.06%); when stratified by intensity, however, sensitivity increased from 41.66% among those with low intensity to 92% among those with high intensity. Specificity, however, was better 21–27 days after trauma (77.82%) compared with 9–15 days (60.98%). Positive Predictive Value of ASD increased, and Negative Predictive Value decreased, with time since exposure and when adding intensity with diagnosis. ASD diagnosis plus intensity achieved better prediction of PTSD and subthreshold PTSD. Conclusion Screening for PTSD should include ASD and its intensity, improving predictive ability for later PTSD, incorporatingfull threshold and subthreshold PTSD. Specificity increases with time since exposure, suggesting a high rate of false positives when assessing ASD soon after trauma. This highlights the need to prioritize individuals for early preventive measures after trauma.
View
... Whereas previous techniques have suggested the use of model ensembles in unsupervised learning 16,20 , these methods have so far been limited to 'shallow' models with a single hidden layer. Moreover, the application of Explainable AI (XAI) in life sciences [21][22][23] , although widespread, often grapples with complex, multidimensional data. In this context, model ensembles offer a substantial advantage, improving the quality and reliability of feature attribution 24 , thereby aligning with the growing emphasis on transparency and comprehension in AI models used for biological data analysis. ...
Deep profiling of gene expression across 18 human cancers
Article
Full-text available
  • Dec 2024
  • Nat. Biomed. Eng.
Clinical and biological information in large datasets of gene expression across cancers could be tapped with unsupervised deep learning. However, difficulties associated with biological interpretability and methodological robustness have made this impractical. Here we describe an unsupervised deep-learning framework for the generation of low-dimensional latent spaces for gene-expression data from 50,211 transcriptomes across 18 human cancers. The framework, which we named DeepProfile, outperformed dimensionality-reduction methods with respect to biological interpretability and allowed us to unveil that genes that are universally important in defining latent spaces across cancer types control immune cell activation, whereas cancer-type-specific genes and pathways define molecular disease subtypes. By linking latent variables in DeepProfile to secondary characteristics of tumours, we discovered that mutation burden is closely associated with the expression of cell-cycle-related genes, and that the activity of biological pathways for DNA-mismatch repair and MHC class II antigen presentation are consistently associated with patient survival. We also found that tumour-associated macrophages are a source of survival-correlated MHC class II transcripts. Unsupervised learning can facilitate the discovery of biological insight from gene-expression data.
View
... The underlying causes of a successful predictive model may provide data on the likelihood of ADHD risk at the individual level. Studies of ML-based mental health prediction using large-scale data have been growing rapidly over the past few years, including successful prediction of a wide range of disorders such as depression [37][38][39], opioid use disorder [29,40,41], and post-traumatic stress [42]. ADHD prediction using ML has been pioneered using neuroimaging data (for a review, see [43]). ...
Early identification of children with Attention-Deficit/Hyperactivity Disorder (ADHD)
Article
Full-text available
  • Nov 2024
Signs and symptoms of Attention-Deficit/Hyperactivity Disorder (ADHD) are present at preschool ages and often not identified for early intervention. We aimed to use machine learning to detect ADHD early among kindergarten-aged children using population-level administrative health data and a childhood developmental vulnerability surveillance tool: Early Development Instrument (EDI). The study cohort consists of 23,494 children born in Alberta, Canada, who attended kindergarten in 2016 without a diagnosis of ADHD. In a four-year follow-up period, 1,680 children were later identified with ADHD using case definition. We trained and tested machine learning models to predict ADHD prospectively. The best-performing model using administrative and EDI data could reliably predict ADHD and achieved an Area Under the Curve (AUC) of 0.811 during cross-validation. Key predictive factors included EDI subdomain scores, sex, and socioeconomic status. Our findings suggest that machine learning algorithms that use population-level surveillance data could be a valuable tool for early identification of ADHD.
View
... These patterns include distinct groups such as consistently low symptomatology (resilient), initial high symptoms followed by gradual improvement (recovery), and persistent high symptomatology (chronic). Similarly, Schultebraucks et al. [12,13,14] have contributed significant findings on predicting PTSD trajectories using a variety of biomedical and psychosocial data, which could potentially refine our understanding of PTSD development over time. Specially, Galatzer-Levy et al. [15] identified three distinct trajectories (nonremitting, slow remitting, and rapid remitting) during a 15-month follow-up, evaluating predictors from emergency room assessments and 10 days postevents between remitters and nonremitters. ...
Predictors of 2-Year Trajectory of Post-Traumatic Stress Disorder Following Physical Injury
Article
Full-text available
Objectives. This study investigated post-traumatic stress disorder (PTSD) trajectories and their predictors over a 2-year period, in individuals recovering from physical injuries. Materials and Methods. Between June 2015 and January 2021, 1,142 patients from a South Korean University Hospital Trauma Center underwent baseline evaluations, including PTSD-related measures and sociodemographic characteristics. They were subsequently followed up for PTSD using the Clinician-Administered PTSD Scale at 3, 6, 12, and 24 months. The analyzed sample consisted of 1,014 patients who were followed up at least once after the baseline and 3-month evaluations. Latent class growth analysis and logistic regression models were used. Results. Five distinctive trajectories were identified: resilient, worsening/recovery, worsening, recovery, and chronic groups. The worsening/recovery trajectory was associated with previous traumatic events and traffic-related injuries, while the worsening trajectory was linked to higher education and elevated depressive symptoms. The recovery trajectory was characterized by female sex, childhood abuse, traffic-related injuries, dissociative subtype, and higher levels of anxiety and depressive symptoms. The chronic trajectory was predicted by the dissociative subtype and heightened anxiety symptoms. Conclusion. These findings highlighted the heterogeneity of PTSD symptom development and, thus, the importance of considering individual characteristics when assessing and addressing PTSD following physical injuries.
View
View
Post‐traumatic stress disorder: evolving conceptualization and evidence, and future research directions
Article
  • Jan 2025
  • WORLD PSYCHIATRY
The understanding of responses to traumatic events has been greatly influenced by the introduction of the diagnosis of post‐traumatic stress disorder (PTSD). In this paper we review the initial versions of the diagnostic criteria for this condition and the associated epidemiological findings, including sociocultural differences. We consider evidence for post‐traumatic reactions occurring in multiple contexts not previously defined as traumatic, and the implications that these observations have for the diagnosis. More recent developments such as the DSM‐5 dissociative subtype and the ICD‐11 diagnosis of complex PTSD are reviewed, adding to evidence that there are several distinct PTSD phenotypes. We describe the psychological foundations of PTSD, involving disturbances to memory as well as to identity. A broader focus on identity may be able to accommodate group and communal influences on the experience of trauma and PTSD, as well as the impact of resource loss. We then summarize current evidence concerning the biological foundations of PTSD, with a particular focus on genetic and neuroimaging studies. Whereas progress in prevention has been disappointing, there is now an extensive evidence supporting the efficacy of a variety of psychological treatments for established PTSD, including trauma‐focused interventions – such as trauma‐focused cognitive behavior therapy (TF‐CBT) and eye movement desensitization and reprocessing (EMDR) – and non‐trauma‐focused therapies, which also include some emerging identity‐based approaches such as present‐centered and compassion‐focused therapies. Additionally, there are promising interventions that are neither psychological nor pharmacological, or that combine a pharmacological and a psychological approach, such as 3,4‐methylenedioxymethamphetamine (MDMA)‐assisted psychotherapy. We review advances in the priority areas of adapting interventions in resource‐limited settings and across cultural contexts, and of community‐based approaches. We conclude by identifying future directions for work on trauma and mental health.
View
View
View
Post-traumatic Stress Disorder Following Acute Stroke
Article
Full-text available
  • Mar 2020
Purpose of Review Stroke is a devastating event that annually affects over 800,000 million individuals in the USA and is associated with significant individual and social costs. In this narrative review, we summarize current evidence regarding post-traumatic stress disorder (PTSD) following acute stroke. Recent Findings In addition to the long-term physical consequences, it is increasingly recognized that psychological distress is common after stroke and transient ischemic attack (TIA). Nearly 1 in 4 survivors of TIA and stroke report elevated symptoms of PTSD in the first year following their cerebrovascular event. Those individuals with PTSD symptoms are at elevated risk for not only sustained psychological distress but also increased risk for non-adherence to medication. Factors in the emergency department, such as crowding, may influence the development of PTSD following acute stroke and TIA. We also summarize the rationale and clinical importance of developing and implementing quantitative predictive models of post-stroke PTSD symptoms in the ED setting. Summary The potential of promising early interventions to prevent PTSD critically hinges on the accurate and precise identification of patients at risk for PTSD. Predictive modeling of PTSD risk may greatly facilitate the prospective management of mental health care in ED patients after stroke.
View
Heightened Biological Stress Response During Exposure to a Trauma Film Predicts an Increase in Intrusive Memories
Article
Full-text available
  • Jul 2019
Some people develop symptoms of posttraumatic stress disorder (PTSD) after having experienced a traumatic event, whereas others do not. Intrusive memories are a cardinal symptom of PTSD and a better understanding of encoding and consolidation of intrusive memory may yield important insights on differences in the response to trauma. The primary aim of this study is to investigate whether psychosocial stress induction (Trier Social Stress Test) versus active control (placebo version) leading to respective biological stress responses during the encoding and consolidation of a film-elicited analogue trauma influences the development of intrusive memories over the course of 7 consecutive days. We hypothesized that the activation of the biological stress system increases the number of intrusive memories over the course of 7 days. This single-blind randomized placebo-controlled study examined 122 young healthy women. Biological stress response was measured by salivary cortisol, salivary α-amylase activity, and heart rate variability. Generalized linear mixed models were used to analyze longitudinal effects of activation of biological stress response on self-reported number of intrusive memories. Cross-validated regularized regression (least absolute shrinkage and selection operator) was applied for data-driven feature selection including known biological and psychological predictors. Corroborating our hypothesis, biological stress-responders to the Trier Social Stress Test reported significantly more intrusive memories after trauma film. A priori designed post hoc tests point at significantly more intrusions on Day 1 and 2 in biological stress responders. Least absolute shrinkage and selection operator regression revealed salivary cortisol, salivary α-amylase activity, heart rate variability, subjectively rated distress, fear, and (on trend level) dissociation during the trauma film as relevant predictors of intrusive memories. A heightened biological stress response in young women is associated with more intrusive memories the first days after experiencing a trauma analogue. (PsycINFO Database Record (c) 2019 APA, all rights reserved).
View
Increased Skin Conductance Response in the Immediate Aftermath of Trauma Predicts PTSD Risk
Article
Full-text available
  • Apr 2019
Background: Exposure to a traumatic event leads to posttraumatic stress disorder (PTSD) in 10-20% of exposed individuals. Predictors of risk are needed to target early interventions to those who are most vulnerable. The objective of the study was to test whether a noninvasive mobile device that measures a physiological biomarker of autonomic nervous system activation could predict future PTSD symptoms. Methods: Skin conductance response (SCR) was collected during a trauma interview in the emergency department within hours of exposure to trauma in 95 individuals. Trajectories of PTSD symptoms over 12 months post-trauma were identified using Latent Growth Mixture Modeling. Results: SCR was significantly correlated with the probability of being in the chronic PTSD trajectory following trauma exposure in the ED (r=0.489, p<0.000001). Lasso regression with elastic net was performed with demographic and clinical measures obtained in the ED, demonstrating that SCR was the most significant predictor of the chronic PTSD trajectory (p<0.00001). Conclusions: The current study is the first prospective study of PTSD showing SCR in the immediate aftermath of trauma predicts subsequent development of chronic PTSD. This finding points to an easily obtained, and neurobiologically informative, biomarker in emergency departments that can be disseminated to predict the development of PTSD.
View
Deep forest
Article
Full-text available
  • Jan 2019
Current deep-learning models are mostly built upon neural networks, i.e. multiple layers of parameterized differentiable non-linear modules that can be trained by backpropagation. In this paper, we explore the possibility of building deep models based on non-differentiable modules such as decision trees. After a discussion about the mystery behind deep neural networks, particularly by contrasting them with shallow neural networks and traditional machine-learning techniques such as decision trees and boosting machines, we conjecture that the success of deep neural networks owes much to three characteristics, i.e. layer-by-layer processing, in-model feature transformation and sufficient model complexity. On one hand, our conjecture may offer inspiration for theoretical understanding of deep learning; on the other hand, to verify the conjecture, we propose an approach that generates deep forest holding these characteristics. This is a decision-tree ensemble approach, with fewer hyper-parameters than deep neural networks, and its model complexity can be automatically determined in a data-dependent way. Experiments show that its performance is quite robust to hyper-parameter settings, such that in most cases, even across different data from different domains, it is able to achieve excellent performance by using the same default setting. This study opens the door to deep learning based on non-differentiable modules without gradient-based adjustment, and exhibits the possibility of constructing deep models without backpropagation. © The Author(s) 2018. Published by Oxford University Press on behalf of China Science Publishing & Media Ltd.
View
Deep Super Learner: A Deep Ensemble for Classification Problems
Chapter
  • Apr 2018
  • Lect Notes Comput Sci
Deep learning has become very popular for tasks such as predictive modeling and pattern recognition in handling big data. Deep learning is a powerful machine learning method that extracts lower level features and feeds them forward for the next layer to identify higher level features that improve performance. However, deep neural networks have drawbacks, which include many hyper-parameters and infinite architectures, opaqueness into results, and relatively slower convergence on smaller datasets. While traditional machine learning algorithms can address these drawbacks, they are not typically capable of the performance levels achieved by deep neural networks. To improve performance, ensemble methods are used to combine multiple base learners. Super learning is an ensemble that finds the optimal combination of diverse learning algorithms. This paper proposes deep super learning as an approach which achieves log loss and accuracy results competitive to deep neural networks while employing traditional machine learning algorithms in a hierarchical structure. The deep super learner is flexible, adaptable, and easy to train with good performance across different tasks using identical hyper-parameter values. Using traditional machine learning requires fewer hyper-parameters, allows transparency into results, and has relatively fast convergence on smaller datasets. Experimental results show that the deep super learner has superior performance compared to the individual base learners, single-layer ensembles, and in some cases deep neural networks. Performance of the deep super learner may further be improved with task-specific tuning.
View
The AURORA Study: a longitudinal, multimodal library of brain biology and function after traumatic stress exposure
Article
  • Nov 2019
  • MOL PSYCHIATR
Adverse posttraumatic neuropsychiatric sequelae (APNS) are common among civilian trauma survivors and military veterans. These APNS, as traditionally classified, include posttraumatic stress, postconcussion syndrome, depression, and regional or widespread pain. Traditional classifications have come to hamper scientific progress because they artificially fragment APNS into siloed, syndromic diagnoses unmoored to discrete components of brain functioning and studied in isolation. These limitations in classification and ontology slow the discovery of pathophysiologic mechanisms, biobehavioral markers, risk prediction tools, and preventive/treatment interventions. Progress in overcoming these limitations has been challenging because such progress would require studies that both evaluate a broad spectrum of posttraumatic sequelae (to overcome fragmentation) and also perform in-depth biobehavioral evaluation (to index sequelae to domains of brain function). This article summarizes the methods of the Advancing Understanding of RecOvery afteR traumA (AURORA) Study. AURORA conducts a large-scale (n = 5000 target sample) in-depth assessment of APNS development using a state-of-the-art battery of self-report, neurocognitive, physiologic, digital phenotyping, psychophysical, neuroimaging, and genomic assessments, beginning in the early aftermath of trauma and continuing for 1 year. The goals of AURORA are to achieve improved phenotypes, prediction tools, and understanding of molecular mechanisms to inform the future development and testing of preventive and treatment interventions.
View
Creating a Learning Health System through Rapid-Cycle, Randomized Testing
Article
  • Sep 2019
  • NEW ENGL J MED
Randomized tests of quality-improvement interventions were embedded within an existing health care system. Quality and efficiency were improved by abandoning ineffective interventions and continuing successful quality-improvement strategies.
View
Association of Prospective Risk for Chronic PTSD Symptoms With Low TNFα and IFNγ Concentrations in the Immediate Aftermath of Trauma Exposure
Article
  • Jul 2019
Objective: Although several reports have documented heightened systemic inflammation in posttraumatic stress disorder (PTSD), few studies have assessed whether inflammatory markers serve as prospective biomarkers for PTSD risk. The present study aimed to characterize whether peripheral immune factors measured in blood samples collected in an emergency department immediately after trauma exposure would predict later chronic development of PTSD. Methods: Participants (N=505) were recruited from a hospital emergency department and underwent a 1.5-hour assessment. Blood samples were drawn, on average, about 3 hours after trauma exposure. Follow-up assessments were conducted 1, 3, 6, and 12 months after trauma exposure. Latent growth mixture modeling was used to identify classes of PTSD symptom trajectories. Results: Three distinct classes of PTSD symptom trajectories were identified: chronic (N=28), resilient (N=160), and recovery (N=85). Multivariate analyses of covariance revealed a significant multivariate main effect of PTSD symptom trajectory class membership on proinflammatory cytokines. Univariate analyses showed a significant main effect of trajectory class membership on plasma concentrations of proinflammatory tumor necrosis factor α (TNFα) and interferon-γ (IFNγ). Concentrations of proinflammatory TNFα and IFNγ were significantly lower in individuals in the chronic PTSD class compared with those in the recovery and resilient classes. There were no significant differences in interleukin (IL) 1β and IL-6 concentrations by PTSD symptom trajectory class. Anti-inflammatory and other cytokines, as well as chemokines and growth factor concentrations, were not associated with development of chronic PTSD. Conclusions: Overall, the study findings suggest that assessing the proinflammatory immune response to trauma exposure immediately after trauma exposure, in the emergency department, may help identify individuals most at risk for developing chronic PTSD in the aftermath of trauma.
View
PTSD: Risk Assessment and Early Management
Article
  • Jul 2019
Posttraumatic stress disorder (PTSD) is a debilitating and pernicious disorder that occurs in a significant proportion of survivors exposed to trauma. Early interventions may mitigate the development of PTSD among survivors who have risk factors. Beyond PTSD risk, survivors' initial distress may require immediate interventions. This article outlines clinically pertinent predictors of PTSD risk that inform early intervention decisions, delineates stress management approaches at the early aftermath of trauma, and discusses barriers to receiving mental health care by those at high risk for PTSD, and ways to mitigate them. [ Psychiatr Ann. 2019;49(7):299–306.]
View
Machine Learning for Prediction of Posttraumatic Stress and Resilience Following Trauma: An Overview of Basic Concepts and Recent Advances
Article
  • Mar 2019
Posttraumatic stress responses are characterized by a heterogeneity in clinical appearance and etiology. This heterogeneity impacts the field's ability to characterize, predict, and remediate maladaptive responses to trauma. Machine learning (ML) approaches are increasingly utilized to overcome this foundational problem in characterization, prediction, and treatment selection across branches of medicine that have struggled with similar clinical realities of heterogeneity in etiology and outcome, such as oncology. In this article, we review and evaluate ML approaches and applications utilized in the areas of posttraumatic stress, stress pathology, and resilience research, and present didactic information and examples to aid researchers interested in the relevance of ML to their own research. The examined studies exemplify the high potential of ML approaches to build accurate predictive and diagnostic models of posttraumatic stress and stress pathology risk based on diverse sources of available information. The use of ML approaches to integrate high-dimensional data demonstrates substantial gains in risk prediction even when the sources of data are the same as those used in traditional predictive models. This area of research will greatly benefit from collaboration and data sharing among researchers of posttraumatic stress disorder, stress pathology, and resilience. © 2019 International Society for Traumatic Stress Studies.
View