In this article, we discuss precision psychiatry for trauma- and stress-related psychopathologies. Precision psychiatry promises personalized screening, interventions' outcome prediction, and more efficient and effective resource allocation. Algorithm-based screening can provide personalized scalable and cost-sparing disorder-likelihood weights supporting decisions. Facilitated by remote assessment via mobile devices, digital health additionally offers scalable and cost-efficient capability to monitor patients' symptom progression and treatment response. Specific features of posttraumatic psychopathology, such as identifiable point-of-onset and typical paths of early development, make it uniquely suitable for precision prognostication and prediction.
According to a recent National Institute of Health definition, precision medicine is “an emerging approach for disease treatment and prevention that accounts for individual variability in genes, environment, and lifestyle to allow care providers and researchers to predict more accurately which treatment and prevention strategies for a particular disease will work in which subset of people. It is in contrast to a ‘one-size-fits-all’ approach, in which disease treatment and prevention strategies are developed for the average person, with less consideration for the differences between individuals.”1
The specific implementation of the above in the field of mental health, namely precision psychiatry, has become a productive field of research across mental health conditions.2–4 Among the latter, the typical characteristics of trauma- and stress-related disorders further enhance the need for, and the expected yield of, this approach.
Traumatic events are inherently complex, multifactorial, and diverse (eg, combat exposure, sexual assault, traffic accidents, captivity, torture, relocation). The presumed and overlapping pathogenic mechanisms underlying traumatic stress disorders (eg, fear-driven learning, emotion deregulation, alteration in cognition and mood)5 may differ in patients in yet unspecified ways. The causes of stress-vulnerability or resilience are similarly complex and multilayered, encompassing genetic,1,6 metabolomic,7 immune,8 psychological,5 and environmental9 contributing factors.
Furthermore, the clinical presentations of stress-triggered psychopathology (ie, posttraumatic stress disorder [PTSD] formal symptom criteria, depression, or anxiety) are eminently heterogeneous and unstable. They vary with time,10 circumstances (eg, exposure to reminders such as anniversary dates), and eventual therapeutics, and all these variations (eg, treatment success and stability) may differ in patients.
Such levels of complexity defy the ability of conventional multivariable statistics to provide personalized risk assessment, predict treatment outcome, and observe symptom progression. Instead, data-intensive computational science and advanced biostatistical modelling1 provides many of the necessary tools for precision psychiatry.
Advanced biostatistical models, and particularly machine learning (ML), were made possible by the exponential increase in computing power in recent decades. Until recently, researchers were forced to choose from only a limited number of preferred hypotheses that could be tested for each disorder. Such choices relied heavily on expertise, prior knowledge, and time-honored diagnostic templates and hypothetical mechanisms such as PTSD and the hypothalamic–pituitary–adrenal axis via peripheral blood, urine, and saliva hormones. Limitations to computing power additionally precluded a full account of the complex contextual dependence of each “main” or hypothesized variable.
In contrast, current computational power allows a researcher to simultaneously test hundreds or thousands of hypothetical links within a given dataset, both “traditional” and newly discovered ones. The ability to question the data in such a way gave birth to data-driven research, and within data-driven research to repeatedly experiment with testing and confirming and gaining new knowledge. Thus, ML can rapidly expand current knowledge and understanding of disorders' risk factors, pathogenesis, natural course, and responses to treatment. Within the domain of stress disorders, ML-based studies have already mapped PTSD phenotypical, etiological, and temporal course heterogeneities and their implication for case identification and treatment (see below).
This article outlines the recent advances in precision psychiatry of trauma- and stress-related psychopathology achieved by data-intensive computational approaches. Because “precision psychiatry” has many meanings, we separately discuss precision approaches to diagnosis, prognosis, and forecasting treatment outcomes using a newly proposed distinction between prognostic and predictive models.11 After introducing the reader to the principles of data-driven methods and ML, we present new ML-based research addressing its implementation as a clinical decision-support tool for (1) forecasting clinical symptom development, (2) predicting responses to early intervention, and (3) digitally monitoring of patients' clinical progression.12
Precision Psychiatry and PTSD
PTSD poses important challenges and opportunities for precision psychiatry. PTSD symptoms encompass heterogeneous neurobehavioral domains such as memory, mood, cognitive alterations, and emotion regulation.5,13,14 Additionally, 636,120 permutations of current Diagnostic and Statistical Manual of Mental Disorders, fifth edition,15 criteria are compatible with a diagnosis of PTSD,13 so symptom presentation may greatly vary in patients. Moreover, PTSD symptoms, trajectories greatly differ in trauma survivors, some of whom recover quickly, whereas others may not recover for years.16 Indeed, among the numerous survivors with initial PTSD symptoms few develop clinically pertinent relevant symptoms (ie, meet formal PTSD criteria, show clinically significant distress or dysfunction). A recent pooled-analysis combining six longitudinal studies of adult civilian survivors of trauma (N = 3,083) who were admitted to emergency departments (ED) and observed for more than 9 months showed that most of them report persistently low symptoms (64.5%) or remitting symptoms over time (16.9%), whereas a minority shows moderate (6.7%) or high PTSD symptoms (6.5%) or a delayed-onset of PTSD (5.5%).14
Given the inherent variability in the symptom presentation, traditional analytic modalities—providing reasonable estimates of group average longitudinal course, prognosis, and treatment effect—cannot properly estimate the individual risk and response to treatment. Hence, the need for “individualized” or “precision” approaches to risk assessment, understanding of biological diversity, and predicting treatment outcomes is increasingly recognized and, therefore, novel computational and methodological approaches are needed for precision psychiatry.
As with other mental disorders, PTSD biological heterogeneity has been recognized and codified by the National Institute of Mental Health Research Domain Criteria (RDoC),17 a research template that defines cross-disorder domains of neurobehavioral dysfunction for use in psychiatric research. The introduction of the RDoC in 2013 has productively emphasized the multiplicity of pathogenic and pathophysiological component underlying mental disorders (including PTSD), and numerous studies have used the RDoC matrix as a source of a priori hypotheses for hypotheses-driven research. The RDoC translation to precision psychiatry, however, is incomplete without computational models commensurate with its predicted complexity. Indeed, the RDoC matrix, which has 6 domains, 19 subdomains, and 8 levels of expression, from genes to self-report, calls for models addressing its various combinations in groups of patients and in individuals.
The Rationale for Precision Prevention of PTSD
A primary mode of prevention is the early detection of risks. Whereas primary prevention aims to increase resistance to potential trauma exposure or to prevent the trauma itself, secondary prevention aims to reduce the risk of PTSD development once exposure to traumatic events, such as natural disasters, serious injury, or sexual violence has already happened.
Because trauma exposure is an identifiable event before PTSD onset, the first hours to weeks in the aftermath of traumatic events provide opportunities for early prevention of clinical PTSD symptoms.18 In the United States, it is estimated that up to 40 million survivors of trauma are admitted to the ED every year.19 Lacking individual forecasting, blanket interventions offered to all survivors of trauma who presented to the ED have generally failed.20 Similarly, neither psychological interventions21 nor pharmacological interventions22 yielded sufficient benefit for their general recommendation in clinical practice.20,23 Moderate evidence exists, however, for early administration of hydrocortisone according to which 7 to 13 patients must be treated to prevent the incidence of PTSD in 1 patient.24 A study by Galatzer-Levy et al.25 found that cortisol was a relevant prognostic factor only in a subgroup of people (ie, those who had experienced childhood trauma). This indicates that treatment with hydrocortisone may have more potential when targeted at subgroups than in the overall population of survivors of trauma and is an important step toward precision psychiatry.
To improve the clinical utility of early interventions, the allocation of early interventions must be precisely targeted to survivors of trauma with the highest risk of PTSD.3,18,26 Several prognostic factors have been studied to identify patients in EDs at the highest risk of PTSD. For example, the acute stress reaction as measured by clinical symptoms of acute stress disorder26 or biological measures such as skin conductance27 is a prognostic indicator of PTSD risk. However, no single prognostic factor provides sufficient sensitivity (ie, the ability to correctly identify those people with PTSD; true positive rate) and specificity (ie, the ability to correctly identify those people without PTSD; true negative rate) to be used as a clinical screening tool in practice. A new alternative approach to overcome this difficulty is the use of computational ML methods that integrate multiple risk factors in a data-driven way.3,26 Although individualized treatment recommendations are not yet possible, risk stratification using automatable algorithms is particularly promising as clinical screening performed by ED personnel can require significant additional resources in the ED for the implementation of screening procedures. In contrast, ML methods have the potential to develop and implement algorithms for the automated screening of patients in the ED for PTSD risk using the information present in the electronic health record (EHR) at the time of an ED visit.3,26,28
In addition, ML methods provide techniques to mitigate limitations of conventional statistical analyses that arise when datasets encompass more prognostic factors than research participants (so-called “high-dimensional datasets”) or when prognostic factors contain high degrees of mutual information (so-called “multicollinearity”).12 For instance, in the conventional statistical analysis of a randomized controlled trial, the treatment effect is typically reported as the average response of all research participants in the intervention arm compared to the average response in the control arm. In contrast, ML methods allow examining the so-called heterogeneous treatment effect that systematically explores whether the observed treatment effect varies in subgroups and which variables predict the variation in the treatment effect.29 The knowledge that the treatment response may vary in particular subgroups can then allow the clinician to make better informed clinical decisions in the individual case as compared to only focusing on the average treatment response.
In the proximate future, the potential of ML methods is valuable for prognostic screening using EHR data and for identification of subgroups that respond best to treatment both illustrate that it may yield important improvements for clinical prognosis of the expected disease course and for more precise clinical prediction of treatment response (Figure 1).
Schematic depiction of the automated algorithm-based risk stratification for predicting individualized treatment allocation.
ML Methods for Identifying more Precise Predictors of Treatment Response
Spurred by statistical advancements,29 ML methods can be employed to identify both prognostic markers as well as to identify predictive markers. It is sometimes useful to distinguish between the use of a risk factor as a predictor versus as a prognostic marker. In precision cancer research, the distinction between a prognostic and predictive marker is well-established for regulatory approval.30 In psychiatry, both terms are often used interchangeably. However, to use a risk factor as a predictor of treatment response, when it is, in fact, a prognostic marker of a remitting symptom progression (and vice versa) can obscure the true effect of any medical intervention and can result in clinical and ethical drawbacks.11 ML methods can help to disentangle for each candidate variable its unique prognostic and its predictive value.11 This has a high potential for precision psychiatry. New computational techniques can directly examine interactions between clinical variables and whether a moderation effect between predictive markers and the clinical outcome of interest gives rise to heterogeneous treatment effects.29
These recent advancements in ML, such as causal forests,29 allow studying such moderator variables systematically without the pitfalls of conventional statistics for post-hoc analyses; however, these new approaches need large data-sets. These methods are useful to answer the question “Does clinically important heterogeneity exist in the response to an early intervention administered in trauma survivors to prevent the onset of PTSD?” In clinical trials, observed clinical response of a patient is a function of the treatment effect and of patient characteristics. Because it is often unknown which of the complex molecular, metabolomic, or epigenetic factors implicated in stress pathology are likely predictors of treatment response, it is difficult and often unfeasible to design clinical trials that test for difference of the treatment effect across a priori defined subgroup. These methods are inspired by econometrics, where companies aim to stratify the price of their product depending on customer characteristics (ie, replacing the “average product price” with different prices for different customers—heterogeneous product prices). Similarly, clinicians may not want to treat each patient the same way but may want to stratify the treatment depending on patient characteristics. ML methods allow us to re-analyze data from existing clinical trials (exploratory, hypothesis-generating secondary data analysis) to systematically examine the existence of heterogeneous treatment effects in data-driven ways.29 Using this method, it has been shown that although examples of these methods in precision medicine exist,31 these methods have not yet been applied for precision psychiatry. The data-driven identification of predictive markers bears a very high potential to discover novel hypotheses for the field of precision psychiatry to improve the effectiveness and clinical utility of early intervention for PTSD in survivors of trauma.
ML-Based Identification of More Precise Prognostic Markers
Precision psychiatry can also greatly benefit from the application of ML for the identification of prognostic markers. As early interventions are less effective as blanket interventions, prognostic models are needed to identify survivors of trauma with the highest risk for PTSD.18 However, extensive screening for clinical symptoms of psychological distress, acute stress disorder, or PTSD is not always feasible directly in the aftermath of traumatic events. One potential solution to increase screening rates is through a data-driven prognostic algorithm using the information present in the EHR at the time of an ED visit.28 Such an algorithm can provide an automated clinical read-out of PTSD risk without requiring significant additional resources in the ED.28 Recently, a prognostic model using clinical variables from EHR in combination with a clinical screening instrument for acute stress reaction was developed and validated in survivors of trauma admitted to two Trauma Level 1 EDs, showing high discriminatory accuracy to distinguish risk for PTSD symptoms.3 Identifying risk for PTSD during the evaluation of survivors of trauma in the ED is promising as manifold pathogenic processes take place during this time, such as the neuroendocrine stress sensitization. Using algorithms in the “golden hours” of ED evaluation combined with data from an accessible EHR, does hold promise for providing new opportunities for cost-effective and scalable methods of risk assessment.26
From a clinical perspective, such an algorithm can fill important gaps and improve the clinical potential of early interventions. For example, a central point to consider for the use of the early intervention in survivors of trauma is the question “How many trauma survivors do we need to treat to prevent the occurrence of one incidence of PTSD?” Once the number-needed-to-be-treated is established for a patient population such as survivors of trauma admitted to the ED, then the prognostic model can be used to rule out potential candidates whose prognosis holds that the most likely outcome without intervention is “resilience,” “rapid remission,” or “subthreshold symptomatology.” The prognostic value of automated clinical read-outs can be used to optimize therapeutic strategies using a risk-based approach that aims to target early intervention more precisely to those patients who are most in need of secondary prevention of PTSD onset. Future research in this area of precision psychiatry will have to show whether this approach to precision psychiatry through the development of ML-based prognostic models can improve upon current secondary pharmacological and/or psychological prevention strategies.18,20–24,32
Future Directions for Precision Psychiatry
The future of an individualized (precision) approach to traumatic stress disorders and their treatment lies in the availability of new, large studies with high-quality, comprehensive data that allows the novel methodology to further expand, examine, and uncover new and pertinent knowledge as well as allow us to better create individualized arrays portraying each individual likelihood of becoming ill, remaining ill, and benefitting from interventions.
Large-scale consortia, such as International Consortium for the prediction of PTSD (ICPP) or AURORA (Advancing Understanding of RecOvery afteR traumA),19 are important efforts that will promote precision psychiatry in this sense.
The multi-site AURORA consortium is enrolling a target sample size of 5,000 survivors of trauma from Trauma Level 1 EDs. As the AURORA study encompasses a wide range of self-report, neurocognitive, physiologic, digital phenotyping, psychophysical, neuroimaging, and genomic assessments,19 it yields high potential for leveraging advances in ML methods to characterize “deep phenotypes” for precision psychiatry (ie, comprehensive analysis of phenotypic abnormalities in an individual across multiple levels of organization of human stress physiology).
Large-scale consortia such as ICPP and AURORA are important for precision psychiatry as the computational methods perform better with a larger sample size (given the same degree of data quality). This is particularly important as sophisticated ML methods are necessary to discover new deep phenotypes of PTSD and to explain interindividual heterogeneity in symptom progression. However, capitalizing on clever methodological approaches, there are clinical benefits from methods such as deep learning even on smaller datasets. For instance, transfer learning can be used on smaller datasets in certain circumstances. Transfer learning is the adaption of an already-existing algorithm to perform a specific task of interest that is simply a more specific version of the task that the algorithm is already able to perform (eg, adopting face recognition algorithms to recognize emotional facial expressions).
A new and innovative application of deep learning methods in precision psychiatry is the use of visual and audio signals to monitor and diagnose mental disorders.33 Digital monitoring of stress symptoms offers opportunities to develop digital health applications that run on available mobile devices, which can greatly multiply the possibilities to monitor symptom progression with unprecedented low increases in cost. The use of digital phenotyping in the context of precision psychiatry is an important emerging field. Initial studies have shown that digital biomarkers can be used to accurately classify provisional depression and PTSD diagnosis.33 Such digital biomarkers can be extracted from video and audio data of a short trauma-related open-ended qualitative interview. As many people now have smartphones with high-resolution cameras, accurate diagnosis using this easily obtainable video and audio data offers an unparalleled opportunity to monitor psychopathologies. If prognostic models are successfully developed using such digital data streams, this will make it possible to monitor symptom progression remotely and to perform risk assessment even under adverse conditions to offer timely help at a low-threshold level when it is most needed.
Initial studies on precision psychiatry for stress and trauma psychopathologies have produced promising results demonstrating the high potential of ML methods as a scientific instrument to promote evidence-based prognosis, treatment allocation, and individualized risk stratification. Computational psychiatry is an essential component of this approach and the key to identifying precise predictive and prognostic markers that are going to support clinical decision-making in routine practice.
- National Research Council. National Academies Press. Toward precision medicine: building a knowledge network for biomedical research and a new taxonomy of disease. Accessed December 3, 2020. https://www.nap.edu/catalog/13284/toward-precision-medicine-building-a-knowledge-network-for-biomedical-research
- Schultebraucks K, Qian M, Abu-Amara D, et al. Pre-deployment risk factors for PTSD in active-duty personnel deployed to Afghanistan: a machine-learning approach for analyzing multivariate predictors [published online ahead of print June 2, 2020]. Mol Psychiatry. doi:10.1038/s41380-020-0789-2 [CrossRef] PMID:32488126
- Schultebraucks K, Shalev AY, Michopoulos V, et al. A validated predictive algorithm of post-traumatic stress course following emergency department admission after a traumatic stressor. Nat Med. 2020;26(7):1084–1088. doi:10.1038/s41591-020-0951-z [CrossRef] PMID:32632194
- Fernandes BS, Karmakar C, Tamouza R, et al. Precision psychiatry with immunological and cognitive biomarkers: a multi-domain prediction for the diagnosis of bipolar disorder or schizophrenia using machine learning. Transl Psychiatry. 2020;10(1):162. doi:10.1038/s41398-020-0836-4 [CrossRef] PMID:32448868
- Shalev A, Liberzon I, Marmar C. Post-traumatic stress disorder. N Engl J Med. 2017;376(25):2459–2469. doi:10.1056/NEJMra1612499 [CrossRef] PMID:28636846
- Ng SB, Turner EH, Robertson PD, et al. Targeted capture and massively parallel sequencing of 12 human exomes. Nature. 2009;461(7261):272–276. doi:10.1038/nature08250 [CrossRef] PMID:19684571
- Mellon SH, Gautam A, Hammamieh R, Jett M, Wolkowitz OM. Metabolism, metabolomics, and inflammation in posttraumatic stress disorder. Biol Psychiatry. 2018;83(10):866–875. doi:10.1016/j.bio-psych.2018.02.007 [CrossRef] PMID:29628193
- Michopoulos V, Beurel E, Gould F, et al. Association of prospective risk for chronic PTSD symptoms with low TNF and IFN concentrations in the immediate aftermath of trauma exposure. Am J Psychiatry. 2020;177(1):58–65. doi:10.1176/appi.ajp.2019.19010039 [CrossRef] PMID:31352811
- Koenen KC, Nugent NR, Amstadter AB. Gene-environment interaction in posttraumatic stress disorder: review, strategy and new directions for future research. Eur Arch Psychiatry Clin Neurosci. 2008;258(2):82–96. doi:10.1007/s00406-007-0787-2 [CrossRef] PMID:18297420
- Galatzer-Levy IR, Huang SH, Bonanno GA. Trajectories of resilience and dysfunction following potential trauma: a review and statistical evaluation. Clin Psychol Rev. 2018;63:41–55. doi:10.1016/j.cpr.2018.05.008 [CrossRef] PMID:29902711
- Sechidis K, Papangelou K, Metcalfe PD, Svensson D, Weatherall J, Brown GJB. Distinguishing prognostic and predictive biomarkers: an information theoretic approach. Bioinformatics. 2018;34(19):3365–3376. doi:10.1093/bioinformatics/bty357 [CrossRef] PMID:29726967
- Schultebraucks K, Galatzer-Levy IR. Machine learning for prediction of posttraumatic stress and resilience following trauma: an overview of basic concepts and recent advances. J Trauma Stress. 2019;32(2):215–225. doi:10.1002/jts.22384 [CrossRef] PMID:30892723
- Galatzer-Levy IR, Bryant RA. 636,120 ways to have posttraumatic stress disorder. Perspect Psychol Sci. 2013;8(6):651–662. doi:10.1177/1745691613504115 [CrossRef] PMID:26173229
- Lowe SR, Ratanatharathorn A, Lai BS, et al. Posttraumatic stress disorder symptom trajectories within the first year following emergency department admissions: pooled results from the International Consortium to predict PTSD [published online ahead of print February 3, 2020]. Psychol Med. doi:10.1017/S0033291719004008 [CrossRef] PMID:32008580
- American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders. 5th ed. American Psychiatric Publishing; 2013.
- Koek RJ, Schwartz HN, Scully S, et al. Treatment-refractory posttraumatic stress disorder (TRPTSD): a review and framework for the future. Prog Neuropsychopharmacol Biol Psychiatry. 2016;70:170–218. doi:10.1016/j.pnpbp.2016.01.015 [CrossRef] PMID:26854815
- Insel TR. The NIMH research domain criteria (RDoC) project: precision medicine for psychiatry. Am J Psychiatry. 2014;171(4):395–397. doi:10.1176/appi.ajp.2014.14020138 [CrossRef] PMID:24687194
- Shalev AY, Barbano AC. PTSD: Risk assessment and early management. Psychiatr Ann. 2019;49(7):299–306. doi:10.3928/00485713-20190605-01 [CrossRef]
- McLean SA, Ressler K, Koenen KC, et al. The AURORA Study: a longitudinal, multi-modal library of brain biology and function after traumatic stress exposure. Mol Psychiatry. 2020;25(2)283–296. doi:10.1038/s41380-019-0581-3 [CrossRef] PMID:31745239
- Roberts NP, Kitchiner NJ, Kenardy J, Robertson L, Lewis C, Bisson JI. Multiple session early psychological interventions for the prevention of post-traumatic stress disorder. Cochrane Database Syst Rev. 2019;8(8):CD006869. doi:10.1002/14651858.CD006869.pub3 [CrossRef] PMID:31425615
- Freedman SA. Psychological interventions to prevent PTSD. Psychiatr Ann. 2019;49(7):314–319. doi:10.3928/00485713-20190528-01 [CrossRef]
- Frijling J, Olff M, van Zuiden MJPA. Pharmacological prevention of PTSD: current evidence for clinical practice. 2019;49(7):307–313. doi:10.3928/00485713-20190604-01 [CrossRef]
- Astill Wright L, Sijbrandij M, Sinnerton R, Lewis C, Roberts NP, Bisson JI. Pharmacological prevention and early treatment of post-traumatic stress disorder and acute stress disorder: a systematic review and meta-analysis. Transl Psychiatry. 2019;9(1):334. doi:10.1038/s41398-019-0673-5 [CrossRef] PMID:31819037
- Amos T, Stein DJ, Ipser JC. Pharmacological interventions for preventing post-traumatic stress disorder (PTSD). Cochrane Database Syst Rev. 2014;(7):CD006239. doi:10.1002/14651858.CD006239.pub2 [CrossRef] PMID:25001071
- Galatzer-Levy IR, Ma S, Statnikov A, Yehuda R, Shalev AY. Utilization of machine learning for prediction of post-traumatic stress: a re-examination of cortisol in the prediction and pathways to non-remitting PTSD. Transl Psychiatry. 2017;7(3):e0. doi:10.1038/tp.2017.38 [CrossRef] PMID:28323285
- Vermetten E, McFarlane AC. Predicting future risk of PTSD. Nat Med. 2020;26(7):1012–1013. doi:10.1038/s41591-020-0981-6 [CrossRef] PMID:32632192
- Hinrichs R, van Rooij SJ, Michopoulos V, et al. Increased skin conductance response in the immediate aftermath of trauma predicts PTSD risk. Chronic Stress (Thousand Oaks). 2019;3:2470547019844441. doi:10.1177/2470547019844441 [CrossRef] PMID:31179413
- Patterson BW, Engstrom CJ, Sah V, et al. Training and interpreting machine learning algorithms to evaluate fall risk after emergency department visits. Med Care. 2019;57(7):560–566. doi:10.1097/MLR.0000000000001140 [CrossRef] PMID:31157707
- Wager S, Athey S. Estimation and inference of heterogeneous treatment effects using random forests. J Am Stat Assoc. 2018;113(523):1228–1242. doi:10.1080/01621459.2017.1319839 [CrossRef]
- Abbas A, Schultebraucks K, Galatzer-Levy IR. Digital measurement of mental health: challenges, promises, and future directions. Psychiatr Ann. 2021;51(1):14–20. doi:10.3928/00485713-20201207-01 [CrossRef]
- Scarpa J, Bruzelius E, Doupe P, Le M, Faghmous J, Baum AJJno. Assessment of risk of harm associated with intensive blood pressure management among patients with hypertension who smoke: a secondary analysis of the systolic blood pressure intervention trial. JAMA Netw Open. 2019;2(3):e190005. doi:10.1001/jamanetworkopen.2019.0005 [CrossRef] PMID:30848803
- Kearns MC, Ressler KJ, Zatzick D, Rothbaum BO. Early interventions for PTSD: a review. Depress Anxiety. 2012;29(10):833–842. doi:10.1002/da.21997 [CrossRef] PMID:22941845
- Schultebraucks K, Yadav V, Shalev AY, Bonanno GA, Galatzer-Levy IR. Deep learning-based classification of posttraumatic stress disorder and depression following trauma utilizing visual and auditory markers of arousal and mood [published online ahead of print August 3, 2020]. Psychol Med. doi:10.1017/S0033291720002718 [CrossRef] PMID:32744201