Faculty Bibliography

Introduction/UNASSIGNED:Motor abnormalities have been shown to be a distinct component of schizophrenia symptomatology. However, objective and scalable methods for assessment of motor functioning in schizophrenia are lacking. Advancements in machine learning-based digital tools have allowed for automated and remote "digital phenotyping" of disease symptomatology. Here, we assess the performance of a computer vision-based assessment of motor functioning as a characteristic of schizophrenia using video data collected remotely through smartphones. Methods/UNASSIGNED:Eighteen patients with schizophrenia and 9 healthy controls were asked to remotely participate in smartphone-based assessments daily for 14 days. Video recorded from the smartphone front-facing camera during these assessments was used to quantify the Euclidean distance of head movement between frames through a pretrained computer vision model. The ability of head movement measurements to distinguish between patients and healthy controls as well as their relationship to schizophrenia symptom severity as measured through traditional clinical scores was assessed. Results/UNASSIGNED:= 0.04), primarily with negative symptoms of schizophrenia. Conclusions/UNASSIGNED:Remote, smartphone-based assessments were able to capture meaningful visual behavior for computer vision-based objective measurement of head movement. The measurements of head movement acquired were able to accurately classify schizophrenia diagnosis and quantify symptom severity in patients with schizophrenia.

Each year, more than 800,000 persons die by suicide, making it a leading cause of death worldwide. Recent innovations in information and communication technology may offer new opportunities in suicide prevention in individuals, hereby potentially reducing this number. In our project, we design digital indices based on both self-reports and passive mobile sensing and test their ability to predict suicidal ideation, a major predictor for suicide, and psychiatric hospital readmission in high-risk individuals: psychiatric patients after discharge who were admitted in the context of suicidal ideation or a suicidal attempt, or expressed suicidal ideations during their intake. Specifically, two smartphone applications -one for self-reports (SIMON-SELF) and one for passive mobile sensing (SIMON-SENSE)- are installed on participants' smartphones. SIMON-SELF uses a text-based chatbot, called Simon, to guide participants along the study protocol and to ask participants questions about suicidal ideation and relevant other psychological variables five times a day. These self-report data are collected for four consecutive weeks after study participants are discharged from the hospital. SIMON-SENSE collects behavioral variables -such as physical activity, location, and social connectedness- parallel to the first application. We aim to include 100 patients over 12 months to test whether (1) implementation of the digital protocol in such a high-risk population is feasible, and (2) if suicidal ideation and psychiatric hospital readmission can be predicted using a combination of psychological indices and passive sensor information. To this end, a predictive algorithm for suicidal ideation and psychiatric hospital readmission using various learning algorithms (e.g., random forest and support vector machines) and multilevel models will be constructed. Data collected on the basis of psychological theory and digital phenotyping may, in the future and based on our results, help reach vulnerable individuals early and provide links to just-in-time and cost-effective interventions or establish prompt mental health service contact. The current effort may thus lead to saving lives and significantly reduce economic impact by decreasing inpatient treatment and days lost to inability.

Medication non-adherence represents a significant barrier to treatment efficacy. Remote, real-time measurement of medication dosing can facilitate dynamic prediction of risk for medication non-adherence, which in-turn allows for proactive clinical intervention to optimize health outcomes. We examine the accuracy of dynamic prediction of non-adherence using data from remote real-time measurements of medication dosing. Participants across a large set of clinical trials (n = 4,182) were observed via a smartphone application that video records patients taking their prescribed medication. The patients' primary diagnosis, demographics, and prior indication of observed adherence/non-adherence were utilized to predict (1) adherence rates ≥ 80% across the clinical trial, (2) adherence ≥ 80% for the subsequent week, and (3) adherence the subsequent day using machine learning-based classification models. Empirically observed adherence was demonstrated to be the strongest predictor of future adherence/non-adherence. Collectively, the classification models accurately predicted adherence across the trial (AUC = 0.83), the subsequent week (AUC = 0.87) and the subsequent day (AUC = 0.87). Real-time measurement of dosing can be utilized to dynamically predict medication adherence with high accuracy.

This article reports on a study aimed to elucidate the complex etiology of post-traumatic stress (PTS) in a longitudinal cohort of police officers, by applying rigorous computational causal discovery (CCD) methods with observational data. An existing observational data set was used, which comprised a sample of 207 police officers who were recruited upon entry to police academy training. Participants were evaluated on a comprehensive set of clinical, self-report, genetic, neuroendocrine and physiological measures at baseline during academy training and then were re-evaluated at 12 months after training was completed. A data-processing pipeline-the Protocol for Computational Causal Discovery in Psychiatry (PCCDP)-was applied to this data set to determine a causal model for PTS severity. A causal model of 146 variables and 345 bivariate relations was discovered. This model revealed 5 direct causes and 83 causal pathways (of four steps or less) to PTS at 12 months of police service. Direct causes included single-nucleotide polymorphisms (SNPs) for the Histidine Decarboxylase (HDC) and Mineralocorticoid Receptor (MR) genes, acoustic startle in the context of low perceived threat during training, peritraumatic distress to incident exposure during first year of service, and general symptom severity during training at 1 year of service. The application of CCD methods can determine variables and pathways related to the complex etiology of PTS in a cohort of police officers. This knowledge may inform new approaches to treatment and prevention of critical incident related PTS.

BACKGROUND:Visual and auditory signs of patient functioning have long been used for clinical diagnosis, treatment selection, and prognosis. Direct measurement and quantification of these signals can aim to improve the consistency, sensitivity, and scalability of clinical assessment. Currently, we investigate if machine learning-based computer vision (CV), semantic, and acoustic analysis can capture clinical features from free speech responses to a brief interview 1 month post-trauma that accurately classify major depressive disorder (MDD) and posttraumatic stress disorder (PTSD). METHODS:N = 81 patients admitted to an emergency department (ED) of a Level-1 Trauma Unit following a life-threatening traumatic event participated in an open-ended qualitative interview with a para-professional about their experience 1 month following admission. A deep neural network was utilized to extract facial features of emotion and their intensity, movement parameters, speech prosody, and natural language content. These features were utilized as inputs to classify PTSD and MDD cross-sectionally. RESULTS:Both video- and audio-based markers contributed to good discriminatory classification accuracy. The algorithm discriminates PTSD status at 1 month after ED admission with an AUC of 0.90 (weighted average precision = 0.83, recall = 0.84, and f1-score = 0.83) as well as depression status at 1 month after ED admission with an AUC of 0.86 (weighted average precision = 0.83, recall = 0.82, and f1-score = 0.82). CONCLUSIONS:Direct clinical observation during post-trauma free speech using deep learning identifies digital markers that can be utilized to classify MDD and PTSD status.

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 treatment^6-9 to mitigate subsequent psychopathology in high-risk populations^10,11. 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.

BACKGROUND:Though lifetime exposure to traumatic events is significant, only a minority of individuals develops symptoms of posttraumatic stress disorder (PTSD). Post-trauma alterations in neurocognitive and affective functioning are likely to reflect changes in underlying brain networks that are predictive of PTSD. These constructs are assumed to interact in a highly complex way. The aim of this exploratory study was to apply machine learning models to investigate the contribution of these interactions on PTSD symptom development and identify measures indicative of circuit related dysfunction. METHODS:N = 94 participants admitted to the emergency room of an inner-city hospital after trauma exposure completed a battery of neurocognitive and emotional tests 1 month after the incident. Different machine learning algorithms were applied to predict PTSD symptom severity and clusters after 3 months based. RESULTS:Overall, model accuracy did not differ between PTSD clusters, though the importance of cognitive and emotional domains demonstrated both key differences and overlap. Alterations in higher-order executive functioning, speed of information processing, and processing of emotionally incongruent cues were the most important predictors. CONCLUSIONS:Data-driven approaches are a powerful tool to investigate complex interactions and can enhance the mechanistic understanding of PTSD. The study identifies important relationships between cognitive processing and emotion recognition that may be valuable to predict and understand mechanisms of risk and resilience responses to trauma prospectively.

Importance/UNASSIGNED:Suicide is a public health problem, with multiple causes that are poorly understood. The increased focus on combining health care data with machine-learning approaches in psychiatry may help advance the understanding of suicide risk. Objective/UNASSIGNED:To examine sex-specific risk profiles for death from suicide using machine-learning methods and data from the population of Denmark. Design, Setting, and Participants/UNASSIGNED:A case-cohort study nested within 8 national Danish health and social registries was conducted from January 1, 1995, through December 31, 2015. The source population was all persons born or residing in Denmark as of January 1, 1995. Data were analyzed from November 5, 2018, through May 13, 2019. Exposures/UNASSIGNED:Exposures included 1339 variables spanning domains of suicide risk factors. Main Outcomes and Measures/UNASSIGNED:Death from suicide from the Danish cause of death registry. Results/UNASSIGNED:A total of 14 103 individuals died by suicide between 1995 and 2015 (10 152 men [72.0%]; mean [SD] age, 43.5 [18.8] years and 3951 women [28.0%]; age, 47.6 [18.8] years). The comparison subcohort was a 5% random sample (n = 265 183) of living individuals in Denmark on January 1, 1995 (130 591 men [49.2%]; age, 37.4 [21.8] years and 134 592 women [50.8%]; age, 39.9 [23.4] years). With use of classification trees and random forests, sex-specific differences were noted in risk for suicide, with physical health more important to men's suicide risk than women's suicide risk. Psychiatric disorders and possibly associated medications were important to suicide risk, with specific results that may increase clarity in the literature. For example, stress disorders among unmarried men older than 30 years were important factors for suicide risk in the presence of depression (risk, 0.54). Generally, diagnoses and medications measured 48 months before suicide were more important indicators of suicide risk than when measured 6 months earlier. Individuals in the top 5% of predicted suicide risk appeared to account for 32.0% of all suicide cases in men and 53.4% of all cases in women. Conclusions and Relevance/UNASSIGNED:Despite decades of research on suicide risk factors, understanding of suicide remains poor. In this study, the first to date to develop risk profiles for suicide based on data from a full population, apparent consistency with what is known about suicide risk was noted, as well as potentially important, understudied risk factors with evidence of unique suicide risk profiles among specific subpopulations.