PROJECT SUMMARY
Post-traumatic stress disorder (PTSD) often has complex profiles of co-occurring medical conditions and is
associated with high risk of self-harm, including suicidality, which is a leading cause of death, particularly
among Veterans. There is a critical lack of advancement in PTSD pharmacotherapy, as illustrated by increased
use of off-label medications and polypharmacy (multiple drugs used simultaneously) with limited evidence on
their relative risks and benefits. Moreover, PTSD and suicidal and nonsuicidal self-harm often remain
undocumented in electronic health records (EHR). There is also poor predictability of disease outcomes since
there are frequent changes in pharmacological treatment and multiple modifying co-occurring conditions
including depression, bipolar disorder, schizophrenia, substance use disorders, traumatic brain injury, and
sleep disorders. Our long-term goal is to improve diagnostics, secondary/tertiary prevention, and treatment
outcomes of PTSD and its co-occurring conditions via enhanced EHR utilization. To achieve our objectives, we
will analyze EHR and administrative claims data from Veterans Health Administration (VHA) and non-VHA
databases, collectively covering >1.8M patients with PTSD. Specifically, we aim to: (1) Identify undetected and
uncoded co-occurring mental health phenotypes that impact PTSD outcomes using machine learning and
characterize disparities in their documentation; (2) Create robust models, accounting for biases and
co-occurring conditions, to identify clinical trajectories of PTSD decompensation/recovery in response to
time-varying treatments; and (3) Compare risk of self-harm and hospitalization among PTSD treatments using
coded and imputed phenotypes through an international network study. We will compare the effectiveness of
PTSD psychotropic monotherapies, polypharmacy, and psychotherapy to guide the choice of treatment for
improved patient outcomes. By enhancing and validating a positive-unlabeled machine learning approach
developed by our team, we will impute unrecorded/undetected mental health conditions co-occurring with
PTSD in both VHA and non-VHA databases, and characterize factors associated with documentation
disparities. We will model disease trajectories with enhanced latent class / latent trajectory analysis, focusing
on self-harm, substance use disorders, and psychiatric hospitalization in PTSD. Finally, we will perform the
largest comparative effectiveness studies to date of PTSD treatments on >100 monotherapy and
polypharmacy regimens, in addition to psychotherapy interventions, using causal models and methods for
addressing biases. These studies will provide high-quality evidence on the risk of hospitalizations and suicidal
acts/self-harm. Successful completion of these investigations will improve the quality of clinical psychiatric
decision-making, and guide improved service delivery to the Veteran and non-Veteran populations with
PTSD/TBI, and/or high risk of self-harm/suicidality.
