JMIR Publications

ABSTRACT

Background:

Growing evidence suggests that rest-activity rhythms may serve as relevant markers of posttraumatic stress disorder (PTSD). Despite the emergence of machine learning methods applied to actigraphy and self-report data, few studies have utilized these approaches to identify individuals with clinically diagnosed PTSD. Prior work has focused on predicting probable PTSD based on self-report measures, yet discrepancies exist between clinical diagnosis and probable PTSD derived from self-report.

Objective:

The current study examined whether wrist actigraphy and sleep logs could be used to accurately predict clinician-rated PTSD diagnosis and probable diagnosis of PTSD based on established self-report cutoffs (PTSD Checklist for DSM-5 [PCL-5] ≥31 and ≥38) among trauma-exposed service members and veterans. We also explored which features were most strongly predictive of each outcome and whether models were able to predict PTSD diagnosis even when accounting for other mental health disorders.

Methods:

Wrist actigraphy data and daily sleep logs were collected over one week from trauma-exposed male service members and veterans (N = 36; mAge = 41, SDage = 5.3). Univariate feature selection was performed to identify the top three features to be used to predict each outcome. Using extreme gradient boosting (XGBoost), classification models were trained to predict diagnosis of PTSD and probable diagnosis of PTSD based on common self-report cutoffs. Linear regression was used to assess the discriminant validity of model-predicted scores and each PTSD outcome specifically, relative to other mental health diagnoses.

Results:

Machine learning models predicting PTSD diagnosis and probable PTSD based on the PCL-5 ≥ 31 threshold both demonstrated strong performance. The diagnosis model achieved an AUC of 0.83, with high accuracy (88%) and specificity (96%), and moderate sensitivity (63%). The PCL-5 ≥ 31 model yielded comparable performance (AUC = 0.84), with balanced sensitivity (73%) and specificity (82%). For both models, a combination of subjective and objective markers were the most influential features. These models were able to predict PTSD even when accounting for non-PTSD mental health diagnosis, as model-specific scores were significantly associated with two outcomes, clinician-rated PTSD (B = 0.19, p < .01) and probable PTSD based on a PCL-5 ≥ 31 cut-off (B = 0.24, p < .01). In contrast, the model predicting probable PTSD based on the PCL-5 ≥ 38 threshold performed poorly (AUC = 0.47), with a non-significant relationship between predicted scores and the outcome (B < 0.01, p = .89).

Conclusions:

Subjective and objective rest-activity features may help identify individuals with PTSD, particularly when using clinician-administered diagnoses or more liberal self-report cut-offs (PCL-5 ≥ 31). Important features differed across outcomes, with only subjective restfulness being shared across each model. Findings support the potential of integrating wearable sensor data with subjective sleep information for PTSD assessment.