2 - (SYM 77) How Well Can We Predict Dropout from Trauma-focused Therapy for PTSD? Differences Between Black and White Patients

Background: Dropout poses a substantial barrier to maximizing the impact of trauma-focused therapies for PTSD (Varker et al., 2021). Some evidence indicates that Black patients are more likely to drop from trauma-focused therapy than their White counterparts (Lester et al., 2010), indicating a need to better understand the factors associated with dropout across racial groups. To address this, the present study used a machine learning approach to examine whether dropout could be more accurately predicted from baseline clinical and demographic variables when modeling White and Black patients separately, and whether predictors were similar across racial groups.

Methods: The study sample consisted of 463 patients (n = 294 White, n = 169 Black) enrolled in one of four randomized controlled trials examining cognitive processing therapy and prolonged exposure for PTSD. To predict dropout status, a machine learning approach was used. Specifically, an elastic net algorithm was adopted, and 10-fold nested cross-validation was performed to conduct a thorough hyperparameter search without potential data leakage.

Results: An initial model with all patients showed acceptable classification accuracy in predicting dropout, with an Area Under the Curve (AUC) of .71. The strongest predictors of dropout were lower income, Black race, a history of physical assault in adulthood, and an index trauma of adult physical assault. When predicting dropout only among White patients (dropout rate = 28%), classification accuracy (AUC = .72) and the strongest predictors of dropout were nearly identical to the initial model. When predicting dropout only among Black patients (dropout rate = 51%), classification accuracy was poor (AUC = 0.58), with lower income being the strongest individual predictor. 

Conclusions: Results show that demographic and baseline clinical variables can be used to predict likelihood of dropout from trauma-focused therapy among White patients, but these variables do not enable accurate prediction of dropout for Black patients. Identifying measures that better capture the unique barriers to treatment completion among Black patients is needed.