JMIR Publications

ABSTRACT

Background:

Online depression screening tools may increase uptake of evidence-based care and consequent symptom reduction. However, results of the DISCOVER [1] trial suggested no effect of automated results feedback compared with no feedback after online depression screening with respect to depressive symptom reduction six months after screening. Interpersonal variation in symptom representation, healthcare needs, and treatment preferences may nonetheless have led to differential response to feedback mode on an individual level.

Objective:

The aim of this study was to examine heterogeneity of treatment effects (HTE), i.e., differential response to two feedback modes (tailored or non-tailored) vs no feedback (control) following online depression screening.

Methods:

We utilized causal random forests, a machine-learning method that applies recursive partitioning to estimate conditional average treatment effects (CATEs). In this secondary data analysis of the DISCOVER trial, eligible participants had not been diagnosed or treated for depression in the preceding year. The primary outcome was depression severity change, i.e., the difference from initial screening to six months after, measured with the Patient Health Questionnaire-9. Analysis comprised exploration of average treatment effects (ATE), HTE, and difference in ATE when allocating feedback based on predicted CATE. We extracted top predictors of depression severity change given feedback and high-CATE covariate profiles. Prior to analysis, data was split into training and test sets (1:1) to minimize risk of overfitting and evaluate predictions in held-out test data.

Results:

Data from a subset of the DISCOVER trial without missing data (n = 946) were analyzed. We did not detect HTE in the DISCOVER trial (control vs. non-tailored feedback, P=.41; HTE control vs. tailored feedback, P=.93; HTE control vs. any feedback, P=.72). There was no evidence of alteration to the average treatment effect in the test set when providing feedback (tailored or non-tailored) based on the predicted CATE. Examining covariate profiles, we observed a potentially detrimental role of treatment control beliefs when providing tailored/non-tailored feedback compared with control.

Conclusions:

We applied causal random forests to describe higher level interaction of a broad range of predictors to detect HTE. In absence of evidence for HTE, treatment prioritization based on trained models did not improve ATEs. We did not find evidence for harm or benefit of providing tailored, or non-tailored feedback after online depression screening regarding depression severity change after six months. Future studies may test if screening alone prompts behavioral activation and downstream depression severity reduction, considering observed uniform changes across groups. Clinical Trial: ClinicalTrials.gov NCT04633096