Prediction of Treatment Benefit With Internet-Based Cognitive Behavioral Therapy for Depression

Abstract

Importance: Treatment benefit prediction models may help identify which patients will benefit most from internet-based cognitive behavioral therapy (iCBT) for depression; it is unclear whether such models can support useful treatment allocation across programs. Objective: To develop a model predicting individual treatment benefit from iCBT for depression across programs and to externally validate it in independent samples. Design, Setting, and Participants: This prognostic study used individual participant data from 5 randomized clinical trials of 3 iCBT programs conducted between 2012 and 2023, with internal validation via repeated nested cross-validation and external validation across 3 independent trials. Participants were adults with depressive symptoms or depressive disorders recruited from community and outpatient clinical settings in Germany and Austria. Analyses were finalized in June 2026. Exposure: Three iCBT programs (guided or unguided) were compared with treatment as usual with a waitlist or active sham control over 8 to 12 weeks. Main Outcomes and Measures: The outcome of interest was change in depression severity (assessed using Patient Health Questionnaire-9-item). Elastic net regression, ordinary least squares, and causal forest models were evaluated for agreement between predicted and observed treatment benefit and differences between predicted high- and low-benefit groups. Results: Of 2037 participants enrolled, 1589 (78.0%; mean [SD] age, 41.5 [11.7] years; 1189 [74.8%] female) with complete outcome data were included. Elastic net regression best predicted outcomes and benefit (R2 range, 0.19 to 0.50) using 12 predictors. Baseline depression severity was the strongest effect modifier, with each unit increase associated with 0.16 (95% CI, 0.06 to 0.25) points of improvement under treatment. The differences between predicted highest- and lowest-benefit groups excluded zero in only 1 sample (4.37 [95% CI, 0.10 to 8.64] points) and selectively treating patients by model recommendation performed worse than a simple treat-all strategy (range, -2.01 to -0.25 PHQ-9 points). Conclusions and Relevance: In this prognostic study of treatment benefit from iCBT for depression, more severe depression was associated with greater benefit from iCBT. The model ranked patients by likely benefit but not precisely enough to support model-based targeting. These findings support wider use of iCBT in patients with more severe depression and highlight the need to assess all dimensions of a model's performance.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 1 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.