Causal inference under interference with prognostic scores for dynamic group therapy studies.

Area of Science:

• Behavioral Health
• Psychotherapy Research
• Causal Inference

Background:

• Group therapy is a primary treatment for behavioral health issues.
• Dynamic therapy groups feature continuous patient entry/exit, complicating outcome analysis.
• Causal effect estimation is challenging due to non-randomization and patient interference.

Purpose of the Study:

• To define and estimate the causal effect of high versus low session attendance on patient outcomes in dynamic group therapy.
• To address challenges of interference and non-randomization in group therapy settings.
• To apply a novel causal inference strategy to real-world group therapy data.

Main Methods:

• Utilized the Rubin causal model framework for defining causal effects.
• Proposed prognostic score stratification to identify individual, peer, and total effects.
• Employed simulation studies to validate the proposed methodology.
• Applied the method to a group cognitive behavioral therapy (gCBT) trial for depression in substance use disorder patients.

Main Results:

• The proposed prognostic score stratification method effectively identifies causal effects of attendance.
• Simulation results demonstrated the validity and performance of the approach.
• Analysis of gCBT trial data provided insights into attendance-outcome relationships.

Conclusions:

• The developed causal inference strategy is effective for analyzing dynamic group therapy.
• High session attendance is causally linked to improved patient outcomes.
• This framework offers a robust approach for future research in group therapy settings.