Transportability Methods for Time-to-Event Outcomes: Application in Adjuvant Colon Cancer Trials

Area of Science:

• Oncology
• Biostatistics
• Causal Inference

Background:

• Observed variations in 5-year overall survival benefits across 12 phase III adjuvant colon cancer trials (ACCENT group).
• Need to understand the drivers of treatment effect heterogeneity in clinical trials.

Approach:

• Applied state-of-the-art transportability methods based on causal inference.
• Compared transportability methods with conventional meta-analysis using prediction errors.
• Evaluated identifiability conditions and method performance across trials.

Key Points:

• One-trial-at-a-time transportability analysis showed prediction errors consistent with treatment effect discrepancies.
• Leave-one-trial-out transportability and meta-analysis yielded similar, significantly lower prediction errors (>40% reduction) compared to one-trial-at-a-time.
• Heterogeneity quantification was reduced in leave-one-trial-out transportability and meta-analysis versus one-trial-at-a-time.

Conclusions:

• Discrepancies in treatment effects are likely due to unobserved covariates or study-level factors, not model specification or observed covariates.
• Leave-one-trial-out transportability and conventional meta-analysis are more robust for predicting target population treatment effects.
• These methods offer a more reliable approach to understanding treatment efficacy across diverse clinical trial settings.