Multi-source analyses of average treatment effects with failure time outcomes

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Multi-source data analyses (e.g., meta-analyses, pooled studies) estimate overall average treatment effects.
• Common methods may lack clear causal interpretation when treatment effects differ across data sources.
• Target population causal inference requires careful consideration of effect heterogeneity.

Purpose of the Study:

• To provide methods for identifying and estimating average treatment effects in a target population from multi-source data.
• To develop valid causal inference approaches when average treatment effects vary across data sources.
• To address point treatment settings for failure time outcomes with potential right-censoring.

Main Methods:

• Derivation of efficient influence functions for source-specific average treatment effects.
• Development of a novel doubly robust estimator for target population average treatment effects.
• Evaluation of the proposed estimator's finite-sample performance through simulation studies.

Main Results:

• The proposed methods allow for valid identification and estimation of average treatment effects under effect heterogeneity.
• The doubly robust estimator demonstrates good performance in simulation studies.
• The methods are applied to real-world data from the HALT-C multi-center trials.

Conclusions:

• The study offers a robust framework for causal inference from multi-source data with heterogeneous treatment effects.
• The proposed doubly robust estimator provides a reliable tool for estimating target population average treatment effects.
• These methods enhance the causal interpretability of findings from combined data sources in clinical research.