Related Experiment Videos
Estimation of the average causal effect among subgroups defined by post-treatment variables.
Yutaka Matsuyama1, Satoshi Morita
1Department of Biostatistics, School of Health Sciences and Nursing, University of Tokyo, Tokyo, Japan. matuyama@epistat.m.u-tokyo.ac.jp
Clinical Trials (London, England)
|March 17, 2006
Summary
This study introduces a new causal inference method for clinical trials. The proposed estimator accurately determines treatment effects in patient subgroups, unlike standard methods that lack causal interpretation.
Area of Science:
- Biostatistics
- Clinical Trial Methodology
- Causal Inference
Background:
- Standard estimators in clinical trials lack causal interpretation when analyzing subgroups defined by post-randomization events.
- Comparing treatments within specific patient subgroups requires methods that respect the causal structure.
Purpose of the Study:
- To formulate clinically relevant causal estimands using the principal stratification framework.
- To propose a novel estimation method for principal causal effects in subgroup analyses.
- To estimate the local average treatment effect (LATE) among patients who would respond to either treatment.
Main Methods:
- Utilized the principal stratification framework by Frangakis and Rubin.
- Developed a weighted average estimator based on the probability of response under alternative treatment.
- Assumed independence of counterfactual response indicator conditional on covariates.
Main Results:
- Applied the method to a phase III trial for non-small-cell lung cancer.
- Estimated average duration of response difference among responders as 16.1 days (proposed) vs. 9.5 days (standard).
- Simulation studies confirmed the proposed estimator is unbiased, while the standard estimator is biased.
Conclusions:
- Developed an easily constructible estimation method for the local average treatment effect.
- The proposed estimator provides a valid causal interpretation for treatment effects among patients with a specific event in either treatment group.
- Applicable to various outcome variables in clinical trial subgroup analyses.