Inference on covariate effect types for treatment effectiveness in a randomized trial with a binary outcome

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Translational Oncology

Background:

• Randomized clinical trials (RCTs) often evaluate covariate effects beyond simple prediction or prognosis.
• Existing methods may not fully capture the nuanced impact of covariates on treatment response.
• A need exists for a more precise framework to characterize how covariates influence outcomes in RCTs.

Purpose of the Study:

• To introduce and define novel covariate effect types: activated-, inert-, causative-, and preventive-responders.
• To propose a Bayesian approach for assessing these covariate effect types in binary outcome RCTs.
• To differentiate this new classification from traditional predictive and prognostic assessments.

Main Methods:

• Proposed four potential response types for binary outcomes: activated-, inert-, causative-, and preventive-responders.
• Utilized a Bayesian method to derive posterior distributions of response proportions within covariate subgroups.
• Assessed covariate effect types by examining differences in response proportions between subgroups (e.g., "augmented-causative", "neutral-activated").

Main Results:

• Applied the method to an RCT of nivolumab vs. docetaxel in non-small-cell lung cancer.
• PD-L1 status was identified as "augmented-causative" for nivolumab, with a 24.3% effect on causative-responders.
• PD-L1 status was found to be "neutral-activated" for activated-responders.

Conclusions:

• The proposed approach offers a detailed characterization of covariate effect types and their magnitude.
• This framework provides a valuable supplement to standard subgroup and regression analyses in binary outcome RCTs.
• The findings demonstrate the utility of this novel classification in understanding treatment 효과 and biomarker roles.