A simulation study on estimating biomarker-treatment interaction effects in randomized trials with prognostic...

Area of Science:

• Biostatistics
• Clinical Trial Design
• Translational Medicine

Background:

• Individualizing treatment decisions requires identifying biomarker-treatment interactions.
• Randomized clinical trials (RCTs) are often used to analyze these interactions.
• Statistical power for interaction tests in RCTs can be limited.

Purpose of the Study:

• To investigate how considering additional prognostic variables impacts the power and bias of biomarker-treatment interaction estimates.
• To evaluate different analytical approaches for handling prognostic factors in time-to-event outcomes within RCTs.

Main Methods:

• A simulation study was conducted to assess the effect of prognostic factors on biomarker-treatment interaction estimates.
• Scenarios varied in censored observations, covariate correlations, and interaction strength.
• Approaches included ignoring, including all, or using variable selection for covariates.

Main Results:

• Including prognostic variables associated with the outcome increases the probability of detecting true biomarker-treatment interactions.
• Omitting relevant prognostic variables leads to biased interaction estimates.
• Inadequate variable selection or inclusion of too many predictors increases false-positive rates.

Conclusions:

• Conducting a thorough literature search for prognostic variables pre-analysis is recommended.
• Pre-specifying analyses enhances power for detecting interactions and avoids selective reporting.
• Adequate consideration of prognostic variables improves the reliability of biomarker-treatment interaction findings in RCTs.