Semi-parametric Bayesian regression for subgroup analysis in clinical trials.

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Precision Medicine

Background:

• Accurate estimation of differential treatment effects in subgroups is crucial for precision medicine.
• Traditional regression models may yield overly conservative or anti-conservative results due to prior assumptions and potential model mis-specification.
• Over-shrinking of outlying subject data can bias subgroup mean response estimations.

Purpose of the Study:

• To investigate the use of nonparametric Bayes, specifically Dirichlet process priors, for semi-parametric models in clinical trial subgroup analysis.
• To address limitations of traditional regression models in estimating subgroup treatment effects.
• To develop models that accommodate response heterogeneity and unaccounted-for terms without excessive shrinkage.

Main Methods:

• Utilized Dirichlet process priors to create semi-parametric models.
• Modeled uncertainty in the prior distribution for overall response.
• Accommodated heterogeneity among individual subgroups and accounted for unaccounted terms.

Main Results:

• The proposed models avoid excessive shrinkage of estimates.
• Achieved improved precision with narrower credible intervals.
• Simulations demonstrated favorable bias, mean squared error, coverage probability, and credible interval widths.
• Applied the method to simulated data from a cystic fibrosis Phase 2 trial.

Conclusions:

• Nonparametric Bayes methods with Dirichlet process priors provide a robust approach for subgroup analysis in clinical trials.
• These models enhance the accuracy of treatment effect estimation in subgroups.
• The approach supports more reliable decision-making in personalized medicine by better handling subgroup heterogeneity.