Sample size calculation for a historically controlled clinical trial with adjustment for covariates

Area of Science:

• Biostatistics
• Clinical Trial Design
• Medical Device Research

Background:

• Historically controlled trials (HCTs) often use retrospective control groups.
• Non-randomized controls necessitate advanced statistical methods (e.g., hierarchical regression, propensity scores) to mitigate bias from heterogeneity.
• Existing sample size calculations for HCTs, often adapted from randomized trials, neglect key factors like parameter estimation, observation correlation, and covariate uncertainty.

Purpose of the Study:

• To develop a Bayesian approach for optimal sample size determination in historically controlled clinical trials.
• To address deficiencies in current sample size calculation methods for HCTs, particularly for medical device trials.
• To provide a robust methodology that accounts for complexities inherent in retrospective control groups.

Main Methods:

• A Bayesian framework was employed for sample size optimization.
• Stochastic optimization techniques were utilized to overcome limitations of traditional methods.
• The methodology was demonstrated using a power-based objective function.
• Analytic approximations for covariate-free analyses were developed to understand power function characteristics.

Main Results:

• Proposed Bayesian methodology provides more accurate sample size calculations compared to existing approximations.
• Stochastic optimization offers a computationally efficient approach for complex sample size determination.
• The developed methods account for parameter estimation, observation correlation, and covariate distribution uncertainty.
• Exact sample size calculations can differ significantly from current approximations, impacting trial power and resource allocation.

Conclusions:

• The novel Bayesian approach with stochastic optimization enhances the accuracy of sample size determination for historically controlled clinical trials.
• This methodology is particularly relevant for medical device trials utilizing retrospective controls.
• Accurate sample size calculations are essential for ensuring adequate statistical power and reliable trial outcomes.
• Stochastic optimization presents a practical computational solution for complex sample size problems in clinical research.