Borrowing from Historical Control Data in Cancer Drug Development: A Cautionary Tale and Practical Guidelines

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacometrics

Background:

• Clinical trialists, particularly in rare or pediatric diseases, propose using historical control data from completed trials.
• Sole reliance on historical controls can lead to erroneous conclusions, as demonstrated by a case study showing a false significant treatment effect.

Purpose of the Study:

• To evaluate the conditions under which borrowing historical information in clinical trials is advisable.
• To assess the performance of Bayesian methods in analyzing historical and concurrent control data.

Main Methods:

• Utilized a Markov Chain Monte Carlo (MCMC)-driven Bayesian hierarchical parametric survival modeling approach.
• Analyzed data from a colorectal cancer study and simulated data to compare borrowing strategies.
• Evaluated effective historical sample size, bias, credible interval widths, and coverage probabilities.

Main Results:

• The Bayesian approach consistently identified significant differences between historical and concurrent controls, even after adjusting for covariates and standard-of-care improvements.
• Simulation studies explored bias and coverage probabilities across various borrowing scenarios.
• Bayesian methods effectively moderated data borrowing, especially when historical and concurrent controls were similar.

Conclusions:

• Bayesian hierarchical modeling provides a robust framework for incorporating historical data in clinical trials.
• Careful consideration of priors and covariate adjustment is crucial for appropriate data borrowing.
• These findings inform the design of future clinical trials, particularly those with limited concurrent control data.