Handing missing data using multiple imputation in hybrid control clinical trials with modified power prior

Area of Science:

• Biostatistics
• Clinical Trial Design
• Real-World Data Analysis

Background:

• Randomized Controlled Trials (RCTs) are the benchmark for clinical evidence.
• Hybrid control trials incorporating Historical Data (HD) can enhance evidence generation.
• Real-world data (RWD) used for HD often contains missing covariate information, potentially introducing bias.

Purpose of the Study:

• To propose and evaluate a statistical method for addressing missing covariate data in HD for hybrid control trials.
• To assess the performance of multiple imputation combined with propensity score matching and power prior methods.
• To investigate bias and precision under different missing data mechanisms and data similarity scenarios.

Main Methods:

• Employed multiple imputation under the Missing At Random (MAR) assumption to handle missing covariates.
• Utilized propensity score matching and a modified power prior approach for data analysis.
• Conducted simulations to compare proposed methods against complete case analysis and applied the method to real clinical trial data.

Main Results:

• Complete case analysis exhibited bias with missing at random and covariate-dependent missingness.
• Multiple imputation yielded nearly unbiased estimates and improved precision when HD was similar to current trial data (MAR assumption).
• The method dynamically borrowed information from HD based on outcome similarity, enhancing accuracy and power while controlling Type I error.

Conclusions:

• The proposed method effectively handles missing covariate data in HD for hybrid control trials.
• Multiple imputation offers a robust solution, outperforming complete case analysis.
• The approach demonstrates practical utility and potential for improving hybrid trial designs using RWD.