Including multiple imputation in a sensitivity analysis for clinical trials with treatment failures

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Medical Research

Background:

• Treatment failures in clinical trials necessitate changes in therapy, complicating direct comparisons between original treatment arms.
• Standard analytical methods like intent-to-treat can introduce significant bias when handling post-failure treatment modifications.
• Existing approaches for managing treatment failures include subject removal, data deletion post-failure, and last observation imputation, each with limitations.

Purpose of the Study:

• To evaluate the utility of multiple imputation for handling observations after treatment failure in clinical trials.
• To compare multiple imputation against traditional methods for managing treatment failures as a sensitivity analysis.
• To demonstrate these methods using a real-world dataset from the Asthma Clinical Research Network.

Main Methods:

• Multiple imputation was employed to replace observations occurring after treatment failure.
• Sensitivity analyses were conducted comparing multiple imputation to: complete case analysis, post-failure data removal, and last observation carried forward (LOCF).
• An intent-to-treat analysis based on original randomized assignments was also performed for comparison.

Main Results:

• Multiple imputation provides a viable approach to mitigate bias introduced by treatment changes following failure.
• Comparisons revealed that multiple imputation can yield different results than methods involving data deletion or LOCF.
• The application to the Asthma Clinical Research Network data illustrated the practical implementation and potential impact of the methods.

Conclusions:

• Multiple imputation is a valuable statistical technique for handling treatment failures in clinical trials, preserving data and reducing bias.
• This method offers a more robust sensitivity analysis compared to traditional approaches when treatment regimens are altered post-failure.
• The findings support the use of multiple imputation for more accurate and reliable treatment effect estimations in complex clinical trial scenarios.