Reference-based multiple imputation for missing data sensitivity analyses in trial-based cost-effectiveness analysis

Area of Science:

• Health Economics
• Biostatistics
• Clinical Trials

Background:

• Missing data are prevalent in cost-effectiveness analysis (CEA) alongside randomized trials.
• Standard methods often assume data are 'missing at random', an assumption frequently questionable.
• Sensitivity analyses are crucial to evaluate the impact of deviations from the missing at random assumption.

Purpose of the Study:

• To extend and illustrate the reference-based multiple imputation (MI) approach within the context of CEA.
• To provide a framework for assessing the robustness of CEA conclusions to various missing data assumptions.
• To introduce principles of reference-based imputation and propose its extension for CEA.

Main Methods:

• Reference-based multiple imputation (MI) is proposed as an attractive method for sensitivity analyses in CEA.
• This approach frames missing data assumptions intuitively by referencing other trial arms.
• The method is illustrated using the CEA of the CoBalT trial for treatment-resistant depression, with Stata code provided.

Main Results:

• Reference-based MI provides a relevant and accessible framework for sensitivity analyses in CEA.
• The approach allows for plausible 'not at random' missing data mechanisms to be assessed.
• Demonstrates the utility of reference-based MI in evaluating the impact of missing data on economic evaluations.

Conclusions:

• Reference-based multiple imputation is a valuable tool for enhancing the rigor of cost-effectiveness analyses.
• It facilitates a more transparent and robust assessment of missing data assumptions in health economic evaluations.
• The proposed method improves the reliability of CEA findings, particularly when dealing with complex missing data scenarios.