Implementing multiple imputations for addressing missing data in multireader multicase design studies

Area of Science:

• Medical Imaging Analysis
• Biostatistics
• Machine Learning in Healthcare

Background:

• Missing data is a common challenge in multireader multicase (MRMC) studies, often arising from reader errors or technical issues.
• Improper handling of missing data in MRMC designs can introduce significant bias into study results.
• Existing research on addressing missing data within the MRMC framework is limited.

Purpose of the Study:

• To introduce and evaluate a novel approach for handling missing data in MRMC studies.
• To compare the performance of the proposed multiple imputation MRMC (MI-MRMC) method against traditional complete case analysis.
• To validate the practicality of the MI-MRMC approach using a real-world computer-aided diagnosis (CAD) study.

Main Methods:

• Developed a novel MI-MRMC approach integrating multiple imputation with MRMC analysis.
• Conducted an extensive simulation study to compare MI-MRMC with complete case analysis.
• Applied both methods to a CAD study on aneurysm detection using head and neck CT angiograms.

Main Results:

• The MI-MRMC approach yielded nearly unbiased estimates of diagnostic capability in simulations, even with small sample sizes.
• MI-MRMC demonstrated satisfactory statistical power and type I error rates compared to complete case analysis.
• In a real CAD study, MI-MRMC showed superior performance in point estimates and confidence intervals over complete case analysis.

Conclusions:

• The MI-MRMC approach effectively mitigates bias caused by missing data in MRMC settings.
• Adopting MI-MRMC facilitates the achievement of unbiased and robust diagnostic capability estimates.
• This method enhances the reliability of findings in CAD studies utilizing MRMC designs.