Meta-analysis of test accuracy studies using imputation for partial reporting of multiple thresholds

Area of Science:

• Biostatistics
• Medical Informatics
• Diagnostic Test Accuracy Studies

Background:

• Primary studies often report test performance at various thresholds, leading to missing data in meta-analyses.
• Standard meta-analysis (no imputation) ignores this missing data.
• Single imputation (SI) was previously proposed to address missing threshold results.

Purpose of the Study:

• To introduce a novel method, multiple imputation of missing threshold results using discrete combinations (MIDC).
• To compare MIDC with no imputation (NI) and single imputation (SI) in meta-analysis of test accuracy.

Main Methods:

• MIDC imputes missing thresholds by selecting from discrete combinations between known thresholds.
• Imputed and observed results are synthesized, repeated multiple times, and combined using Rubin's rules.
• Simulations were conducted to compare NI, SI, and MIDC approaches.

Main Results:

• Both imputation methods (SI and MIDC) outperformed the no imputation (NI) method in simulations.
• MIDC showed advantages in estimating between-study variances and providing better coverage.
• Imputation methods reduced bias in summary receiver operating characteristic curves, especially with selective threshold reporting.

Conclusions:

• Single imputation (SI) and multiple imputation using discrete combinations (MIDC) are valuable for analyzing missing threshold data in test accuracy meta-analyses.
• MIDC is particularly effective for addressing missing threshold data in diagnostic test accuracy studies.