RIbench: A Proposed Benchmark for the Standardized Evaluation of Indirect Methods for Reference Interval Estimation

Area of Science:

• Clinical laboratory science
• Biostatistics
• Medical informatics

Background:

• Indirect methods estimate reference intervals using real-world data, an active research area with recent method development.
• A lack of standardized tools hinders the evaluation and comparison of these indirect methods.

Purpose of the Study:

• To introduce RIbench, a novel benchmarking suite for the quantitative evaluation of indirect methods for reference interval estimation.
• To provide a standardized platform for comparing existing and new indirect methods.

Main Methods:

• RIbench utilizes simulated test sets for 10 biomarkers with varying non-pathological distributions (normal, skewed, heavily skewed, skewed-and-shifted).
• Test sets incorporate diverse sample sizes and pathological distribution characteristics (location, overlap, fraction) to assess method robustness.
• Performance is evaluated using an overall benchmark score and sub-scores based on z-score deviations from true reference limits.

Main Results:

• Modern indirect methods showed strong dependence on pathological fraction and sample size.
• With a pathological fraction up to 20% and a minimum sample size of 5000, most indirect methods achieved results comparable or superior to direct methods.
• RIbench was used to compare Hoffmann, TML, kosmic, TMC, and refineR methods against a nonparametric direct method.

Conclusions:

• RIbench is an open-source R-package designed for systematic evaluation of indirect reference interval estimation methods.
• This tool facilitates the enhancement of indirect methods and improves the accuracy of reference interval estimation.