RIbench

RIbench benchmarks indirect methods for estimating reference intervals from mixed clinical laboratory measurements to evaluate method accuracy.

Key Features:

• Standardized Evaluation: Provides a benchmarking suite for quantitative comparison of existing and novel indirect methods for reference interval estimation.
• Simulated Test Sets: Includes simulated test sets for ten biomarkers that mimic routine measurements containing mixed non-pathological (reference) and pathological values.
• Distribution Types: Covers four distribution types—normal, skewed, heavily skewed, and skewed-and-shifted—to represent common measurement distributions.
• Diverse Scenario Parameters: Incorporates test sets with varying sample sizes and pathological distributions that differ in location, extent of overlap, and fraction.
• Performance Metrics: Uses an overall benchmark score and sub-scores derived from absolute z-score deviations between estimated and true reference limits.
• Method Comparison: Supports comparison of indirect methods against each other and against a nonparametric direct method.

Scientific Applications:

• Diagnostic Accuracy Assessment: Evaluates how indirect methods recover true reference limits under mixed-pathology scenarios to inform clinical interpretation of laboratory results.
• Method Selection and Validation: Enables selection and validation of indirect reference-interval estimation methods for specific biomarkers and clinical contexts.

Methodology:

Compares various indirect methods against each other and against a nonparametric direct method using simulated data; illustrative evaluations included the Hoffmann method and four modern indirect methods—TML (Truncated-Maximum-Likelihood), kosmic, TMC (Truncated-Minimum-Chi-Square), and refineR—and reported that with pathological fraction up to 20% and minimum sample size of 5000 most modern indirect methods achieved performance comparable or superior to the direct method.