An integrated framework for universal coagulation test reference intervals: Cross-platform validation of eleven...

The conventional direct method for establishing coagulation reference intervals (RIs) faces challenges, including difficulties in recruiting healthy populations, non-Gaussian distributions, and complex age-related effects. Indirect methods utilizing real-world data (RWD) are limited by a lack of consensus on algorithm selection and insufficient cross-platform validation.

To develop a multi-algorithm collaborative framework for systematically evaluating the performance of 11 outlier detection algorithms, thereby optimizing coagulation RIs for the Chinese population and validating their cross-platform applicability.

We retrospectively analysed 630,946 coagulation test records from Shenzhen Hospital of Southern Medical University. Outlier removal was performed using eleven algorithms. Algorithm stability and sex/age effects were quantified via reference change value (RCV) and variance component models. Validation was conducted across Mindray CX-9000 and Stago STA R Max platforms.

Among the 11 algorithms, Z-Score and Tukey demonstrated optimal performance for 8 coagulation parameters, yielding RIs with minimal deviation from manufacturer standards. Sex-based analysis revealed minimal differences (Standard deviation ratio (SDR) < 0.40, Standardized effect size (Cohen's d) <0.50), eliminating the need for sex stratification. Age stratification was required for D-Dimer (≥60 years subgroup: SDR = 0.62, Cohen's d = 0.54). Cross-platform consistency confirmed algorithm generalizability.

This RCV-driven multi-algorithm framework establishes the Z-Score and Tukey methods as optimal for coagulation RIs derivation, advocates age-specific D-Dimer RIs, and deems sex stratification unnecessary. This scalable approach enhances standardization in coagulation testing.