Performance Assessment of Centrifugation-Based Hematocrit Testing in Primary Care Units: Evidence from Seventy-Two Proficiency Testing Datasets over Eight Years.

Abstract

Background: Centrifugation-based hematocrit (Hct) testing remains widely used in decentralized primary care settings where access to automated hematology analyzers may be limited. Appropriate proficiency testing (PT) performance assessment therefore requires consideration of the analytical variability associated with this measurement method. This study evaluated longitudinal PT data to characterize method-specific variability and compare alternative approaches to RSD estimation for centrifugation-based Hct testing. Methods: Seventy-two datasets from 32 PT rounds conducted during 2017-2024 were analyzed, with 12 to 310 participants per dataset. PT materials were assessed for homogeneity and stability in accordance with ISO 13528. The relative standard deviation (RSD) was estimated using two approaches: Method A, based on all statistically processed valid results to reflect total interlaboratory variability, and Method B, based on the satisfactory-performance subset (∣z∣ ≤ 2), within which robust statistics were recalculated. Temporal trends were evaluated by linear regression using dataset-specific RSD values. Sensitivity analysis evaluated Method B using alternative z-score inclusion thresholds of ∣z∣ ≤ 1.5, 2.0, 2.5, and 3.0. Performance classifications in 2024 were compared descriptively. Results: All PT materials met predefined homogeneity and stability criteria. The final cumulative pooled relative standard deviation (RSD_pooled) was 10.81% for Method A and 4.86% for Method B. was 10.81% for Method A and 4.86% for Method B. Method A showed an increasing temporal trend, whereas Method B showed minimal temporal change; however, participant numbers increased substantially over the study period and may have contributed to the Method A trend. Sensitivity analysis showed that Method B was threshold-dependent, with final RSD_pooled values of 3.90%, 4.86%, 5.59%, and 6.28% at ∣z∣ ≤ 1.5, 2.0, 2.5, and 3.0, respectively. In the 2024 datasets, satisfactory rates ranged from 75.68-99.12% for Method A and 43.15-95.61% for Method B. Conclusions: Methods A and B provide complementary information whereas Method B characterizes dispersion among participants meeting a predefined satisfactory-performance criterion. The Method B RSD_pooled of 4.86% is specific to inclusion criterion and should not be interpreted as a threshold-independent estimate of analytical imprecision. Further studies using prospectively defined statistical methods and appropriately structured participant-level data are needed to determine its potential utility for PT/EQA performance assessment.