Clin Chem Lab Med. 2026 Sep 3. doi: 10.1515/cclm-2026-1099. Online ahead of print.
ABSTRACT
OBJECTIVES: Using HbA1c for diabetes diagnosis directly links analytical uncertainty to patient classification. Since traditional performance metrics provide limited insight into these clinical implications, we developed a probabilistic framework to translate HbA1c concentrations and components of analytical uncertainty into clinically interpretable diagnostic risk.
METHODS: The framework was applied to 100,472 HbA1c results obtained from a real-world Laboratory Information System. Analytical bias, analytical imprecision, and biological variation were integrated to estimate the probability that an HbA1c result would be assigned to an alternative diagnostic category based on medical decision limits. Misclassification risk was evaluated across a broad range of analytical performance conditions, allowing the contribution of analytical performance to be distinguished from the background risk arising solely from biological variation. Benchmarking analyses were performed using five analytical platforms, based on External Quality Assessment (EQA) data.
RESULTS: Analytical bias increased misclassification risk more than imprecision. Class-specific analyses showed that analytical error redistributes risk asymmetrically rather than uniformly increasing it; thus, a platform’s clinical impact depends strongly on the diagnostic composition of the tested population. EQA benchmarking showed that platforms with apparently acceptable conventional performance produced markedly different EAR profiles. Furthermore, probabilistic reporting scenarios revealed substantial divergence from deterministic classification near diagnostic thresholds.
CONCLUSIONS: The proposed Clinical Misclassification Risk Framework (CMRF) quantifies the direct contribution of analytical uncertainty and demographics to patient misclassification. Applicable beyond HbA1c, this framework might support a critical transition in laboratory medicine from assessing metrological quality alone to actively managing clinical decision risk.
PMID:42685300 | DOI:10.1515/cclm-2026-1099