Comparison of LC–MS/MS and CLIA Methods for Vitamin D Measurement in an Unselected Outpatient Cohort: Statistical Evaluation and Impact on Clinical Classification

Abstract

Background: This study aimed to compare a chemiluminescence immunoassay (CLIA; Siemens Atellica IM 1600) with an automated liquid chromatography–tandem mass spectrometry (LC–MS/MS) analyser (Thermo Scientific Cascadion SM) for serum 25-hydroxyvitamin D [25(OH)D] measurement in unselected outpatients, and its impact on clinical classification. Methods: 25(OH)D was measured on 1064 paired serum samples (828 women, 236 men; age 3–111 years); both platforms participated in the Vitamin D Standardisation Program (VDSP). Agreement was assessed by the Wilcoxon signed-rank test, Passing–Bablok regression, Bland–Altman analysis on log-transformed values, Breusch–Pagan testing for heteroscedasticity, consistency-based intraclass correlation coefficient (ICC), Lin’s concordance correlation coefficient, and Cohen’s kappa with McNemar’s test on six- and two-category clinical classifications. Results: Methods differed significantly (p<2.2×10−16). Passing–Bablok regression showed proportional bias (slope 1.33; 95 % confidence interval [CI], 1.28–1.38; LC–MS/MS higher). Consistency ICC was 0.928 (95 % CI, 0.919–0.936). Log-Bland–Altman bias was 35.5 % (limits of agreement, −18.3 % to +124.5 %; heteroscedasticity, p=5.18×10−11). Cohen’s kappa was 0.32 unweighted and 0.72 weighted on six clinical categories, and 0.56 on a two-class scheme (McNemar p<2.2×10−16). In discordant pairs, CLIA classified deficiency 118-fold more often than LC–MS/MS. Conclusions: In our study, despite preserved rank ordering, a non-linear concentration-dependent bias and different limits of agreement caused systematic differences in the classification of vitamin D deficiency. Adoption of automated LC–MS/MS can substantially reclassify outpatients toward sufficiency, with implications for supplementation and laboratory harmonisation.