Verification of the harmonization of human epididymis protein 4 assays

Serum human epididymis protein 4 (HE4) has gained relevance as an ovarian cancer (OC) biomarker and new automated methods have replaced the first released manual EIA by tracing results to it. We verified agreement and bias of automated methods vs. EIA as well as possible effects on patients' management.

One hundred and fifteen serum samples were measured by Abbott Architect i2000, Fujirebio Lumipulse G1200, Roche Modular E170, and Fujirebio EIA. Passing-Bablok regression was used to compare automated assays to EIA and agreement between methods was estimated by Lin's concordance correlation coefficient (CCC). The bias vs. EIA was estimated and compared to specifications derived from HE4 biological variation.

Median (25th-75th percentiles) HE4 concentrations (pmol/L) were 84.5 (60.1-148.8) for EIA, 82.7 (50.3-153.9) for Abbott, 89.1 (55.2-154.9) for Roche, and 112.2 (67.8-194.2) for Fujirebio. Estimated regressions and agreements (95% confidence interval) were: Abbott=1.01(0.98-1.03) EIA-4.8(-7.5/-2.6), CCC=0.99(0.99-1.00); Roche=0.91(0.89-0.93) EIA+5.7(4.2/8.0), CCC=0.98(0.98-0.99); Fujirebio=1.20(1.17-1.24) EIA+ 2.4(-0.6/4.9), CCC=0.97(0.96-0.98). The average bias vs. EIA resulted within the desirable goal for Abbott [-3.3% (-6.1/-0.5)] and Roche [-0.2% (-3.0/2.5)]. However, while for Abbott the bias was constant and acceptable along the measurement concentration range, Roche bias increased up to -28% for HE4 values >250 pmol/L. Lumipulse showed a markedly positive bias [25.3% (21.8/28.8)].

Abbott and Roche assays exhibited a good comparability in the range of HE4 values around the previously recommended 140 pmol/L cut-off. For patient monitoring, however, the assay used for determining serial HE4 must not be changed as results from different systems in lower and higher concentration ranges can markedly differ.