Beyond correlation: Intraclass correlation and bland-altman analysis reveal poor absolute agreement across multiplex cytokine immunoassay platforms

Abstract

Cross-platform comparisons of serum cytokine immunoassays have often focused on rank correlation, but the extent of absolute agreement and systematic bias between platforms remains insufficiently characterized. Few independent head-to-head evaluations have compared multiple immunoassay platforms on the same clinical samples using overlapping cytokines. Here, serum samples from 48 adult participants were analyzed in parallel for seven cytokines (IL-6, IL-7, IL-8, IL-10, IL-15, TNF-α, IFN-γ), four chemokines (CCL2, CCL3, CCL4, CXCL10), and one growth factor (VEGF-A) using three multiplex platforms: OLINK proximity extension assay-based, LUMINEX bead-based, and V-PLEX electrochemiluminescence-based immunoassays. LLOQs, absolute concentrations, rank correlations, ICCs, and Bland-Altman biases were compared between platforms. Absolute serum cytokine concentrations differed by up to two orders of magnitude. Correlations between OLINK and V-PLEX were generally strong, whereas correlations involving LUMINEX were often weak or non-significant. ICCs were poor (<0.5) for most analytes and platform pairs, with IL-8 as a notable exception. Bland-Altman analyses on log-transformed concentrations revealed large, analyte-specific multiplicative biases and wide ratio-scale limits of agreement, indicating poor absolute agreement. These findings show that OLINK, LUMINEX, and V-PLEX measurements are not interchangeable at the level of absolute concentrations and underscore key analytical limitations that currently hinder the broad adoption of cytokine immunoassays, particularly multiplex panels, as regulator-approved tests.