Paired Human-Serum Evaluation of HER2 Immunosensors: Effects of Electrode Architecture and Trastuzumab/Pertuzumab Biorecognition on Agreement with ELISA

Abstract

A biosensor intended for serum measurement must be assessed against a comparator on the same specimens, because calibration behavior in buffer cannot establish agreement in a complex clinical matrix. This study compared nine label-free electrochemical immunosensors for serum human epidermal growth factor receptor 2 extracellular domain (HER2-ECD) in a paired subgroup of 25 women: eight healthy controls and 17 patients with breast cancer. Three working-electrode architectures—activated glassy carbon (GCE), poly(p-anisidine)-modified GCE (GCE/PPA), and ammonium alginate–tungsten trioxide-coated GCE/PPA [GCE/PPA/(AmAlg–WO3)]—were each combined with trastuzumab (AbT), pertuzumab (AbP), or an equal-total-mass trastuzumab/pertuzumab mixture (AbTP). For every coded specimen, 10 μL of undiluted serum was incubated on the sensor disk for 30 min. The current change was calculated as ΔIserum = I0 − Iserum and converted with the inverse calibration equation specific to that sensor. Matched HER2 concentrations were obtained by sandwich ELISA. Agreement was evaluated with Lin’s concordance correlation coefficient (CCC), Bland–Altman bias and 95% limits of agreement, mean absolute error (MAE), root mean square error (RMSE), Deming regression, and stratified analyses. Absolute errors were also compared by Friedman and Holm-corrected paired tests. GCE/PPA/(AmAlg–WO3)/AbTP showed the strongest descriptive balance: CCC 0.926 (95% CI, 0.678–0.970), bias −6.16 ng mL−1, MAE 9.12 ng mL−1, and RMSE 12.14 ng mL−1. GCE/PPA/(AmAlg–WO3)/AbP was close (CCC 0.917; MAE 10.03 ng mL−1). Bare-GCE configurations tended toward positive bias, whereas PPA-containing configurations tended toward negative bias. Most Friedman comparison families were not significant; the only Holm-corrected pairwise difference involved AbP versus AbTP within GCE/PPA. The nanocomposite mixed-antibody configuration is therefore a promising candidate, but the small paired sample, broad confidence intervals, and nontrivial limits of agreement do not support interchangeability with ELISA or general statistical superiority.