Analytical validation of a multi-biomarker algorithmic test for prediction of progressive kidney function decline in patients with early-stage kidney disease

Abstract

Background: The KidneyIntelX™ test applies a machine learning algorithm that incorporates plasma biomarkers and clinical variables to produce a composite risk score to predict a progressive decline in kidney function in patients with type 2 diabetes (T2D) and early-stage chronic kidney disease (CKD). The following studies describe the analytical validation of the KidneyIntelX assay including impact of observed methodologic variability on the composite risk score. Methods: Analytical performance studies of sensitivity, precision, and linearity were performed on three biomarkers assayed in multiplexed format: kidney injury molecule-1 (KIM-1), soluble tumor necrosis factor receptor-1 (sTNFR-1) and soluble tumor necrosis factor receptor-2 (sTNFR-2) based on Clinical Laboratory Standards Institute (CLSI) guidelines. Analytical variability across twenty (20) experiments across multiple days, operators, and reagent lots was assessed to examine the impact on the reproducibility of the composite risk score. Analysis of cross-reactivity and interfering substances was also performed. Results: Assays for KIM-1, sTNFR-1 and sTNFR-2 demonstrated acceptable sensitivity. Mean within-laboratory imprecision coefficient of variation (CV) was established as less than 9% across all assays in a multi-lot study. The linear range of the assays was determined as 12–5807 pg/mL, 969–23,806 pg/mL and 4256–68,087 pg/mL for KIM-1, sTNFR-1 and sTNFR-2, respectively. The average risk score CV% was less than 5%, with 98% concordance observed for assignment of risk categories. Cross-reactivity between critical assay components in a multiplexed format did not exceed 1.1%. Conclusions: The set of analytical validation studies demonstrated robust analytical performance across all three biomarkers contributing to the KidneyIntelX risk score, meeting or exceeding specifications established during characterization studies. Notably, reproducibility of the composite risk score demonstrated that expected analytical laboratory variation did not impact the assigned risk category, and therefore, the clinical validity of the reported results.