Adipokines production in metabolic-associated osteoarthritis of the knee joint

Abstract

Purpose: Age, sex, and BMI can help to predict knee osteoarthritis (OA) patients from healthy individuals (HV). The metabolome, the comprehensive output of metabolic processes occurring within an individual, and the levels of individual metabolites can also be used to help with disease diagnosis. However, metabolite selection methods and modeling algorithms that best identify metabolites capable of predicting OA have not been well established. We sought to determine a method that was capable of effectively identifying metabolite signatures that were predictive of OA in demographically-stratified populations. Methods: Phosphatidylcholine (lysoPC) and lyso(PC) analogues from plasma of 152 OA patients undergoing total knee replacement and 194 HV (346 total individuals) were measured by metabolomics. Cohorts were stratified by age, sex and BMI. Analogue signatures were determined by generating univariate area under the receiver operator curve (UAUC) values from 1000 bootstrapped training and test sets. Metabolites with UAUC > 0.5 at the 2.5% quantile of the empirical distribution were selected as capable of predicting OA from HV within strata. Three multivariate classification algorithms were tested using each signature. The most consistent algorithm was determined by the minimum difference between training and test set AUC values, derived from 1000 resamplings. The effect of diabetes mellitus on signature elements was also determined by identifying metabolites that were significantly changed in diabetic patients and removing those signature elements from multivariate analyses.