Two biomarkers improve the diagnosis of pre-radiographic knee osteoarthritis: data from the osteoarthritis initiative

Purpose: To validate and qualify the putative ability of 6 proteins with biomarker potential to predict the appearance of radiographic knee OA. Methods: In the validation phase (Figure 1), 540 sera from the baseline visit belonging to participants from the Osteoarthritis Initiative (OAI) cohort who will develop (incident group, n=209) or not (not-incident group, n=331) radiographic knee OA in a follow-up period of 96 months were randomly selected to blindly quantify 6 potential biomarkers using custom sandwich microarrays by the xMAP technology. Non-parametric Mann-Whitney U tests were carried out to look for statistical differences between the outcome groups. In the qualification phase, the association of the individual biomarkers with the risk of knee OA development was assessed by univariate regression analysis. A clinical prognostic model was defined by stepwise regression analysis using clinical non-radiographic variables significantly associated with the OA incidence. The utility of the potential biomarkers, alone or in combination with other biomarkers included in the same multiplex sandwich immunoassay, was evaluated by comparing the Area Under the Curve (AUC) of the clinical prognostic model with the biomarkers plus clinical prognostic models. In addition, sensitivity, specificity and predictive coefficients (positive and negative) were also assessed. Results: For all the potential biomarkers analyzed, the incident group showed significant higher serum concentrations at the baseline visit (p < 0.05). We also found that 5 of the analytes were significantly associated with the future appearance of radiographic knee OA, yielding Odds Ratios (OR) ≥ 10 per 10 μg/ml increase. Among all the possible combinations, the inclusion of 2 of the potential biomarkers to the clinical prognostic model showed a significant improvement of the predictive capacity (AUCs = 0.78 vs 0.82, p= 0,044) with 65% (95% Confident Interval (95%CI): 60-70%) specificity and 88% (95%CI: 81-91%) sensitivity. Variables included in the regression model and all metrics comparing the biomarkers plus clinical prognostic model with the clinical prognostic model are shown in Figure 2A. The ROC curves of the biomarkers-only model, clinical prognostic model and biomarkers plus clinical prognostic model are represented in Figure 2B. Conclusions: Five of the potential biomarkers that have been analyzed were found associated with the development of knee OA before 96 months. Among all of them, the combination of 2 of the analytes showed a putative utility in the clinical setting by improving the predictive capacity of a clinical prognostic model to identify patients at a higher risk to develop radiographic knee OA.View Large Image Figure ViewerDownload Hi-res image Download (PPT)