Multivariable logistic and linear regression models for identification of clinically useful biomarkers for osteoarthritis

Abstract

AbstractOsteoarthritis (OA) is the most common chronic degenerative joint disease which causes substantial joint pain, deformity and loss of activities of daily living. Currently, there are over 500 million OA cases worldwide, and there is an urgent need to identify biomarkers for early detection, and monitoring disease progression in patients without obvious radiographic damage to the joint. We have used regression modelling to describe the association of 19 of the currently available biomarkers (predictors) with key radiographic and clinical features of OA (outcomes) in one of the largest and best characterised OA cohort (NIH Osteoarthritis Initiative). We demonstrate that of the 19 currently available biomarkers only 4 (serum Coll2-1 NO2, CS846, COMP and urinary CTXII) were consistently associated with established radiographic and/or clinical features of OA. These biomarkers are independent of one another and provide additional predictive power over, and above established predictors of OA such as age, gender, BMI and race. We also show that that urinary CTXII had the strongest and consistent associations with clinical symptoms of OA as well as radiographic evidence of joint damage. Accordingly, urinary CTXII may aid in early diagnosis of OA in symptomatic patients without radiographic evidence of OA.