Prediction model for knee osteoarthritis incidence, including clinical, genetic and biochemical risk factors

Objective

To develop and validate a prognostic model for incident knee osteoarthritis (KOA) in a general population and determine the value of different risk factor groups to prediction.

Methods

The prognostic model was developed in 2628 individuals from the Rotterdam Study-I (RS-I). Univariate and multivariate analyses were performed for questionnaire/easily obtainable variables, imaging variables, genetic and biochemical markers. The extended multivariate model was tested on discrimination (receiver operating characteristic curve and area under the curve (AUC)) in two other population-based cohorts: Rotterdam Study-II and Chingford Study.

Results

In RS-I, there was moderate predictive value for incident KOA based on the genetic score alone in subjects aged <65 years (AUC 0.65), while it was only 0.55 for subjects aged ≥65 years. The AUC for gender, age and body mass index (BMI) in prediction for KOA was 0.66. Addition of the questionnaire variables, genetic score or biochemical marker urinary C-terminal cross-linked telopeptide of type II collagen to the model did not change the AUC. However, when adding the knee baseline KL score to the model the AUC increased to 0.79. Applying external validation, similar results were observed in the Rotterdam Study-II and the Chingford Study.

Conclusions

Easy obtainable 'Questionnaire' variables, genetic markers, OA at other joint sites and biochemical markers add only modestly to the prediction of KOA incidence using age, gender and BMI in an elderly population. Doubtful minor radiographic degenerative features in the knee, however, are a very strong predictor of future KOA. This is an important finding, as many radiologists do not report minor degenerative changes in the knee.