Development of a prediction model for future risk of radiographic hip osteoarthritis

Objective: To develop and validate a prognostic model for incident radiologic hip osteoarthritis (HOA) and determine the value of previously identified predictive factors. Design: We first validated previously reported predictive factors for HOA by performing univariate and multivariate analyses for all predictors in three large prospective cohorts (total sample size of 4548 with 653 incident cases). The prognostic model was developed in 2327 individuals followed for 10 years from the Rotterdam Study-I (RS-I) cohort. External validation of the model was tested on discrimination in two other cohorts: RS-II (n = 1435) and the Cohort Hip and Cohort Knee (CHECK) study (n = 786). Results: From the total number of 28 previously reported predictive factors, we were able to replicate 13 factors, while 15 factors were not significantly predictive in a meta-analysis of the three cohorts. The basic model including the demographic, questionnaire, and clinical examination variables (area under the receiver-operating characteristic curve (AUC) = 0.67) or genetic markers (AUC = 0.55) or urinary C-terminal cross-linked telopeptide of type II collagen (uCTX-II) levels (AUC = 0.67) alone were poor predictors of HOA in all cohorts. Imaging factors showed the highest predictive value for the development of HOA (AUC = 0.74). Addition of imaging variables to the basic model led to substantial improvement in the discriminative ability of the model (AUC = 0.78) compared with uCTX-II (AUC = 0.74) or genetic markers (AUC = 0.68). Applying external validation, similar results were observed in the RS-II and the CHECK cohort. Conclusions: The developed prediction model included demographic, a limited number of questionnaire, and imaging risk factors seems promising for prediction of HOA.