Background/Purpose: High variability in patient-reported outcome scores presents a major challenge in the design and interpretation of clinical studies in osteoarthritis (OA). We present a novel successive prediction algorithm that forecasts individual WOMAC scores based on known clinical parameters and identifies risk factors for deviations between observed and predicted scores.

Methods:  Participants from the Osteoarthritis Initiative progression cohort with a baseline Kellgren-Lawrence grade ≥2 for either knee were selected for analysis (N = 1,122). Demographic data, history of pain and depression, BMI, and physical activity parameters at baseline and eight subsequent annual observation times were processed to maximize the number of complete records. WOMAC scores were predicted for study years four through eight, using a mixed effects regression model applied to data up to, but not including, each study year. The predictive model adjusted for continuous time trends for individual subjects as well as the overall cohort, yielding an expected WOMAC score for each participant. To identify subgroups with excessive residual WOMAC scores, subject-specific residuals were analyzed by a separate multivariable regression at each time point.

Results:  We were able to accurately model WOMAC scores and to use the fitted models to predict individual scores for each successive year. The predictive model for each visit showed significant overall fit (p < 0.0001 at all targeted study years). The succession of fitted models adapted to new information gained over time, and adjusted predictions to minimize residuals (Figure 1). Factors associated with lower than expected WOMAC scores on more than one successive visit were long-standing history of knee pain and history of hip pain, with mean residual WOMAC (95% confidence interval) ranging from –3.3 (–5.5, –1.7) to –7.7 (–13.2, –2.3). The factor associated with higher than expected WOMAC scores was high performance on the 20-meter walk test, with mean residual WOMAC ranging from +3.3 (+0.5, +6.0) to +4.5 (+1.7, +7.2).

Conclusion:  A successive prediction model closely predicted WOMAC scores based on traditional OA risk factors and prior clinical assessments. Subjects with long-standing histories of knee pain and associated hip pain, as well as subjects with good physical performance displayed unexpected WOMAC scores over time.