Background/Purpose:   Among persons at high risk for knee osteoarthritis (OA), identifying those who will have function decline is important; instituting prevention strategies in all at high risk is impractical and costly.  Our focus is a cohort at high risk but without knee OA, at the same stage, before disease contaminates risk factors, and when prevention strategies are likely to be effective.  Our objectives were: 1) develop prediction models for poor function outcomes, and 2) test if adjusting for variables linked to behavior change, modified lifestyle to avoid damaging activities to knees, or lack of knee confidence, improves prediction.

Methods:   In 1197 OAI participants KL0 in both knees, baseline predictors analyzed included: demographic, socioeconomic, psychological, comorbidity; WOMAC Pain (P), WOMAC Function (F), KOOS Symptoms (Sx), KOOS QOL (higher worse); KOOS QOL items (higher better).  Proportional hazards (PH) regression models were used to develop prediction models for each outcome, unable to complete sit-to-stand without arms, and slow gait speed (< 1m/sec), over up to 10 years follow-up.  For each outcome, a base model included predictors with univariate hazard ratio p ≤ 0.20; separate models also included P, F, Sx, and QOL.  Comparisons of nested PH models used a likelihood ratio (LR) chi-square test and Schwarz Bayesian Information Criterion (SBC).  We calculated the AUC of the ROC for logistic regression models including variables that improved prediction in the corresponding PH model. Hosmer-Lemeshow Χ² statistics (H-L) tested goodness-of-fit; large H-L p-values and high AUC indicate good calibration and discrimination, respectively.

Results:   For gait speed outcome, the expanded prediction model vs. base model was significantly improved by including WOMAC and KOOS variables, particularly WOMAC-F, KOOS-QOL, and KOOS-QOL items (Table 1, LR tests). For sit-to-stand outcome, prediction was significantly improved by KOOS-QOL and KOOS-QOL items (Table 2)  For both outcomes, among the models, performance was best with base model + modified lifestyle to avoid potentially damaging activities to knees:  for slow gait speed, AUC 0.80 (95% CI 0.76, 0.84), H-L p = 0.55; and for sit-to-stand, AUC 0.78 (95% CI 0.74, 0.82), H-L p = 0.56.

Conclusion:   In persons at high risk for knee OA, improved prediction and best performance for poor function outcome was achieved by including modified lifestyle to avoid potentially damaging activities to knees.  Such a model, achieving excellent AUCs using easily assessed variables, may help early-stage identification of persons at high risk for poor outcome and inform focus of prevention strategies.