Abstract
OBJECTIVE: To use patient-reported outcomes (PROs) and performance measures to develop a multi-component outcome (virtual knee replacement [vKR]) that predicts knee replacement (KR). METHODS: Osteoarthritis Initiative (OAI) participants with baseline radiographs, clinical assessments and health insurance were followed for 60 months. Of 8205 knees (4143 participants), 206 knees (187 participants) had KR. Eighteen clinical measures available at annual visits were considered, (e.g., Knee injury and Osteoarthritis Outcome Score knee pain (KOOS KP), frequency, severity, and Quality of Life (QoL), frequent knee pain, knee pain severity, SF12 Mental Health and Physical Health subscales, physical activity, depression, 20-meter walk and chair stands). Combinations of these were evaluated utilizing logistic regression to predict KRs in the next year. Sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) were assessed. RESULTS: The most accurate models for predicting KR incorporated KOOS KP and KOOS QoL at a specific annual visit, and KOOS KP worsening over the previous year combined as weighted sums. Per 10 points of these assessments the odds of vKR increased by 30 % to 65 %. AUCs ranged from 0.87 to 0.92 (vKR1). Constraining KOOS KP and KOOS QoL to be worse at follow-up than at baseline, AUCs exceeded 0.85 (vKR2). When KOOS KP and KOOS QoL remained unchanged or worsened over one year, AUCs were >0.80 (vKR3). CONCLUSION: The three vKR criteria represent potential clinical multi-component outcomes for clinical trials of knee OA. After further validation, any of them may serve as a standalone clinical outcome or in combination with actual knee replacement.