Abstract| Volume 30, SUPPLEMENT 1, S31-S32, April 2022


      Purpose: There is a paucity of prognostic prediction models for total knee replacement (TKR) and the role of radiographic findings in predicting TKR remains unclear. We aimed to develop and validate predictive models for TKR and assess whether adding radiographic findings improves predictive performance using data from the Multicenter Osteoarthritis Study (MOST) and the Osteoarthritis Initiative (OAI)
      Methods: We included participants who reported knee pain in the past 3 months at their baseline visit and followed them up until 60 months so that participants of MOST and OAI had the same length of follow-up time. We selected only one knee from each participant based on the severity of knee pain measured by either WOMAC pain questionnaire or other instruments. TKRs that occurred during the 60 month follow-up period in both cohorts were identified and adjudicated by medical records and/or post-operative radiographs. The date of TKR was obtained from medical records or self-report by participants when medical records were unavailable. We identified 26 potential predictors, collected and assessed at baseline visit, based on an expert review of the current studies on the risk factors of TKR by a census group of orthopedic surgeons led by the chief investigator (Table 1). We developed two predictive models for the risk of TKR within 60 months by fitting Cox proportional hazard models among participants in the MOST. The first model included socio-demographic and anthropometric factors, medical history and clinical measures (i.e., clinical model). The second model added radiographic findings into predictive model (i.e., radiographic model). A significance level of 0·20 was chosen so that important predictors relevant to TKR would not be missed and to avoid deleting less significant ones that may have practical and clinical implication. We performed a 10-fold cross-validation to minimize overfitting. We evaluated each model’s discrimination and calibration performance and assessed the incremental value of radiographic findings using both category-free net reclassification improvement (NRI) and integrated discrimination improvement (IDI). We tuned the models and externally validated among participants in the OAI.
      Results: We included 2658 participants of the MOST (mean age=62·4 years (SD=8·1), 1646 (61·9%) women) and 4060 participants of the OAI (mean age=60·9 years (SD=9·1), 2379 (58·6%) women). The C statistic was 0·79 (95% CI: 0·76-0·81) for the clinical model and 0·87 (95% CI: 0·85-0·99) for the radiographic model. The calibration slope was 0·95 (95% CI: 0·86-1·05) and 0·96 (0·87-1·04), respectively. Adding radiograph findings significantly improved predictive performance with an NRI of 0·42 (95% CI: 0·34-0·48) and IDI of 0·13 (95% CI: 0·09-0·16). The clinical model and radiographic model were tuned with an adjustment size of 0·839 and 0·941, respectively, using the validation dataset. C statistics of tuned clinical model and radiographic model were 0·78 (95% CI: 0·71-0·85) and 0·88 (95% CI: 0·84-0·92), respectively (Table 2). The calibration slopes of clinical model and radiographic model were 1·004 (95% CI: 0·81-1·19) and 1·03 (95% CI: 0·95-1·10), with a corresponding intercept of -0·004 (95% CI: -0·18-0·18) and -0·02 (95% CI: -0·09-0·04), respectively. Adding radiographic findings significantly improved in predictive performance with NRI of 0·51 (95% CI: 0·39-0·59) and an IDI of 0·09 (95% CI: 0·06-0·11) (Table 2). Free access to the tuned models was provided through We also derived nomograms to graphically present the predictive models (Figure 1).
      Conclusions: While the risk of TKR can be predicted based on common risk factors with good discrimination and calibration , adding radiographic findings of knee OA into the predictive model substantially improves the predictive performance. This study adds empirical evidence that radiographs, a commonly collected information in routine clinical care, significantly improved performance of predicting future TKR. Patients and clinicians should be informed by these findings that the risk of TKR can be accurately estimated using several common risk factor, but radiographs, if available, should be included and provide added value.
      Table 2Performance in discrimination and calibration and measures of improvement of the radiograp
      MOST (Training dataset), n=2658OAI (Validation dataset), n=2932
      Clinical modelRadiographic modelClinical modelRadiographic model
      C-statistic0·79 (0·76-0·81)0·87(0·85-0·89)0·78 (0·71-0·85)0·88 (0·84-0·92)
      Calibration slope0·95 (0·86-1·05)0·96 (0·87-1·04)1·004 (0·81-1·19)1·03 (0·95-1·10)
      Calibration intercept0·01 (-0·08-0·11)0·01 (-0·04-0·05)-0·004 (-0·18-0·18)-0·02 (-0·09-0·04)
      NRIReference0·43 (0·38-0·50)Reference0·51 (0·39-0·59)
      IDIReference0·14 (0·10-0·18)Reference0·09 (0·06-0·11)