ObjectivesThe progression of knee osteoarthritis (OA) can be defined as either radiographic progression or pain progression. This study aimed to construct models to predict radiographic progression and pain progression in patients with knee OA.MethodsWe retrieved data from the FNIH OA Biomarkers Consortium project, a nested case-control study. A total of 600 subjects with mild to moderate OA (Kellgren-Lawrence grade of 1, 2, or 3) in one target knee were enrolled. The patients were classified as radiographic progressors (n = 297), non-radiographic progressors (n = 303), pain progressors (n = 297), or non-pain progressors (n = 303) according to the change in the minimum joint space width of the medial compartment and the WOMAC pain score during the follow-up period of 24–48 months. Initially, 376 variables concerning demographics, clinical questionnaires, imaging measurements, and biochemical markers were included. We developed predictive models based on multivariate logistic regression analysis and visualized the models with nomograms. We also tested whether adding changes in predictors from baseline to 24 months would improve the predictive efficacy of the models.ResultsThe predictive models of radiographic progression and pain progression consisted of 8 and 10 variables, respectively, with area under curve (AUC) values of 0.77 and 0.76, respectively. Incorporating the change in the WOMAC pain score from baseline to 24 months into the pain progression predictive model significantly improved the predictive effectiveness (AUC = 0.86).ConclusionsWe identified risk factors for imaging progression and pain progression in patients with knee OA over a 2- to 4-year period, and provided effective predictive models, which could help identify patients at high risk of progression.