Using two predictive models captures two types of poor outcomes in knee arthroplasty: A multisite longitudinal cohort study

Objectives Poor outcome following knee arthroplasty (KA), a common major surgery worldwide, reportedly occurs in approximately 20% of cases. These patients demonstrate minimal improvement, at least moderate knee pain, and difficulty performing many routine daily activities. The purposes of our study were to comprehensively determine poor outcome risk following KA and to identify predictors of poor outcome. Methods Data from 565 participants with KA in the Osteoarthritis Initiative and the Multicenter Osteoarthritis studies were used. Previously validated latent class analyses (LCA) of good versus poor outcome trajectories of WOMAC Pain and Disability were generated to describe minimal improvement and poor final outcome. The modified Escobar RAND appropriateness system was used to generate classifications of appropriate, inconclusive and rarely appropriate. Multivariable prediction models included LCA‐based good versus poor outcome, modified Escobar classifications and evidence‐driven preoperative prognostic variables. Results Modified Escobar appropriateness classifications were non‐significant predictors of WOMAC Pain good versus poor outcomes indicating the methods provide independent outcome estimates. For WOMAC Pain, approximately 34% and for WOMAC Disability, 45% of participants had high probability of either minimal improvement via “rarely appropriate” classifications or poor outcome via LCA. In multivariable prediction models, greater contralateral knee pain consistently predicted poor outcome (e.g., OR = 1.21, 95%CI=1.10, 1.33). Conclusions Appropriateness criteria and LCA estimates provided combined poor outcome estimates that were approximately double the commonly reported poor outcome of 20%. Rates of poor outcome could be reduced if clinicians screened patients using Appropriateness criteria and LCA predictors prior to surgery to optimize outcome.