Cross-validation of good versus poor self-reported outcome trajectory types following knee arthroplasty

Abstract

To determine whether good versus poor outcome trajectories and predictors of poor outcome obtained in a derivation study could be replicated in an independent sample of persons undergoing knee arthroplasty (KA). We used data from TRIO-POPULAR, a prospective cohort study of 926 participants who underwent KA in the United Kingdom. The participants were assessed preoperatively and 6-weeks, 6- and 12-months postoperatively. The Oxford Knee Score was the primary outcome and a variety of pre-operative predictors of outcome were selected. The outcome measure and the predictors were selected to most closely align with a previously published derivation study of good versus poor outcome. Confirmatory two-piece latent class growth curve analyses were used to model outcome and regression was used to identify predictors of outcome class. Trajectories for the Oxford Knee Scores from TRIO-POPULAR replicated trajectories for WOMAC Pain and Function scores from the previously published derivation study. Multivariable predictors of poor outcome were pain catastrophizing (odds ratio=1.125, 95% CI=1.048, .206, p=0.001) and comorbidity (odds ratio=1.134, 95%CI=1.049, 1.227, p=0.002. Pain catastrophizing also predicted poor outcome in the derivation study. Good and poor outcome trajectories replicated those found in the previously published derivation study. Our model-based method produces stable outcome trajectories despite using data from different countries and participants with substantively different characteristics. Predictors of poor outcome were somewhat inconsistent between the cross-validation and derivation studies. Pain catastrophizing was the only consistent poor outcome predictor.