Abstract
Cumulative incidence of revision provides a measure of the failure rate of joint replacements and can be used to project demand for revisions. The most commonly applied survival analysis method (Kaplan-Meier [KM]) does not account for competing risks (eg, death). The authors compared the cumulative incidence function (CIF), a competing risks method, with the KM method through application to population-based cohorts. They measured time to revision, death, or censoring for unilateral total hip arthroplasty (THA; n=12,496) and total knee arthroplasty (TKA; n=19,172) cohorts in administrative databases in Alberta and TKAs (n=80,177) in the Swedish Knee Arthroplasty Register. The authors compared relative differences between the KM and CIF. They fitted Cox, Fine and Gray, and Royston and Parmar regression models and compared coefficients, standard errors, and P values. On sensitivity analysis, the authors included staged bilateral operations. Kaplan-Meier estimates exceeded the CIF at each time point. The magnitude of overestimation increased with follow-up time and was greatest for the Swedish cohort. At 5 years, relative differences between KM and CIF estimates for the Alberta THA and TKA and Swedish TKA cohorts were 1.8%, 2.3%, and 3.8%, respectively. These differences increased to 3.1%, 5.8%, and 8.2%, respectively, at 9 years, reaching 39.1% at 20 years (Swedish cohort). On sensitivity analysis (including staged bilateral operations), the Fine and Gray subdistribution hazard ratio differed from the Cox and Royston and Parmar hazard ratios. When the frequency of competing risks is high, competing risks methods are recommended to obtain accurate cumulative incidence estimates for informing health care planning and decision making. © 2021 Slack Incorporated. All rights reserved.