Alternative survival analysis methods to estimate time to revision following hip and knee arthroplasty: Can the Kaplan-Meier method compete?

  • Lacny S
  • Faris P
  • Bohm E
  • et al.
N/ACitations
Citations of this article
6Readers
Mendeley users who have this article in their library.

Abstract

Purpose: With increased longevity and frequency with which joint replacements are being performed on younger, more physically active patients, it has become common for patients to outlive the life of their prosthesis and require a revision. Measuring the cumulative incidence of revision (i.e., revision rate) provides a measure of the rate of failure of joint replacements and can be used to project future demand for revisions. Due to varying patient follow-up times and censoring, survival analysis is required to estimate revision rates. The Kaplan-Meier (KM) method is the most commonly applied survival analysis method. However, it does not account for the competing risk of death and consequently overestimates the cumulative incidence of revision. This is problematic given the high rate of competing risks, especially in older patients receiving joint replacements. Although alternative methods that account for competing risks have been developed, they are rarely used within the arthroplasty literature or among joint replacement registries. Our objective was to assess alternative methods for estimating the cumulative incidence of revision through application to population-based cohorts. We evaluated these methods using hip and knee replacement data from Alberta, Canada, over a 10 year period and knee replacements recorded in the Swedish Knee Arthroplasty Register (SKAR) over 24 years. We recommend a preferred method for accurately estimating the cumulative incidence of revision that can be used to inform health policy decisions and future resource allocation. Methods: We measured the time to revision, death, or censoring for cohorts of uncemented total hip (n=12,496) and cemented total knee arthroplasties (TKAs) (n=19,172) recorded in administrative inpatient databases in Alberta from 2003 to 2013 and cemented TKAs (n=80,177) recorded in the SKAR from 1989 to 2012. Each cohort included primary unilateral procedures in patients with osteoarthritis. We estimated the unadjusted cumulative incidence of revision using the KM failure function (1-KM) and the cumulative incidence function (CIF), which accounts for competing risks. To quantify the overestimation of the KM method compared to the CIF, we calculated the relative difference (RD) ([{1-KM}-CIF]/CIF) between estimates at five, nine, and 23 years (Swedish knee cohort only) following primary surgery. We also compared three regression models that accounted for competing risks: the Cox proportional hazards model, Fine and Gray subdistribution hazards model, and Royston and Parmar flexible parametric model. We adjusted for age, sex, Charlson Comorbidity Score (Alberta cohorts only) and year of primary operation (Swedish cohort only). To examine differences between regression models, we compared the magnitude of coefficients (hazard ratio [HR] for the Cox and Royston and Parmar models, subdistribution HR for the Fine and Gray model), standard errors (SE), and Wald statistic P-values. Our base case analysis included primary unilateral operations only. In sensitivity analysis, we included unilateral and staged-bilateral operations to compare the performance of each model using larger sample sizes and where the assumption of independence between subjects was violated. Results: The KM method and CIF produced similar estimates at follow- up times shorter than 10 years for all three cohorts, although the KM cumulative incidence was greater than the CIF at each time point. The magnitude of overestimation at each follow-up time was greatest for the Swedish cohort, which also had the highest cumulative incidence of death. The extent of overestimation also increased with follow-up time and became substantial beyond 10 years. At five years, the relative increase in estimation of the cumulative incidence estimated using the KM method compared to the CIF for the Alberta hip, Alberta knee, and Swedish knee cohorts was 1.8%, 2.3%, and 3.8%, respectively. At nine years, the RDs increased to 3.1%, 5.8%, and 8.2%, respectively. At 23 years, the RD for the Swedish cohort reached 39.1%, where the cumulative incidence estimated using the KM method was 7.4% compared to 5.3% for the CIF. Each adjusted regression model yielded similar coefficients, SEs, and P-values for our base case analysis. However, in our sensitivity analysis the Fine and Gray subdistribution HR for sex deviated from the HRs obtained from the Cox and Royston and Parmar models (P=0.365, P=0.039, P=0.045, respectively). Conclusions: Our results support the application of the CIF and competing risks regression models to more accurately estimate the cumulative incidence of revision following joint replacement surgery.

Cite

CITATION STYLE

APA

Lacny, S., Faris, P. D., Bohm, E., Woodhouse, L. J., Robertsson, O., & Marshall, D. A. (2015). Alternative survival analysis methods to estimate time to revision following hip and knee arthroplasty: Can the Kaplan-Meier method compete? Osteoarthritis and Cartilage, 23, A197–A198. https://doi.org/10.1016/j.joca.2015.02.991

Register to see more suggestions

Mendeley helps you to discover research relevant for your work.

Already have an account?

Save time finding and organizing research with Mendeley

Sign up for free