Different competing risks models applied to data from the Australian Orthopaedic Association National Joint Replacement Registry

Purpose Here we describe some available statistical models and illustrate their use for analysis of arthroplasty registry data in the presence of the competing risk of death, when the influence of covariates on the revision rate may be different to the influence on the probability (that is, risk) of the occurrence of revision. Patients and methods Records of 12,525 patients aged 75-84 years who had received hemiarthroplasty for fractured neck of femur were obtained from the Australian Orthopaedic Association National Joint Replacement Registry. The covariates whose effects we investigated were: age, sex, type of prosthesis, and type of fixation (cementless or cemented). Extensions of competing risk regression models were implemented, allowing the effects of some covariates to vary with time. Results The revision rate was significantly higher for patients with unipolar than bipolar prostheses (HR = 1.38, 95% CI: 1.01-1.89) or with monoblock than bipolar prostheses (HR = 1.45, 95% CI: 1.08-1.94). It was significantly higher for the younger age group (75-79 years) than for the older one (80-84 years) (HR = 1.28, 95% CI: 1.05-1.56) and higher for males than for females (HR = 1.37, 95% CI: 1.09-1.71). The probability of revision, after correction for the competing risk of death, was only significantly higher for unipolar prostheses than for bipolar prostheses, and higher for the younger age group. The effect of fixation type varied with time; initially, there was a higher probability of revision for cementless prostheses than for cemented prostheses, which disappeared after approximately 1.5 years. Interpretation When accounting for the competing risk of death, the covariates type of prosthesis and sex influenced the rate of revision differently to the probability of revision. We advocate the use of appropriate analysis tools in the presence of competing risks and when covariates have time-dependent effects.