[Is Kaplan-Meier statistics the most appropriate tool for survivorship measurement of outcomes in orthopaedics?].

PURPOSE OF THE STUDY: An assessment of the results over a time period is an integral part of any treatment evaluation. A standard method used for this purpose is Kaplan-Meier survival analysis. The aim of this study was to show how, for total hip arthroplasty (THA), estimates of prosthesis survival probability can be affected by a factor known as competing events. MATERIAL AND METHODS: A set of input data concerning competing events (death and loss to follow-up), in addition to the investigated event (revision THA), was analysed by both the Kaplan-Meier method and the method of cumulative incidence with the use of worst-case analysis. RESULTS: Our results showed that, for the same input data, the Kaplan-Meier method gave lower estimates of prosthesis survival probability than the method of cumulative incidence. This difference increased with an increasing number of competing events occurring during the clinical study, and with an increasing duration of follow-up. These survival probability estimates can markedly be influenced by the use of worst-case analysis In our set of data this was more than 30%. DISCUSSION: Reports in the orthopaedic literature still show a predominant use of the Kaplan-Meier method, although it is obvious that this is not always optimal for observational clinical studies. Also, it is not clear how beneficial is the use of worst-case analysis, because our and other authors' results suggest that not all patients lost to follow-up should be considered as failed. Therefore, insolvement of such analysis might markedly distort the real survival curves and to disadvantage evaluated the orthopaedic method. CONCLUSIONS: In observational clinical studies involving a higher number of competing events, it is preferable to use the cumulative incidence method rather than Kaplan Meier analysis for statistical evaluation of the results. The former gives more exact estimates of prosthesis survival probability. It is also recommended to avoid modifying survival curves indiscriminately according to the results of worst-case analysis.