11 – Session in Honor of 70th Birthday of Stephen E. Fienberg and His Nearly 50 Years of Statistical Practice

Construction and Validation of a Parametric Model for Predicting Implant Survivorship Beyond Observed Data in Total Joint Arthroplasty

at Biomet Orthopedics

In the orthopaedic space, long-term implant survivorship is a critical measure to determine product performance. Construction of a sound statistical model that will predict implant survival beyond the time window of available data is of strong interest and would allow manufacturers to conduct more proactive surveillance on product performance. An accelerated failure time model was fit to short-term survival data from a multi-center clinical trial for a total joint replacement product. The model was used to estimate the survival probability at all time points, including those outside the time window of available data. The model building, model selection, and model diagnostics were demonstrated. Model validation (both internal and external) was performed. An Exponential model (intercept only) was found to be the best-fit model for the data, and satisfied the model assumptions and goodness-of-fit. The internal validation confirmed that the estimates were relatively stable although censoring patterns appear to affect the estimates to some extent. External validation confirmed that this model reliably estimates survivorship within the range of available follow-up time and for approximately 8 years past this interval. Model-predicted survivorship and associated 95% confidence intervals were calculated within the time window of available data and for 8 years outside this interval. This methodology is a viable alternative to the traditional data acquisition methods, and may allow for faster attainment of long-term survivorship information and hence more proactive surveillance on product performance.