Competing risks in orthopedic data
In close collaboration with the Dutch Arthroplasty Register, we investigated how the competing risks nature of their data needs to be incorporated in analyzing the question: "How long until I'll need revision surgery after hip replacement?"
Total hip replacement is a very common surgery in The Netherlands, with approximately 40.000 procedures registered every year. A question of great interest is how long it takes until revision surgery is needed after the initial procedure. This is a question answered by survival analysis techniques. A particular feature of the data, available from the Dutch Arthroplasty Register, is that there is a competing risk: a patient may die before their prosthesis needs revision.
Neglecting to take into account the competing risk nature of the data may lead to overestimates of the probability of revision, especially as follow-up increases. We provide recommendations for performing adjusted analyses of total joint replacement data.
Another complication arises because patients have two hips, and both can be replaced. Thus a patient can appear multiple times in a data set, creating a dependence issue. There is a risk of immortal time bias when this is handled incorrectly. We described how this immortal time bias may occur, and provide recommendations for practice.
Van der Pas, S., Nelissen, R., & Fiocco, M. (2018). Different competing risks models for different questions may give similar results in arthroplasty registers in the presence of few events: Illustrated with 138,234 hip (124,560 patients) and 139,070 knee (125,213 patients) replacements from the Dutch Arthroplasty Register. Acta Orthopaedica, 89(2), 145-151. [link]
Van der Pas, S. L., Nelissen, R. G., & Fiocco, M. (2017). Patients with staged bilateral total joint arthroplasty in registries: Immortal time bias and methodological options. JBJS, 99(15), e82. [link]