I had a chance to collaborate with Nordic Arthroplasty Register Association in a study published in the Acta Orthopaedica last year. Study was titled “The effect of fixation type on the survivorship of contemporary total knee arthroplasty in patients younger than 65 years of age: a register-based study of 115,177 knees in the Nordic Arthroplasty Register Association (NARA) 2000–2016“. This was the first time I had a chance to analyze a really large total joint replacement data set. During the analyses I become very concerned about the quality of previous TJR studies.
In my analyses I detected several violations in the proportional hazards assumption when I was using all-cause revision as an end-point in the Cox regression analysis. This means that all possible causes of revision, such as infection, instability, aseptic loosening, fracture, pain, polyethylene wear etc., were used as an composite end-point. My finding was, however, very logical.
All different causes for revision are not made equal. For example infection can be acute, subacute, chronic or late. Acute infection may be due to poor patient compliance or comorbidities such as diabetes resulting to wound contamination and subsequent deep infection. Late infection is usually hematogenous meaning some bacteria got into the patients blood stream and then finding it´s way to the knee implant. Instability may be due to an iatrogenic perioperative soft tissue injury or instability may manifest lately due to a knee injury. Even under single cause, there are very different “disease” entities which may lead to revision surgery. Hence it is only natural that hazard for age, gender or BMI are not constant thus not being proportional for whole study period which usually spans 5 to 15 years.
All this made me wonder how often this finding has gone unnoticed in the previous TJR studies. Firstly, if analyst does to test or inspect PH assumptions, any violation will surely go unnoticed. Secondly, sometimes the sample size may be too small to detect any violations. After all, overall revision rate in modern TJR is well below 10% at ten years.
Obvious issues with all-cause revision as an end-point is surely a source of irreproducibility in the TJR research. This is very important aspect which should be discussed more often in the future and be also be investigated in with metaresearch.