CORR Insights®: What Is the Effect of Using a Competing-risks Estimator when Predicting Survivorship After Joint Arthroplasty: A Comparison of Approaches to Survivorship Estimation in a Large Registry

Where Are We Now?

An accurate and reliable measure of any outcome event in healthcare is imperative. Poor clinical and policy decisions can be made if data are inaccurate or biased. A competing risk in survival analysis refers to a scenario in which measuring an outcome event of interest might be influenced by another event, and thus causes a potentially biased estimation. For example, in studying revision surgery after TKA, mortality might be a competing risk; those who die during surveillance cannot, by definition, have revision surgery. The potential bias induced by death unrelated to revision joint arthroplasty has been illustrated by Wongworawat et al. [10]. Competing risks can be an important source of bias when seeking to interpret the results of time-to-event (survival) analyses, which only measure surveillance over a specified observational period. Patients who do not incur the outcome event of interest by the end of the time during which they were observed are “censored,” and their data are only informative for the period of their surveillance. Censoring may occur because a patient is lost to follow-up or experiences another event that obviates ongoing surveillance, or because the surveillance period ends. A traditional statistical approach is to censor the surveillance of patients at the time of their death, which assumes censoring is “independent” and that such patients would otherwise have an outcome event rate similar to those who did not die. Because these traditional approaches are thought to overestimate revision rates, statistical methods to account for the competing risk are often used in joint arthroplasty [4, 5, 8].

Understanding the competing risk of mortality and its effect on other relevant events such as reoperation, all-cause readmission, pulmonary embolism, or revision surgery is important, especially as health policy moves toward weighing reimbursement based on surgical outcomes; it is equally important in reporting performance indicators across providers or institutions. Previous studies have described the overestimation bias from failing to account for competing risks and have recommended more robust methods to account for this bias in arthroplasty [6, 7].

The nicely detailed analysis by Cuthbert et al. [3] adds to this knowledge by comparing survivorship after TKA based on whether a competing-risk model was used, over a range of age groups, and by setting covariates to a representative sample for “low-risk,” “average-risk,” and “high-risk” cohorts. They found that the traditional Cox proportional-hazards survival regression model with censoring slightly overestimated revisions of TKA among older patients, but the magnitude was small and fairly negligible. For THA, the traditional approach also overestimated the rate of revision because of periprosthetic fracture, and this overestimation was particularly evident in the "high-risk" representative patients. Thus, a competing-risk approach is more important in situations where both the frequency of the endpoint in question (for example, revision for periprosthetic fracture) and the competing risk (such as mortality) are common. These findings can have a direct bearing on quality and safety reporting, as well as for policymakers relying on revision events as an outcome. Because a competing risk can hinder or modify the estimate of an outcome event of interest, there are many situations in orthopaedics where it becomes especially important.

Where Do We Need To Go?

Efforts should extend these models to patients with other orthopaedic conditions, especially in cohorts that are disproportionately older and frail, such as patients with femoral neck fractures, distal femur fractures, or cervical odontoid fractures. Future efforts should follow guidelines for when a competing-risk approach is most needed. For example, as a “rule of thumb” for which mortality as a competing risk should be considered, Wongworawat et al. [10] recommend an event rate greater than 10% and when surveillance approaches 10 years. Similarly, Martin et al. [7] illustrated its importance among older cohorts. A future direction might be to demonstrate the bias among those with greater comorbidity. How does the duration of surveillance of an outcome event of interest play into the requirement for a competing-risk model? Because the ability to confirm bias might directly inform public policy, how does the size of a study cohort impact the ability to detect bias induced by not incorporating a competing-risk method? How do relatively higher rates of loss to follow-up or censored data impact the competing-risk analysis? Finally, an important future direction is to understand how competing-risk models might be integrated into clinical decision-making, such as the choice of implant among high-risk patients, and to inform policymaking, such as ranking hospital readmission rates.

How Do We Get There?

Registries such as the Australian Orthopaedic Association Joint Replacement Registry [2], the American Academy of Orthopaedic Surgeons American Joint Replacement Registry [1], or the Swedish Knee Arthroplasty Register [9], to name just a few, provide an ideal setting for methodologic research on competing risks. Registries may combine data across multiple health systems and generally include a large number of patients to draw from. Although registries generally do not control for unobserved confounding, they are more likely to be generalized to the real world or a pragmatic setting and can better facilitate methodologic studies of competing risk and surveillance bias. The future growth and interest of orthopaedic registries spearheaded by professional societies will enable future studies to confirm, expand, and further elucidate the understanding of competing risks for outcome events in orthopaedic patients. In doing so, they will be better equipped to provide reliable, unbiased, and robust evaluations of outcome events of interest in the context of competing risks.