Background: In arthroplasty data, patients with staged bilateral total joint arthroplasty (TJA) pose a problem in statistical analysis. Subgroup analysis, in which patients with unilateral and bilateral TJA are studied separately, is sometimes considered an appropriate solution to the problem; we aim to show that this is not true because of immortal time bias.
Methods: We reviewed patients who underwent staged (at any time) bilateral TJA. The logical fallacy leading to immortal time bias is explained through a simple artificial data example. The cumulative incidences of revision and death are computed by subgroup analysis and by landmark analysis based on hip replacement data from the Dutch Arthroplasty Register and on simulated data sets.
Results: For patients who underwent unilateral TJA, subgroup analysis can lead to an overestimate of the cumulative incidence of death and an underestimate of the cumulative incidence of revision. The reverse conclusion holds for patients who underwent staged bilateral TJA. Analysis of these patients can lead to an underestimate of the cumulative incidence of death and an overestimate of the cumulative incidence of revision. Immortal time bias can be prevented by using landmark analysis.
Conclusions: When examining arthroplasty registry data, patients who underwent staged bilateral TJA should be analyzed with caution. An appropriate statistical method to address the research question should be selected.
1Mathematical Institute of Leiden University, Leiden, the Netherlands
2Department of Medical Statistics, Leiden University Medical Center, Leiden, the Netherlands
3Department of Orthopaedic Surgery, Leiden University Medical Center, Leiden, the Netherlands
E-mail address for S.L. van der Pas: firstname.lastname@example.org