Does the method of handling immortal time bias impact the estimated clinical effectiveness of ICDs in patients with heart failure?
Handling immortal time bias by exclusion overestimates the treatment effect of ICDs, whereas the landmark method provides results comparable to the time-dependent Mantel-Byar method.
OBJECTIVES: To assess the extent of immortal time bias in estimating the clinical effectiveness of implantable cardioverter-defibrillators (ICDs) and the impact of methods of handling immortal time bias. STUDY DESIGN AND SETTING: Retrospective population-based cohort study of patients with heart failure in a national registry linked to Medicare claims (2003-2008). We compared three methods of handling immortal time bias, namely the Mantel-Byar (or time-dependent exposure assignment), the landmark, and the exclusion methods. RESULTS: Of the 5,226 study patients, 1,274 (24.4%) received ICD therapy. Total person-years in the Mantel-Byar method were 2,639, or 490 more than that in the exclusion method, reflecting potential immortal time in the study. The exclusion method yielded a hazard ratio of 0.71 (95% confidence interval CI: 0.63-0.80), which was 16% lower than the Mantel-Byar method (0.84; 95% CI: 0.75-0.95). The 120-day landmark method yielded similar results to those produced by the Mantel-Byar method (0.82; 95% CI: 0.72-0.95). CONCLUSIONS: Immortal time bias was detected in the ICD clinical effectiveness study, which might have led to substantial bias overestimating the treatment effect if handled by exclusion. When an appropriate landmark was selected, that method yielded similar hazard ratios to those obtained by the Mantel-Byar method, supporting the validity of the landmark method.
Mi et al. (Tue,) studied this question.