The influence of immortal time bias in observational studies examining associations of antifibrotic therapy with survival in idiopathic pulmonary fibrosis: A simulation study

Abstract

BackgroundImmortal time bias (ITB) has been overlooked in idiopathic pulmonary fibrosis (IPF). We aimed to identify the presence of ITB in observational studies examining associations between antifibrotic therapy and survival in patients with IPF and illustrate how ITB may affect effect size estimates of those associations.MethodsImmortal time bias was identified in observational studies using the ITB Study Assessment Checklist. We used a simulation study to illustrate how ITB may affect effect size estimates of antifibrotic therapy on survival in patients with IPF based on four statistical techniques including time-fixed, exclusion, time-dependent and landmark methods.ResultsOf the 16 included IPF studies, ITB was detected in 14 studies, while there were insufficient data for assessment in two others. Our simulation study showed that use of time–fixed [hazard ratio (HR) 0.55, 95% confidence interval (CI) 0.47–0.64] and exclusion methods (HR 0.79, 95% CI 0.67–0.92) overestimated the effectiveness of antifibrotic therapy on survival in simulated subjects with IPF, in comparison of the time–dependent method (HR 0.93, 95% CI 0.79–1.09). The influence of ITB was mitigated using the 1 year landmark method (HR 0.69, 95% CI 0.58–0.81), compared to the time–fixed method.ConclusionThe effectiveness of antifibrotic therapy on survival in IPF can be overestimated in observational studies, if ITB is mishandled. This study adds to the evidence for addressing the influence of ITB in IPF and provides several recommendations to minimize ITB. Identifying the presence of ITB should be routinely considered in future IPF studies, with the time–dependent method being an optimal approach to minimize ITB.

Methodological biases in observational hospital studies of COVID-19 treatment effectiveness: pitfalls and potential

“…In studies evaluating time-varying or time-dependent treatments addressing immortal-time bias is crucial, for which several options are available. Two commonly used approaches that can lead to severe immortal-time bias and result in flawed estimates of treatment effect: (i) including person-time and classifying patients as treated from time zero, even if they receive treatment later during follow-up, and (ii) excluding person-time, which is the time from baseline to treatment initiation for the exposure group ( 16 , 19 , 37 , 38 ). The landmark analysis is a design-based method involving setting fixed time as the landmark time and classifying patients according to their treatment status at the landmark ( 17 ).…”

“…The landmark analysis is a design-based method involving setting fixed time as the landmark time and classifying patients according to their treatment status at the landmark ( 17 ). Patients are then followed from the landmark time regardless of subsequent changes in their treatment status ( 17 , 37 ). However, this approach has two principal limitations: (i) the choice of the landmark time and (ii) the exclusion of patients who had an outcome before the landmark time from the analysis ( 15 , 28 ).…”

“…This usually requires defining a clinically meaningful grace period ( 33 , 34 ). In the sequential trial approach, a sequence of multiple nested trials with all potential time zeros is modeled ( 37 ). Each method has its own assumptions and limitations, which should be considered when interpreting study results.…”