Identifying, handling and impact of immortal time bias on addressing treatment effects in observational studies using...

Immortal time bias (ITB) represents a methodological challenge in evaluating treatment effects in observational studies using routinely collected data (RCD). However, the prevalence of ITB, the strategies used to address ITB and its impact remain inadequate. This study aimed to investigate how ITB was identified and handled in observational studies using RCD, and to assess its impact on treatment effect estimates.

A systematic search was performed in PubMed for observational studies published from 2018 to 2020 that used RCD to evaluate drug treatment effects. We examined the synchronization of three time points (eligibility, treatment assignment, and the start of follow-up) to identify ITB and assessed the risk of ITB. For low-risk studies, we summarized the handling approaches. For high-risk studies, we conducted quantitative bias analyses to correct for ITB and calculate ITB-controlled estimates. These ITB-controlled estimates were then compared with original estimates to quantify the impact of ITB.

Among the 256 studies initially identified, 162 cohort studies with time-to-event outcomes were included. 13 studies (8.0%) lacked sufficient reporting to assess ITB. Of the remaining studies, 35 studies (21.6%) were classified as high risk for ITB, while 114 studies (70.4%) were classified as low risk, with 15 having naturally synchronized time points and 99 using design or analytical approaches to synchronize them. For the 99 low-risk studies, the commonly employed approaches were the active comparator new-user design and the time-varying exposure definition, accounting for 56.6% and 19.2%, respectively. Of the 35 high-risk studies, 16 studies that provided sufficient information for correction were included in the quantitative bias analyses. Among these, 4 studies (25%) showed statistically significant differences between ITB-controlled and original estimates, and 4 studies (25%) yielded conflicting conclusions regarding the statistical significance of these two estimates. Only 5 of the 35 high-risk studies (14.3%) discussed that the results may be affected by ITB.

ITB is a critical methodological issue in observational studies using RCD, with the potential to significantly distort conclusions. To enhance the validity of treatment effect estimates, researchers should thoroughly examine the presence of ITB and employ appropriate strategies to mitigate its impact.