Abstract
Objectives
Immortal time bias (ITB) occurs when a period during which, by design, participants cannot experience the outcome (like death) is incorrectly included in the treatment group’s follow-up, artificially making the treatment look better than it truly is. We aimed to identify a systematic sample of cases of ITB in the literature of studies using survival analysis and assess the impact of ITB on the results.
Design
Meta-epidemiological study (PROSPERO[CRD42022356073]).
Setting
We searched PubMed/MEDLINE, Embase and Cochrane Database of Systematic Reviews from database inception to August 2024. Systematic reviews with quantitative syntheses that allowed subgroup analysis by the presence of ITB for any available exposure-outcome pairs (‘topics’) were eligible for inclusion.
Participants
Participants included in the systematic reviews.
Main outcome measures
Information on ITB and effect sizes (ESs) with 95% confidence interval for individual studies in forest plots were extracted to run re-analysis using generic inverse variance fixed- and random-effects methods. After extracting data, we conducted subgroup analysis by the presence of ITB for all available topics and assessed the impact of ITB on the heterogeneity (I2), vulnerability of evidence (or conclusion), statistical significance of the finding, and altering ES in favour of intervention/exposure.
Results
The median (interquartile range (IQR)) number of studies included for a topic was 6 (4–10). Across 25 topics (including 182 studies), 44.0% of the eligible studies (80 studies) were affected by ITB. Among the 21 topics where both studies with ITB and studies without ITB were available (four topics only had studies unaffected by ITB), 57.1% (12/21) demonstrated statistically significant results only in studies with ITB (n = 11 topics) or only in studies without ITB (one topic). In 23.8% (5/21), the overall summary results changed from statistically significant to non-statistically significant or vice versa after excluding studies with ITB. The ratio of ES – summary ES from studies with ITB relative to summary ES from studies without ITB – was 0.71 (95% CI, 0.66-0.78), suggesting that the ES from studies with ITB was larger by an average of 29% in favour of the intervention/exposure. Excluding studies involving ITB reduced between-study heterogeneity (I2) by 21.4% on average.
Conclusions
ITB can be common among studies in some medical areas, and its presence may substantially inflate the ESs and lead to misleading, exaggerated evidence.
Keywords: Immortal time bias, guarantee time bias, meta-epidemiology, meta-analysis, cancer research, drug
Introduction
Since observational studies are cost-effective and feasible to conduct in real-world settings, they have offered a practical measure to assess the effectiveness and safety of treatments, and have become the main sources of informing clinical decisions in advance of clinical trial results.1–3 They often measure time-to-event outcomes such as overall survival, yet a well-described bias in observational research that is heavily relevant to time-to-event outcomes – immortal time bias (ITB) – has frequently been overlooked.1,3
The ITB stems from an immortal time interval between enrolment (or cohort entry) and receipt of the study drug. 4 The immortal time, also called guarantee time, refers to a period of follow-up during which death or the outcome of interest cannot occur; those who are allocated in the drug or exposed group inevitably survive (alive and event-free) from cohort entry to the timepoint where first exposure/prescription is defined. 3 When follow-up time of exposed and unexposed groups begins identically at, for instance, cohort entry, and if there is some delay for a prescription to be dispensed, the drug group gains immortal survival period (i.e. until drug prescription), whereas the unexposed group (i.e. drug non-user) does not secure any guaranteed survival time. 3 Such misclassification of the unexposed period (immortal time) in the exposed group would lengthen the overall survival period and favour the drug by conferring a spurious survival advantage.3,5
ITB has been acknowledged most frequently in pharmacoepidemiology, 3 but it is not specific to pharmacoepidemiologic studies. It can affect any type of cohort measuring any type of association, if there is a divergence between the time of onset of follow-up and onset of exposure. The prevalence of ITB could be high, but its estimation is hampered by the fact that most authors either do not recognise it or do not acknowledge it.3,6 Herein, we conducted a meta-epidemiology study that aimed to systematically assess cases of ITB in studies that used survival analysis and estimate the impact of ITB. To achieve this objective, we systematically overviewed meta-analyses where their authors considered ITB and provided data that allowed estimation of effect sizes (ESs) in studies with and without ITB.
Methods
Identification of relevant meta-analyses
We systematically searched ITB in PubMed/MEDLINE, Embase and the Cochrane Database of Systematic Reviews (CDSR) from database inception to August 2024, without any language restriction, to collect systematic reviews with quantitative syntheses that provided subgroup analysis by presence of ITB or judged ITB for any available exposure-outcome pairs (‘topics’). One exposure-outcome pair consists of two exposure-outcome subgroup analyses – with and without ITB. The search terms included: (immortal-time-bias[tiab] OR time-to-treatment-bias[tiab] OR guarantee-time-bias[tiab]) AND meta-analysis[pt]. We excluded duplicate studies, and systematic reviews without quantitative synthesis and those that did not evaluate ITB for included studies were also excluded from the meta-epidemiological review. Two authors (MSK and DKY) reviewed titles and abstracts to isolate relevant studies. Next, the authors accessed the full text and supplementary materials to identify studies that contained at least one ITB-based subgroup analysis. Two authors (MSK and JIS) arbitrated all potential discrepancies. The study protocol was published in PROSPERO (CRD42022356073).
Data extraction
For all eligible studies, we manually extracted several characteristics: author, publication year, exposure (i.e. drugs), outcome measures (i.e. overall survival), outcome conditions (i.e. ovarian cancer), number of studies included in the meta-analysis where ITB was present, number of studies included in the meta-analysis where ITB was absent, the traditional statistical significance of findings (‘positive’, if p < 0.05), summary ESs (i.e. hazard ratio (HR) or relative risk (RR)) with 95% confidence intervals (CIs) and heterogeneity (I2 metric) for overall and for ITB subgroup results. The quality of the included systematic reviews was evaluated using the AMSTAR2. 7 While compiling the eligible evidence, we encountered instances when a single systematic review presented multiple topics and summary ESs. We kept all the available topics to capture the relevant data and results separately for studies with and without ITB. When multiple meta-analyses were published for the same topic, we selected one with the largest number of included studies.
Meta-analysis
The authors of systematic reviews used different models (i.e. fixed- or random-effects model) and different types of weighting when summarising the results. To bring all 46 meta-analyses at level ground and make them more comparable, we performed by ourselves all meta-analyses using the generic inverse variance fixed- and random-effects methods throughout. Reanalysis with both models allowed us to isolate ITB's effects from variability introduced by differing original methodologies. Fixed-effect models assume all included studies share a single true ES. Any differences among observed effects are attributed to random (within-study) sampling error. In contrast, random-effects models assume the true effect can vary across studies due to differences in populations, interventions, settings, or other factors. This introduces both within-study and between-study variance into the model. Then, we performed subgroup analysis by the presence of ITB for all topics and yielded ES and heterogeneity for ITB subgroups. More than half of the meta-analyses did not present heterogeneity value (i.e. I2) for ITB subgroups, and as such, we could fill many missing gaps in data with complete set of meta-analyses.
Assessment of the impact of ITB
From the extracted information, we evaluated the impact of ITB on the amount of evidence, results, and conclusions. First, we calculated the prevalence of ITB by calculating the proportion of studies that involved ITB among all included individual studies (182 studies). Second, we calculated the discrepancy rate of topics where the results differed in the presence or absence of statistical significance (p < 0.05) in the summary ES from studies with ITB versus those without ITB. Third, we calculated the proportion of topics where the overall summary results changed from statistically significant to non-statistically significant or vice versa after excluding studies with ITB. Descriptive analyses show the percentage, mean with standard deviation (SD), or median with interquartile range (IQR).
Furthermore, we added the following exploratory analyses, acknowledging that they may be underpowered as the number of topics were limited. First, we assessed whether it was more common to have statistically significant results in the summary ES of studies with ITB versus studies without ITB using the McNemar test. Second, we assessed whether it was more common to have statistically significant results when ITB ratio – the proportion of studies with ITB among all studies included in meta-analysis for a given topic – is high using logistic regression model. Third, we examined whether it was more common to have more favourable estimates of benefit with the intervention in the summary ES of studies with ITB versus studies without ITB by calculating the ratio of ES.8,9 The ratio of ES – summary ES from studies with ITB relative to summary ES from studies without ITB – for topics were calculated, and then synthesised across all the topics with fixed- and random-effects inverse variance meta-analysis, so as to calculate the amount of average exaggeration of the ES with ITB. 8 Finally, to explore whether ITB is a potential source of heterogeneity in meta-analysis, we calculated heterogeneity (I2) before and after excluding studies affected by ITB, and descriptively demonstrated the change in I2 following the removal of studies with ITB.
Statistical software
All analyses were performed using ‘meta’ package of R software (version 3.6.0; R foundation, Vienna, Austria) and IBM SPSS statistics version 21.0 (SPSS Inc.; Chicago, IL, USA).
Results
Study selection
We identified 355 items potentially relevant to ITB from PubMed/MEDLINE, 764 from Embase, and 45 from CDSR. Among a total of 1164 items, 398 duplicates were removed, and 748 other items were further excluded after title and abstract screening (Figure 1). We assessed full text and supplementary materials for eligible studies and excluded two systematic reviews without quantitative synthesis or meta-analysis and four systematic reviews that have not evaluated ITB for their included observational studies. Finally, 12 systematic reviews that contained 25 eligible topics were retained. We obtained subgroup ESs and heterogeneity estimates for these 25 topics in studies with and without ITB (Supplementary Figure 1). For four topics, there was no study with ITB. We used all 25 topics and included individual studies to quantitatively assess the prevalence of ITB, and used 21 topics with both ITB subgroups to examine the impact of ITB.
Figure 1.
Flowchart for evidence search and study selection. (a) Overall survival. (b) Cancer-specific survival. (c) Relative risk of disease.
Characteristics of the 25 topics
The 25 topics are shown in Supplementary Table 1. The median (IQR) number of studies included for a topic was 6 (4–10); 3 (1–6) studies with ITB and 3 (2–5) studies without ITB per topic. The most prevalent endpoint was overall survival (n = 15 topics, 60%), followed by cancer-specific survival (n = 5, 20%), risk of cancer (n = 2, 8%), risk of Parkinson’s disease (n = 1, 4%), risk of cardiovascular disease (CVD) (n = 1, 4%), and progression-free survival (n = 1, 4%) (Table 1). For most topics, exposure was a drug, except in two studies that investigated the effect of immune-related thyroid dysfunction on cancers (Table 1). The vast majority of the assessed conditions were cancers (22/25), except for three topics that focused on Parkinson's disease, cardiovascular disease, and all deaths related to allopurinol use (Table 1). As shown also in Supplementary Table 1, of the 25 topics, the overall results of a meta-analysis including all studies showed a statistically significant reduction in the outcome of interest in nine of the 25 topics and a point estimate in the direction of a favourable result for the exposure in 22 of the 25 topics. Supplementary Table 1 shows, for each topic, the summary ES and heterogeneity estimate, reported separately for studies with ITB and those without.
Table 1.
Characteristics of 25 topics (exposure-outcome pairs) and descriptive analysis.
| Estimates | |
|---|---|
| Number of topics (number of exposure-outcome subgroup analyses/number of all included observational studies) | 25 (46/182) |
| Number of studies included per topic, median (IQR) | 6 (4–10) |
| Exposure-outcome subgroup analyses | |
| Immortal time bias (ITB) | |
| Exposure-outcome subgroup with ITB | 21 (46) |
| Exposure-outcome subgroup without ITB | 25 (54) |
| Results of exposure-outcome subgroup analyses‡ | |
| Statistically significant results | 20 (43) |
| Statistically not significant results | 26 (57) |
| Topics | |
| Field of topics | |
| Oncology | 22 (88) |
| Others | 3 (12) |
| Endpoints | |
| Overall survival | 15 (60) |
| Cancer-specific survival | 5 (20) |
| Progression-free survival | 1 (4) |
| Relative risk | 4 (16) |
| Exposures | |
| Immune-related thyroid dysfunction | 2 (8) |
| Beta-blocker | 10 (40) |
| Metformin | 6 (24) |
| Inhaled corticosteroid | 1 (4) |
| Statin | 2 (8) |
| Aspirin | 1 (4) |
| Non-aspirin NSAID | 1 (4) |
| Allopurinol | 2 (8) |
| Outcomes | |
| All cancer | 2 (8) |
| Ovarian cancer | 5 (20) |
| Gastric cancer | 1 (4) |
| Lung cancer | 2 (8) |
| Head and neck cancer | 2 (8) |
| Pancreatic cancer | 1 (4) |
| Colorectal cancer | 2 (4) |
| Melanoma | 2 (8) |
| Breast cancer | 2 (8) |
| Prostate cancer | 3 (8) |
| Parkinson’s disease | 1 (4) |
| Cardiovascular disease (composite) | 1 (4) |
| Death associated with drug | 1 (4) |
Values are numbers (percentages) unless stated otherwise. IQR: interquartile range; NSAID: non-steroidal anti-inflammatory drugs.
Impact of ITB
The 25 topics contained a total of 182 studies, and 80 of them were suspected to involve ITB; the prevalence of ITB was 44.0%. Discrepancy in presence or not of statistical significance between the summary ES in studies with and without ITB was observed in 12/21 (57.1% of topics [topics 3–9, 11–13, 18 and 24]: in 11 topics, the ES was statistically significant in the studies with ITB but not in the studies without ITB, and the opposite was seen in only one topic). In six topics, both types of studies had non-statistically significant results, and in three topics, both had statistically significant results (Supplementary Table 2). A change of the presence of statistical significance after excluding studies with ITB, was observed in five topics (23.8%) [topics 3–5, 7 and 12]. In all of them, statistical significance was lost after exclusion of the studies with ITB, except for topic 3, which had turned from null to statistical significance after excluding studies with ITB (Figure 2). The ES for studies with and without ITB for 19 drug intervention topics are shown in Figure 2 (excluded are two topics associated with immune-related thyroid dysfunction (exposure) to focus on drug use).
Figure 2.
Subgroup analysis by immortal time bias (ITB) for the associations between drugs and overall survival (a), cancer-specific survival (b) and risk of disease (c). Since some studies did not provide subgroup ESs by ITB, we manually re-analysed and pooled effect estimates for ITB subgroups. HR < 1.00 indicates reduced mortality with the drug (favours drug) and RR < 1.00 indicates reduced risk of disease with the drug. A number of cohorts used to pool each result are noted in the right-most column. BB: beta-blocker; NA-NSAID: non-aspirin non-steroidal anti-inflammatory drugs; ICS: inhaled corticosteroids; CVD: cardiovascular diseases.
Figure 2.
Continued.
McNemar matched odds ratios suggested that studies with ITB had 11.0 (95% CI: 1.60–473) higher odds of having statistically significant ES than studies without ITB (Supplementary Table 2). The odds of summary ES having statistically significant results with higher ITB ratio was 1.06 (95% CI: 1.01–1.12, p value 0.03); and the odds of evidence reversal with higher ITB ratio was 1.12 (95% CI: 1.01–1.25, p value 0.03) (Supplementary Table 3). Figure 3 shows the ratio of ES and 95% CI per topic. Across 21 topics with both studies with and without ITB, the summary ratio of ES was 0.71 (95% CI: 0.66–0.78, I2, 55%), suggesting that the ES from studies with ITB was larger by an average of 29%, suggesting a more favourable drug effect.
Figure 3.
The ratio of ESs – ES pooled from studies with immortal time bias (ITB) relative to ES pooled from studies without ITB – for topics. Among 25 topics, only 21 could be divided by ITB (studies with ITB only and without ITB); four topics included only studies without ITB.
The mean (SD) heterogeneity (I2) for the results of all studies was 59.5% (25.0), and for studies with ITB, it was 28.3% (27.7), and for studies without ITB, it was 38.1% (35.9). The average reduction in heterogeneity (I2) by excluding studies with ITB across eligible topics (21 topics) was 21.4% (34.6). I2 decreased with removal of studies with ITB in 13 topics (52%) increased in seven topics (28%) and was unchanged in five topics (20%).
Assessment of systematic reviews and meta-analysis
The AMSTAR 2 evaluations of the 12 systematic reviews that contained the 25 eligible topics are provided in Supplementary Table 4. Three reviews were graded high, five were graded low and four were graded critically low. The most frequent factor downgrading the quality of review was lack of prespecified protocol established prior to the conduct of the review.
Discussion
This meta-epidemiological study found that across 25 topics (including 182 studies) where the systematic reviewers had been sensitised to the possibility of ITB, 44.0% of the eligible studies were affected by ITB. Among the 21 topics where both studies with ITB and studies without ITB were available, in more than half the ITB-affected and the non-affected studies differed in the presence or not of statistical significance. In these cases, almost always the ITB-affected studies were the ones that had statistically significant results. On average, studies with ITB generated substantially biased estimates of effect compared with studies without ITB, exaggerating ES by 29%, making it seem more favourable for the intervention. Excluding studies involving ITB reduced the heterogeneity (I2) of overall results, suggesting that ITB may be a noteworthy source of heterogeneity in meta-analysis.
Previous empirical examples have demonstrated that – within the same study – treating ITB with ITB robust methods can make a substantial difference.3,10,11 For example, Lévesque et al. demonstrated the effect of ITB by replicating the time-fixed (time-independent) analysis conducted by Yee et al. that explored the association between statin prescription and delay in insulin initiation.3,12 In the original time-fixed analysis, the immortal (statin unexposed time) periods accounted for approximately 68% of total follow-up time allocated to statin group, and yielded an adjusted HR of 0.74 (0.58–0.95). 3 By contrast, when Lévesque et al. corrected the misclassified immortal time using a simple time-varying (time-dependent) analysis, the adjusted HR changed to 1.97 (1.53–2.52). 3
ITB is not difficult to recognise and can be prevented. 3 The immortal time could be generally recognised by comparing time points of index date (start date of follow-up, most frequently at cohort entry) and first treatment/exposure during the follow-up. Correct handling of the immortal time could largely attenuate ITB, 13 and numerous statistical methods and designs were introduced to balance out the immortal time. 13 For example, the landmark method reduces ITB by classifying patients into exposed and unexposed groups at a single common, prespecified time point, so-called landmark, and this method is easily implemented with familiar statistical software.1,14 Time-matched analysis, nested case–control analysis, prescription time–distribution matching analysis, or regression analysis (including Cox proportional hazard model) with time-varying covariates may also be utilised to address ITB.1,14,15 These methods have been successful in controlling ITB. Meanwhile, indiscrete exclusion of immortal time could result in selection bias and should be dealt with considerable caution. 3
Authors of observational studies should provide adequate information for readers to assess the risk of ITB. 1 Vail et al. provided the ITB reporting checklist consisting of five criteria 1 ; authors should report (1) time of cohort entry for the exposure/treated and unexposed group, (2) eligible period in which the intervention may occur, (3) index time or start date of the follow-up in both groups (this could be same or different with the time of cohort entry), (4) statistical methods to address ITB and (5) the definition of treatment duration that divides patients into the exposure/treated or unexposed group. 1 Reporting treatment duration definition is particularly important when investigating a drug that is delivered over a time period (i.e. a seven-day regimen of antibiotics) because additional immortal time accumulates during the delivery of the medication. 1 When evaluating time-to-event outcomes (i.e. overall survival, cancer-specific survival, progression-free survival) in an observational study, researchers should be aware of possible ITB involvement, make the most of available statistical methods and adhere to higher reporting standards to counteract ITB.
Our analysis revealed that more than half of the statistically significant results of meta-analyses were no longer statistically significant, when studies with ITB were excluded. This may be due to the stronger ES estimates in studies with ITB but also the loss of power for the overall meta-analysis. Statistical significance alone should not be taken as equivalent to clinical significance. However, statistical significance is often read as a signal that mobilises an interpretation that the evaluated intervention works and should be considered for endorsement by guidelines, clinical practice and reimbursement agencies. In this sense, overlooking ITB could lead to clinical and health economic consequences such as redundant prescription of drugs, adverse drug reactions and increased medical expenditure. Researchers of observational studies should avoid or handle ITB and reviewers and editors should also be cognisant of this bias and take appropriate preventive or corrective actions.1,16 Observational study checklists 17 should include the necessary dedicated information to assess ITB. 1
Our study is subject to several limitations. First, our sample of 25 topics is not necessarily representative of all topics where studies perform time-to-event analyses for exposures. Our sample is derived from systematic reviews where the authors were cognisant of ITB and felt they should address it in their work. This may have selected topics where studies with ITB were more frequent and where the impact of the bias on the overall results was more prominent and thus noticeable. Therefore, we do not claim that our estimates of the prevalence of ITB or bias impact should be generalised to the entire literature of time-to-event analyses. However, our estimates suggest that ITB can be a frequent and serious bias. Second, the judgement for each observational study on whether ITB was involved was based on the discretion of the authors of the systematic reviews. Therefore, there may be some degree of heterogeneity caused by multiple assessors and slightly varying definitions being considered about what constitutes ITB. However, since inspecting the presence of immortal time is neither very complicated nor subjective, the classification of studies in those with and without ITB is unlikely to be uncertain. Third, we used the data already extracted by the systematic reviewers, which may have occasional errors in data extraction or arbitration. However, depending on already extracted data diminished the possibility of adding bias from our own data extractions. Fourth, because we used search terms requiring explicit mention of ITB in the abstract, there is a chance that some systematic reviews covering ITB only in the main text were missed. While comprehensively reviewing every drug-related SR for ITB analyses would be ideal, the sheer volume of literature makes such an undertaking impractical. Thus, we relied on an abstract-based search strategy, which offers a pragmatic solution that captures most relevant studies without significantly diminishing completeness. Our approach strategically balanced methodological rigor with operational reality. Finally, the clinical importance of the differences in results of studies with and without ITB can be subjective and may vary from one topic to another. Nevertheless, the observed differences were large in both statistical significance and magnitude of ES; thus, they are likely to be perceived as important for clinical and other interpretation purposes.
Conclusions
Our study revealed a substantial footprint of ITB on several meta-analyses of diverse topics of drug and other exposures in observational studies with time-to-event outcomes. Adherence to ITB robust statistical methods and improved, dedicated reporting standards by researchers, as well as collective surveillance from readers, reviewers and editors may help diminish the impact of this bias.
Supplemental Material
Supplemental material, sj-pdf-1-jrs-10.1177_01410768251366880 for Quantifying the impact of immortal time bias: empirical evidence from meta-analyses by Min Seo Kim, Dong Keon Yon, Seung Won Lee, Masoud Rahmati, Marco Solmi, Andre F Carvalho, Ai Koyanagi, Lee Smith, Jae Il Shin and John PA Ioannidis in Journal of the Royal Society of Medicine
Supplemental material, sj-pdf-2-jrs-10.1177_01410768251366880 for Quantifying the impact of immortal time bias: empirical evidence from meta-analyses by Min Seo Kim, Dong Keon Yon, Seung Won Lee, Masoud Rahmati, Marco Solmi, Andre F Carvalho, Ai Koyanagi, Lee Smith, Jae Il Shin and John PA Ioannidis in Journal of the Royal Society of Medicine
Footnotes
ORCID iDs: Masoud Rahmati https://orcid.org/0000-0003-4792-027X
Jae Il Shin https://orcid.org/0000-0003-2326-1820
Declarations
Competing Interests
This work was supported by the Yonsei Fellowship by Lee Youn Jae.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Funded by Lee Youn Jae (JIS).
Ethics approval
Not applicable. Publicly available data were involved in the study.
Guarantor
JIS and JPAI.
Contributorship
The study was conceived, and the scope designed by MSK, DKY, JIS, and JPAI. The design of the search strategy and database searches were conducted by MSK and DKY. The quality of the studies was assessed by MSK and DKY. Data extraction and statistical analysis were conducted by MSK and DKY. The first draft of the article was prepared by MSK with supervision from DKY, JIS and JPAI. Expert input on the intellectual content and results was provided by SWL, MR, MS, AFC, AK, LS, JIS, and JPAI.
Provenance
Not commissioned; peer-reviewed by Julie Morris.
Supplemental material
Supplemental material for this article is available online.
References
- 1.Vail EA, Gershengorn HB, Wunsch H, Walkey AJ. Attention to immortal time bias in critical care research. Am J Respir Crit Care Med 2021; 203: 1222–1229. [DOI] [PubMed] [Google Scholar]
- 2.Vail EA, Wunsch H, Pinto R, Bosch NA, Walkey AJ, Lindenauer PK, et al. Use of hydrocortisone, ascorbic acid, and thiamine in adults with septic shock. Am J Respir Crit Care Med 2020; 202: 1531–1539. [DOI] [PubMed] [Google Scholar]
- 3.Lévesque LE, Hanley JA, Kezouh A, Suissa S. Problem of immortal time bias in cohort studies: example using statins for preventing progression of diabetes. BMJ (Clinical Res ed) 2010; 340: b5087. [DOI] [PubMed] [Google Scholar]
- 4.Yadav K, Lewis RJ. Immortal time bias in observational studies. JAMA 2021; 325: 686–687. [DOI] [PubMed] [Google Scholar]
- 5.Suissa S. Immortal time bias in pharmaco-epidemiology. Am J Epidemiol 2008; 167: 492–499. [DOI] [PubMed] [Google Scholar]
- 6.Suissa S. Immortal time bias in observational studies of drug effects. Pharmacoepidemiol Drug Saf 2007; 16: 241–249. [DOI] [PubMed] [Google Scholar]
- 7.Shea BJ, Reeves BC, Wells G, Thuku M, Hamel C, Moran J, et al. AMSTAR 2: a critical appraisal tool for systematic reviews that include randomised or non-randomised studies of healthcare interventions, or both. BMJ (Clin Res ed) 2017; 358: j4008. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Hróbjartsson A, Thomsen ASS, Emanuelsson F, Tendal B, Hilden J, Boutron I, et al. Observer bias in randomised clinical trials with binary outcomes: systematic review of trials with both blinded and non-blinded outcome assessors. BMJ (Clin Res ed) 2012; 344: e1119. [DOI] [PubMed] [Google Scholar]
- 9.Savović J, Jones HE, Altman DG, Harris RJ, Jüni P, Pildal J, et al. Influence of reported study design characteristics on intervention effect estimates from randomized, controlled trials. Ann Intern Med 2012; 157: 429–438. [DOI] [PubMed] [Google Scholar]
- 10.Wei M, Liu Y, Bi Y, Zhang Z-J. Metformin and pancreatic cancer survival: real effect or immortal time bias? Int J Cancer 2019; 145: 1822–1828. [DOI] [PubMed] [Google Scholar]
- 11.Street S, Chute D, Strohbehn I, Rengarajan M, Faje A, Seethapathy H, et al. The positive effect of immune checkpoint inhibitor-induced thyroiditis on overall survival accounting for immortal time bias: a retrospective cohort study of 6596 patients. Ann Oncol 2021; 32: 1050. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Yee A, Majumdar S, Simpson S, McAlister FA, Tsuyuki RT, Johnson JA. Statin use in type 2 diabetes mellitus is associated with a delay in starting insulin. Diab Med 2004; 21: 962–967. [DOI] [PubMed] [Google Scholar]
- 13.Karim ME, Gustafson P, Petkau J, Tremlett H. Comparison of statistical approaches for dealing with immortal time bias in drug effectiveness studies. Am J Epidemiol 2016; 184: 325–335. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Gleiss A, Oberbauer R, Heinze G. An unjustified benefit: immortal time bias in the analysis of time-dependent events. Transpl Int 2018; 31: 125–130. [DOI] [PubMed] [Google Scholar]
- 15.Zhou Z, Rahme E, Abrahamowicz M, Pilote L. Survival bias associated with time-to-treatment initiation in drug effectiveness evaluation: a comparison of methods. Am J Epidemiol 2005; 162: 1016–1023. [DOI] [PubMed] [Google Scholar]
- 16.Targownik LE, Suissa S. Understanding and avoiding immortal-time bias in gastrointestinal observational research. Am J Gastroenterol 2015; 110: 1647–1650. [DOI] [PubMed] [Google Scholar]
- 17.von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP; STROBE Initiative. The strengthening the reporting of observational studies in epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet 2007; 370: 1453–1457. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplemental material, sj-pdf-1-jrs-10.1177_01410768251366880 for Quantifying the impact of immortal time bias: empirical evidence from meta-analyses by Min Seo Kim, Dong Keon Yon, Seung Won Lee, Masoud Rahmati, Marco Solmi, Andre F Carvalho, Ai Koyanagi, Lee Smith, Jae Il Shin and John PA Ioannidis in Journal of the Royal Society of Medicine
Supplemental material, sj-pdf-2-jrs-10.1177_01410768251366880 for Quantifying the impact of immortal time bias: empirical evidence from meta-analyses by Min Seo Kim, Dong Keon Yon, Seung Won Lee, Masoud Rahmati, Marco Solmi, Andre F Carvalho, Ai Koyanagi, Lee Smith, Jae Il Shin and John PA Ioannidis in Journal of the Royal Society of Medicine