Quantifying the impact of immortal time bias: empirical evidence from meta-analyses
Bibliographic record
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 ( I 2 ), 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 ( I 2 ) 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.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Direct model labels (unvalidated)
Per-model category and study-design labels from the labeling rounds. They are machine output, unvalidated, and the disagreement between models ships as data. No study design here is MEDLINE-validated yet.
| Model arm | Categories | Study design | Confidence |
|---|---|---|---|
| gemma | MetaresearchMeta-epidemiology (broad) Domain: Methods · Genre: Empirical About the Canadian research system: no · About a Canadian topic: no | Meta-analysis | low |
| gpt | MetaresearchMeta-epidemiology (narrow)Meta-epidemiology (broad) Domain: Methods · Genre: Empirical About the Canadian research system: no · About a Canadian topic: no | Meta-analysis | high |
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.341 | 0.656 |
| Meta-epidemiology (narrow) | 0.006 | 0.003 |
| Meta-epidemiology (broad) | 0.025 | 0.063 |
| Bibliometrics | 0.020 | 0.015 |
| Science and technology studies | 0.001 | 0.005 |
| Scholarly communication | 0.010 | 0.010 |
| Open science | 0.006 | 0.007 |
| Research integrity | 0.006 | 0.006 |
| Insufficient payload (model declined to judge) | 0.006 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedLabeled directly by 2 models reading the full record.
The models disagree on parts of this classification; every voice is preserved in the section at the end of the page.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".