The DOI matches its Crossref record; 38 of the 38 references it cites were found in a registry or an index; 13 of the 21 sources its claims cite were checked for support.
Validity of data extraction in evidence synthesis practice of adverse events: reproducibility study
Open accessAbstract
Abstract Objectives To investigate the validity of data extraction in systematic reviews of adverse events, the effect of data extraction errors on the results, and to develop a classification framework for data extraction errors to support further methodological research. Design Reproducibility study. Data sources PubMed was searched for eligible systematic reviews published between 1 January 2015 and 1 January 2020. Metadata from the randomised controlled trials were extracted from the systematic reviews by four authors. The original data sources (eg, full text and ClinicalTrials.gov) were then referred to by the same authors to reproduce the data used in these meta-analyses. Eligibility criteria for selecting studies Systematic reviews were included when based on randomised controlled trials for healthcare interventions that reported safety as the exclusive outcome, with at least one pair meta-analysis that included five or more randomised controlled trials and with a 2×2 table of data for event counts and sample sizes in intervention and control arms available for each trial in the meta-analysis. Main outcome measures The primary outcome was data extraction errors summarised at three levels: study level, meta-analysis level, and systematic review level. The potential effect of such errors on the results was further investigated. Results 201 systematic reviews and 829 pairwise meta-analyses involving 10 386 randomised controlled trials were included. Data extraction could not be reproduced in 1762 (17.0%) of 10 386 trials. In 554 (66.8%) of 829 meta-analyses, at least one randomised controlled trial had data extraction errors; 171 (85.1%) of 201 systematic reviews had at least one meta-analysis with data extraction errors. The most common types of data extraction errors were numerical errors (49.2%, 867/1762) and ambiguous errors (29.9%, 526/1762), mainly caused by ambiguous definitions of the outcomes. These categories were followed by three others: zero assumption errors, misidentification, and mismatching errors. The impact of these errors were analysed on 288 meta-analyses. Data extraction errors led to 10 (3.5%) of 288 meta-analyses changing the direction of the effect and 19 (6.6%) of 288 meta-analyses changing the significance of the P value. Meta-analyses that had two or more different types of errors were more susceptible to these changes than those with only one type of error (for moderate changes, 11 (28.2%) of 39 v 26 (10.4%) 249, P=0.002; for large changes, 5 (12.8%) of 39 v 8 (3.2%) of 249, P=0.01). Conclusion Systematic reviews of adverse events potentially have serious issues in terms of the reproducibility of the data extraction, and these errors can mislead the conclusions. Implementation guidelines are urgently required to help authors of future systematic reviews improve the validity of data extraction.
Overview
CoChat Summary
TL;DR
In a reproducibility study of 201 systematic reviews of adverse events, data extraction errors were common: 85.1% of reviews had at least one meta-analysis with an error, and 66.8% of meta-analyses had at least one trial with an error. These errors were classified into five mechanisms and sometimes changed effect magnitude, direction, or statistical significance.
Key findings
- Among 201 systematic reviews containing 829 pairwise meta-analyses and 10,386 randomised controlled trials, 1762 (17.0%) trials could not be reproduced at the study level (extractor-adjusted 15.8%); 554 (66.8%) meta-analyses had at least one trial with a data extraction error (adjusted 65.5%); and 171 (85.1%) systematic reviews had at least one meta-analysis with an error (adjusted 85.1%).
- Errors were classified into five types: numerical error 867/1762 (49.2%), ambiguous error 526 (29.9%), zero assumption error 221 (12.5%), misidentification error 115 (6.5%), and mismatching error 33 (1.9%); numerical, ambiguous, and zero assumption errors accounted for 91% of total errors.
- Drug interventions had higher error proportions than non-drug interventions: total error 19.9% vs 8.9% (P<0.001), ambiguous 6.1% vs 2.4% (P<0.001), numerical 9.4% vs 5.4% (P<0.001), zero assumption 2.6% vs 0.9% (P<0.001), and misidentification 1.5% vs 0.1% (P<0.001).
- In 288 meta-analyses used to assess impact, 37 (12.8%) had moderate changes (≥20%) and 13 (4.5%) had large changes (≥50%) in effect estimates; 10 (3.5%) changed direction and 19 (6.6%) changed P-value significance. Meta-analyses with two or more error types had higher proportions of moderate (28.2% vs 10.4%, P=0.002) and large changes (12.8% vs 3.2%, P=0.01) than those with a single error type.
Methods
- This was a reproducibility study extending prior work: PubMed was searched for systematic reviews of adverse events indexed from 1 January 2015 to 1 January 2020. Eligible reviews included randomised controlled trials with adverse events as the exclusive outcome, at least one pairwise meta-analysis with five or more studies, and 2×2 event count/sample size data; proportions, network meta-analyses, continuous safety outcomes, and non-English/Chinese reviews were excluded. Screening 18,636 records yielded 201 included systematic reviews.
- Four experienced authors extracted data in four equal portions after pilot training, with double-checking. They then reproduced the data from original trial publications and supplementary files, ClinicalTrials.gov, and pharmaceutical company websites. Discrepancies were classified as data extraction errors unless review authors reported successfully contacting original authors for data. Errors were empirically classified into five types by consensus.
- Proportions of errors were summarised at study, meta-analysis, and systematic review levels; a generalised linear mixed model estimated extractor-adjusted proportions; multivariable logistic regression examined associations with duplicated extraction and protocol development. Corrected datasets were used to repeat meta-analyses with the same methods and effect estimators, with prespecified 20% and 50% thresholds for moderate and large impact.
Limitations
- The paper states that data collection was divided into four portions, so judgments might differ between extractors, although the generalised linear mixed model accounted for clustering and suggested no obvious impact. It also states that only reviews published from 2015 to 2020 were covered, so validity in earlier studies is unclear.
- The paper states that a small proportion of studies could not have reproducibility checked and were treated as having no errors, possibly slightly underestimating error overall; it also states that only randomised controlled trials were included, not observational studies.
- The paper states that poor reporting limited investigation of associations between good practice recommendations and data extraction errors, so associations with duplicated extraction and advance protocol should be interpreted cautiously. Evident from the text: the search was restricted to PubMed and to English/Chinese reviews, and inclusion required at least five studies and 2×2 data, so findings may not generalise to other systematic reviews or meta-analysis formats; impact analyses also excluded meta-analyses with ambiguous errors.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures





Verification
Claims the paper makes
8 Valid · 1 Not valid · 5 Not checked · 65 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“Despite this quality assurance mechanism, data extraction error in systematic reviews occurs frequently in the literature. 3”
What the sources say
- Supportedcites [3] source
The Discussion states that extraction errors have a high prevalence, and the Conclusion repeats this finding. The Results also report error rates of 8% to 42% in one included comparison of systematic reviews. These passages support the claim that data extraction errors occur frequently in systematic reviews. (LLM, conf=0.94)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Another important reason for data extraction errors might be the poor reporting of adverse events in randomised controlled trials, which have varying terminology, poorly defined categories, and diverse data sources. 30 31 32”
What the sources say
- Not checked — no full text of the sourcecites [30] source
- Partly supportedcites [31] sourceExperimental — accuracy not yet measured
The Results (CONSORT harms recommendations; page unavailable) find that adverse events were inadequately defined and that their collection, analysis, and coding were often poorly described. The paper does not establish that these reporting problems cause data-extraction errors, or specifically document varying terminology and diverse data sources. (LLM, conf=0.84)
- Not checked — no full text of the sourcecites [32] source
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For the reproducibility of the data used in these meta-analyses, at the study level, we could not reproduce 1762 (17.0%) of 10 386 studies with an extractor addressed proportion of 15.8%.”
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“No systematic reviews reported the information of whether the data extractor was trained or the expertise of the data extractor.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The second most prevalent data extraction error was the ambiguous error, accounting for 526 (29.9%) errors.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
22 more claims
Create a free account to see all 79 claims and every source verdict.
- It may state two separate claims as one. 35
- Probably not a substantial claim. 28
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“However, we addressed this by performing randomized and blinded adjudication, adopting best practices seen in radiology, ophthalmology, and clinical trials 28-31 .”
From Automation of Systematic Reviews with Large Language Models (2025) · cites this paper as [30]
The check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“This adjudication framework was adapted from established protocols in radiology, ophthalmology, and clinical trials 28-31 .”
From Automation of Systematic Reviews with Large Language Models (2025) · cites this paper as [30]
Experimental — accuracy not yet measuredIn the citing paper
“human data extraction is also not error-free; several studies have reported high error rates, often involving misidentification or placing correct data in the wrong fields [29–31].”
From Automated Mass Extraction of Over 680,000 PICOs from Clinical Study Abstracts Using Generative AI: A Proof-of-Concept Study. (2024) · cites this paper as [30]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
“Sixty six (32.8%) systematic reviews used a standard data extraction form, while most (135, 67.2%) did not report this information. For the systematic reviews that used a standard extraction form, six (8.8%) piloted this process.”
Reported 8.8 · recomputed 9.090909
Citations
Cites
Adverse Renal Effects of Immune Checkpoint Inhibitors: A Narrative Review
Wanchoo, R., Karam, S., Uppal, N. N., Barta, V. S., Deray, G., Devoe, C., Launay-Vacher, V., Jhaveri, K. D., & on behalf of Cancer and Kidney International Network Workgroup on Immune Checkpoint Inhibitors (2017) · American Journal of Nephrology · doi:10.1159/000455014 · Cited by 15,256
Nivolumab plus Ipilimumab versus Sunitinib in Advanced Renal-Cell Carcinoma
Motzer, R. J., Tannir, N. M., McDermott, D. F., Arén Frontera, O., Melichar, B., Choueiri, T. K., Plimack, E. R., Barthélémy, P., Porta, C., George, S., et al. (2018) · New England Journal of Medicine · doi:10.1056/nejmoa1712126 · Cited by 4,656
1,500 scientists lift the lid on reproducibility
Baker, M. (2016) · Nature · doi:10.1038/533452a · Cited by 4,209
Combined BRAF and MEK inhibition in melanoma with BRAF V600 mutations
Flaherty, K. T., Infante, J. R., Daud, A., Gonzalez, R., Kefford, R. F., Sosman, J., Hamid, O., Schuchter, L., Cebon, J., Ibrahim, N., et al. (2012) · New England Journal of Medicine · doi:10.1056/nejmoa1210093 · Cited by 2,810
Nivolumab and ipilimumab versus ipilimumab in untreated melanoma
Postow, M. A., Chesney, J., Pavlick, A. C., Robert, C., Grossmann, K., McDermott, D., Linette, G. P., Meyer, N., Giguere, J. K., Agarwala, S. S., et al. (2015) · New England Journal of Medicine · doi:10.1056/nejmoa1414428 · Cited by 2,781
Hodi, F. S., Chesney, J., Pavlick, A. C., Robert, C., Grossmann, K. F., McDermott, D. F., Linette, G. P., Meyer, N., Giguere, J. K., Agarwala, S. S., et al. (2016) · The Lancet Oncology · doi:10.1016/s1470-2045(16)30366-7 · Cited by 968
Power analysis for random-effects meta-analysis
Jackson, D., & Turner, R. (2017) · Research Synthesis Methods · doi:10.1002/jrsm.1240 · Cited by 619
PRISMA harms checklist: improving harms reporting in systematic reviews
Zorzela, L., Loke, Y. K., Ioannidis, J. P., Golder, S., Santaguida, P., Altman, D. G., Moher, D., Vohra, S., & PRISMA harms group (2016) · BMJ · doi:10.1136/bmj.i157 · Cited by 557
Self-citationBlackwell, K. L., Burstein, H. J., Storniolo, A. M., Rugo, H. S., Sledge, G., Aktan, G., Ellis, C., Florance, A., Vukelja, S., Bischoff, J., et al. (2012) · Journal of Clinical Oncology · doi:10.1200/jco.2011.35.6725 · Cited by 547
Austin, P. C., & Steyerberg, E. W. (2017) · Statistical Methods in Medical Research · doi:10.1177/0962280214558972 · Cited by 516
Cited by
Gallardo-Gómez, D., del Pozo-Cruz, J., Pedder, H., Alfonso-Rosa, R. M., Álvarez-Barbosa, F., Noetel, M., Jasper, U., Chastin, S., Ramos-Munell, J., & del Pozo Cruz, B. (2023) · British Journal of Sports Medicine · doi:10.1136/bjsports-2022-106409 · Cited by 59
Tóth, B., Berek, L., Gulàcsi, L., Péntek, M., & Zrubka, Z. (2024) · Systematic Reviews · doi:10.1186/s13643-024-02592-3 · Cited by 70
“SR automation using artificial intelligence (AI) and advanced computing technologies has the potential to speed up the review process, reduce the workload of researchers, prevent human errors, and facilitate reproducibility by diminishing the role of human judgement [18–20].”
MethodXu, C., Fan, S., Tian, Y., Liu, F., Furuya-Kanamori, L., Clark, J., Zhang, C., Li, S., Lin, L., Chu, H., et al. (2025) · BMJ · doi:10.1136/bmj-2024-082068 · Cited by 39
Self-citationZhang, Y., Stokes, J., Anselmi, L., Bower, P., & Xu, J. (2025) · Health Research Policy and Systems · doi:10.1186/s12961-024-01260-1 · Cited by 90
Assessing the robustness of results from clinical trials and meta-analyses with the fragility index
Lin, L., Xing, A., Chu, H., Murad, M. H., Xu, C., Baer, B. R., Wells, M. T., & Sanchez-Ramos, L. (2023) · American Journal of Obstetrics and Gynecology · doi:10.1016/j.ajog.2022.08.053 · Cited by 35
Self-citation
Metadata
- Authors
- Xu, C., Yu, T., Furuya-Kanamori, L., Lin, L., Zorzela, L., Zhou, X., Dai, H., Loke, Y., & Vohra, S.
- Affiliations
- Ministry of Education of the People's Republic of China (CN)
- Sichuan University (CN)
- UQ Centre for Clinical Research, Faculty of Medicine, University of Queensland, Brisbane, QLD, Australia
- Florida State University (US)
- University of Alberta (CA)
- West China Hospital of Sichuan University (CN)
- University of East Anglia (GB)
- Corresponding author
- Vohra, Sunita
- Venue
- BMJ
- Publisher
- BMJ
- Published
- 2022-05-10
- Volume
- 377
- Pages
- e069155
- ISSN
- 1756-1833 (ISSN-L)
- PMC
- PMC9086856
- OpenAlex
- W4280532161
- Semantic Scholar
- 527adcb389a36f796734e21260c9d9aeb9a1b1ef
- Cited by
- 82
- References
- 38
- Citations by year
- 2026: 28 · 2025: 16 · 2024: 20 · 2023: 13 · 2022: 5
- FWCI
- 8.13
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- BMJ (publishedVersion, cc-by-nc)
- PubMed Central (publishedVersion, cc-by-nc)
- PubMed (publishedVersion)
- UEA Digital Repository (University of East Anglia) (acceptedVersion)
- Funders
- National Institutes of Health
- Medical Research Council
- National Health and Medical Research Council (1158469)
- National Institute of Mental Health (MH128727, R03MH128727)
- U.S. National Library of Medicine (R01LM012982, LM012982)
- Topics
- Social Sciences › Decision Sciences › Statistics, Probability and Uncertainty › Meta-analysis and systematic reviews
- Social Sciences › Sociology and Political Science › Delphi Technique in Research
- Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods in Clinical Trials
- Keywords
- data extraction errors, adverse events, safety outcomes, systematic reviews, reproducibility, meta-analysis, randomized controlled trials, classification framework, numerical error, ClinicalTrials.gov
- MeSH terms
- Systematic Reviews as Topic, Humans, Meta-Analysis as Topic, Reproducibility of Results, Randomized Controlled Trials as Topic, Data Mining
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-09-29, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-09-29 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Validity of data extraction in evidence synthesis practice of adverse events: reproducibility study. https://app.cochat.ai/papers/10.1136/bmj-2021-069155
- The paper
- Xu, C., Yu, T., Furuya-Kanamori, L., Lin, L., Zorzela, L., Zhou, X., Dai, H., Loke, Y., & Vohra, S. (2022). Validity of data extraction in evidence synthesis practice of adverse events: reproducibility study. BMJ. https://doi.org/10.1136/bmj-2021-069155
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.