Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one
Contents

Academy · Research Data Analytics & AI · Extracting quantitative data

Extracting Quantitative Data for Systematic Review and Meta-Analysis: Preserving the Denominator

Cochrane asks review authors to collect all five elements of an outcome, yet most extraction forms in circulation capture three. The missing elements are usually the denominator and the measure of variability, precisely the two that cannot be reconstructed later. This article addresses what to extract, why two people must extract independently, how numerical data can be obtained from figures, and what automation can and cannot achieve.

Pubrica AcademyWritten by the medical writing and biostatistics teams Reviewed byPubrica biostatistics team Published9 October 2026 Last checked9 October 2026 Reading time12 minutes Sources36 peer-reviewed

Data extraction is the step at which a review’s numbers are determined, yet it is the step most often delegated to whichever team member is available. The error rate is not hypothetical: Gøtzsche and colleagues documented extraction errors in meta-analyses using standardised mean differences in JAMA, and Buscemi and colleagues showed that single extraction generates more errors than double extraction. The consequential errors are not typographical errors in effect sizes but missing denominators, incorrect variability statistics, and outcomes selectively chosen from among several reported.

Learning objectives

  • The five elements of an outcome that Cochrane expects review authors to collect, and the consequences when one is missing
  • Why double independent extraction is a methods requirement rather than a counsel of perfection
  • How many extractors a review requires, according to the Handbook’s scaling table
  • How to obtain usable numerical data from a figure, and when software outperforms reading values from the axis
  • Why a single publication is often an insufficient source for a single study
  • What automated extraction tools currently achieve, and the large proportion of fields they do not address

Common misconceptions

  • “Extract the effect size and the p-value and you have what you need.” An effect size without its denominator and variability measure cannot be weighted, and inverse-variance weighting is what a meta-analysis runs on.
  • “Double extraction is a nice-to-have if you have the people.” Cochrane’s Methodological Expectations of Cochrane Intervention Reviews (MECIR) standard for data collection is at least two people working independently. Buscemi and colleagues, in the Journal of Clinical Epidemiology, found that single extraction generated more errors than double extraction.
  • “Extraction errors get caught in peer review.” Jones and colleagues found a high prevalence of data extraction and reporting errors in Cochrane reviews — reviews that had already been through methodological editing.
  • “Reading values off a published graph is close enough.” Jelicic Kadic and colleagues found extracting data from figures with software was faster and had higher interrater reliability than manual extraction.
  • “The published paper is the study.” Mayo-Wilson and colleagues’ MUDS work shows that trial registries, regulatory documents and protocols routinely contain outcomes and analyses the publication does not.

The five elements of an outcome

The Cochrane Handbook’s chapter on collecting data sets the requirement simply: review authors should collect all five elements of an outcome 1. In practice, this means that for every outcome in every arm of every study the reviewer requires the number of participants on which the result is based, the number of events or the summary statistic, the measure of variability or precision, and sufficient information to identify exactly which outcome construct and which time point is being recorded. If any one is omitted, the result may be uninterpretable or unusable.

The denominator is the element most often lost, and it is lost in a characteristic way. A paper reports that 18% of the intervention arm experienced the event, and the extractor records 18%. Months later the analyst needs the number of participants contributing to that percentage, and the paper’s baseline table gives a randomised total that does not match the analysed total because of withdrawals. The cell counts can then no longer be reconstructed, and the study either enters the model with an estimated denominator or is excluded from it. In either case, the decision has been made by an extraction form rather than by a reviewer.

The variability element fails differently. Standard deviation, standard error, confidence interval and interquartile range are not interchangeable, and a form with a single field labelled “SD” encourages the extractor to enter whatever statistic the paper reported 23. The statistic actually reported should be recorded, in its own field, on every occasion.

Additional data to be extracted from each report

Outcome data are necessary but not sufficient. Every variable that will later be needed to assign a study to a synthesis group, to run a pre-specified subgroup analysis, or to appraise risk of bias must be on the form from the outset, because adding a field afterwards requires re-reading every included paper.

Intervention description is the field most commonly under-specified. Hoffmann and colleagues’ TIDieR checklist in the BMJ exists because published intervention descriptions are too sparse to permit replication, and the same gaps make it impossible to determine whether two studies delivered the same intervention 12. Extracting against the TIDieR (Template for Intervention Description and Replication) items (what, who provided, how, where, how much, when, tailoring and modifications) raises the question early rather than at the synthesis-grouping stage.

In addition, the form should capture design and unit of randomisation, eligibility criteria as stated, the analysis population for each result, how missing data were handled, funding and conflicts, and the risk-of-bias judgements with the supporting quotation for each. Wood and colleagues’ work in the BMJ on how design features relate to estimated effects explains why the bias fields cannot be treated as secondary 21. Our page on what data to extract covers the full field list.

Independent extraction by two reviewers

Cochrane’s MECIR standard for data collection is to use at least two people working independently 1,2. Independence is the essential requirement: two people completing the same form in sequence, with one checking the other’s entries, is a different and weaker procedure than two people extracting blind and then reconciling. The latter detects a misread value; the former detects only a value that the checker happens to doubt.

Buscemi and colleagues, in the Journal of Clinical Epidemiology, found that single data extraction generated more errors than double data extraction; the finding is stated in the title of the paper and has not been seriously contested since 25. Jones and colleagues, in the same journal, examined Cochrane reviews and found a high prevalence of data extraction and reporting errors, though with low impact on conclusions in their sample 26. The low-impact finding is sometimes interpreted as justification for single extraction. It is more appropriately attributed to chance: an error’s impact depends on the study in which it occurs, and under inverse-variance weighting an error in a heavily weighted study moves the pooled estimate directly.

The Handbook also scales the team to the review. Table 5.4.a suggests a small team of 2 to 3 data extractors for small-scale reviews of fewer than 10 included studies, 4 to 6 for small-to-medium reviews of 10 to 20 studies, and more than 6 extractors for larger reviews 1,36. For a single author planning a review of thirty studies, that table is particularly relevant.

Number of extractors required by review size

Review scaleIncluded studiesData extractors
Small-scaleFewer than 10A small team of 2 to 3
Small to medium10 to 204 to 6
LargerMore than 20More than 6

Cochrane Handbook for Systematic Reviews of Interventions, version 6.5, Table 5.4.a. The MECIR standard for data collection, in the same chapter, is at least two people working independently regardless of review size.

5elements of an outcome that Cochrane expects review authors to collect
2minimum number of people extracting independently (MECIR)
1–7data elements each automated tool was able to extract

Outcome elements and the MECIR minimum: Cochrane Handbook for Systematic Reviews of Interventions, version 6.5, Chapter 5 (Collecting data). Automated extraction range: Jonnalagadda SR, Goyal P, Huffman MD. Automating data extraction in systematic reviews: a systematic review. Systematic Reviews 2015;4:78 — which also reports that well over half of the data elements needed for systematic reviews have not been explored.

Reconciliation as the core of the method

Double extraction is effective only if disagreements are resolved by a stated procedure rather than by the more confident extractor. The procedure should be set before extraction begins: who arbitrates, what happens when both extractors make the same error, and how a reconciliation decision is recorded so that the final dataset can be traced back to a source page.

The disagreement rate should be recorded. It is a quality indicator for the form and the training, not only for the extractors; a field with a high disagreement rate is usually ambiguously worded. The form should also be piloted on three or four studies expected to be difficult, rather than on the clearest ones, because the form must accommodate the difficult cases. Where disagreements are frequent and substantive, the Academy’s page on handling data discrepancies covers the escalation options, including contacting authors.

The completed extraction dataset should be deposited. Page and colleagues, in the Journal of Clinical Epidemiology, found reproducible research practices to be underused in systematic reviews of biomedical interventions, and the extraction sheet is the most valuable single artefact to publish: it allows a reader to verify a number rather than accept it on trust 9. Ip and colleagues’ web-based archive of systematic review data in Systematic Reviews was an early demonstration of what becomes possible once extraction data are shared 6.

Extracting numerical data from figures

A substantial share of the data a reviewer needs exists only in a Kaplan-Meier curve, a scatter plot or a bar chart with error bars. Reading values from the axis by eye is the default and the least reliable option. Jelicic Kadic and colleagues, in the Journal of Clinical Epidemiology, compared the approaches directly and found that extracting data from figures with software was faster and had higher interrater reliability than manual extraction, an uncommon case in which the more rigorous option is also the quicker one 11. Gross, Schirm and Scholz describe one such tool, Ycasd, in BMC Bioinformatics; the general approach is to calibrate the image against two known axis points and digitise from there 14.

Three practices make figure extraction defensible. Both extractors should digitise independently, as with any other field, and compare results. The calibration points and the software used should be recorded, so that the procedure is reproducible. The form should also record that the value came from a figure rather than from text, because that provenance affects how far the value should be trusted and whether it belongs in a sensitivity analysis.

Where the figure cannot provide what the model requires, most often a variability statistic, the appropriate step is to contact the authors before deciding to impute. An imputed value with a stated method is defensible; an imputed value presented as extracted is not.

Multiple documents per study

Mayo-Wilson, Li, Fusco and Dickersin’s practical guidance in Research Synthesis Methods, drawing on the MUDS study, is the clearest statement of a problem that most reviewers encounter and few address systematically: a single trial may be described in a journal article, a conference abstract, a registry record, a protocol, a statistical analysis plan and a regulatory submission, and these sources disagree 8. The choice of source changes the result.

Two related bodies of literature explain why this is a substantive rather than an administrative problem. Chan and colleagues, in JAMA, compared protocols with publications and documented discrepancies in reported outcomes 27. Turner and colleagues, in the New England Journal of Medicine, compared the published literature on antidepressant trials with the regulatory record. Page and colleagues’ Cochrane methodology review on bias due to selective inclusion and reporting of outcomes in systematic reviews shows the problem recurring at the level of the review, and Mayo-Wilson and colleagues demonstrated in the Journal of Clinical Epidemiology that cherry-picking by trialists and meta-analysts can drive conclusions about intervention efficacy 4,10.

The appropriate safeguard is procedural: the protocol should state which source takes precedence when sources conflict, every source consulted should be recorded for each study, and discrepancies should be recorded rather than silently resolved. A review that reports which outcome it chose, from which document, and why, has addressed the objection before it is raised.

Components of a complete extraction record

The difference between an effective form and one that loses data is not its length. It is whether each of the five outcome elements has its own field, whether the variability statistic is named rather than assumed, and whether provenance is recorded. The two versions below extract the same result from the same paper.

The same outcome, extracted two ways

Loses the denominator

Study: Author 20XX. Outcome: infection rate. Intervention group: 18%. Control group: 27%. Result: significant. Notes: see Figure 2.

No denominator. 18% of how many participants? Without the analysed n the study cannot be weighted, and under inverse-variance weighting it therefore cannot enter the model.

No variability statistic, and no label for one. “Significant” is a verdict, not a precision estimate.

Outcome construct and time point unstated. Two studies recorded this way cannot be checked for whether they measured the same thing.

No provenance and no second extractor. “See Figure 2” indicates that a value was read from an axis by one person, without a record of the procedure.

No analysis population. Randomised and analysed totals routinely differ, and the difference is itself a risk-of-bias signal.

Captures all five elements

Study ID: Author 20XX (journal article; registry record and protocol also retrieved and reconciled — see source log). Outcome construct: surgical site infection, CDC definition as stated in the paper. Time point: 30 days post-operative. Intervention arm: events 24, analysed n 133, randomised n 140. Control arm: events 37, analysed n 137, randomised n 141. Analysis population: as stated, participants with complete 30-day follow-up. Variability statistic reported in paper: 95% confidence interval on the risk difference — recorded as CI, not SD. Provenance: event counts from Table 2 text; percentages in Figure 2 digitised independently by both extractors as a cross-check, calibration points recorded. Extractor A / Extractor B agreement: yes. Reconciliation notes: none required.

All five outcome elements are present: counts, both denominators, the named variability statistic, and an unambiguous outcome and timing.

Randomised and analysed n are recorded separately, so attrition is visible without returning to the paper.

The variability field names what the paper actually reported rather than coercing it into an SD column.

Figure-derived values are flagged as such, digitised twice, with calibration recorded — which is what makes them auditable.

The source log records that registry and protocol were consulted, so a later outcome-reporting query has an answer.

The second row takes longer to complete once and cannot be lost. The first row forces a choice months later between estimating a denominator and excluding the study, a decision about the review’s result made by the design of a form rather than by a reviewer.

The current role of automation

Less than commercial claims suggest, but more than nothing. Jonnalagadda, Goyal and Huffman’s systematic review in Systematic Reviews remains the most reliable baseline: the automated extraction tools they reviewed each extracted only a small number of data elements, ranging from one to seven, and well over half of the data elements needed for systematic reviews had not been explored at all 3. Tsafnat and colleagues’ overview of systematic review automation technologies in the same journal maps where in the review pipeline automation is established, largely in searching and screening rather than extraction 5.

The implication for practice is a division of labour rather than a verdict. Automation is reasonable for high-volume, low-ambiguity fields, and for a first pass that a human then verifies against the source. It is not currently a substitute for either extractor in a double-extraction design, and a tool’s output should be treated as one extractor’s entry rather than as the reconciled value. Where a tool is used, the review should report which fields it populated and how its output was verified; this is the same provenance requirement that applies to figures, applied to a different source.

Automated tools are currently most useful at the screening stage, and the Academy’s page on abstract screening covers that separately.

The influence of data extraction on review conclusions

The scale of the downstream consequence should be kept in view. DerSimonian and Laird’s original paper re-analysed eight published clinical reviews containing between 5 and 26 randomised trials each 32. With study numbers of this size, one mis-extracted standard error in a heavily weighted study is not a rounding issue: it changes the weight that study receives, and under inverse-variance weighting the weights are the mechanism by which the pooled estimate is formed 32.

The methodological reviews that reviewers cite are themselves built on extraction. Veroniki and colleagues’ review of confidence-interval methods searched PubMed from inception to April 2016, retrieved 5,628 records with 20 further records from other sources, and found 69 publications meeting the eligibility criteria 1,36. This ratio is typical: the extraction effort falls on a small fraction of the records screened, which is precisely why it merits being done twice. Our methodology walkthrough picks up where this page leaves off, and biostatistics support is where extracted data becomes a model.

Key points

Key points

All five elements of every outcome should be extracted: counts or summary statistic, the denominator for each arm, the named variability statistic, and the outcome construct with its time point, because the denominator and the variability measure cannot be reconstructed later. At least two people should extract independently, as the MECIR standard requires, and the team should be scaled to the review: the Handbook suggests 2 to 3 extractors for fewer than 10 studies, 4 to 6 for 10 to 20, and more than 6 above that. Figures should be digitised with software rather than read from axes, provenance should be recorded for every value, and automated tools should be treated as one extractor rather than as a complete solution.

Application to your study

Application to your study

  1. Each of the five outcome elements should have its own field. These are events or summary statistic, analysed n, randomised n, named variability statistic, and outcome construct with time point, never a single combined “result” cell.
  2. The variability field should be labelled according to the statistic reported. Standard deviation (SD), standard error (SE), confidence interval (CI) and interquartile range (IQR) should be separate fields, not one field with a default heading.
  3. The form should be piloted on the three most complex studies. Crossover designs, multi-arm trials and outcomes reported only in figures are the cases the form must accommodate.
  4. Extraction should be independent, followed by reconciliation according to a stated rule. Sequential checking does not constitute independent extraction, and the arbitration procedure should be specified in the protocol.
  5. The disagreement rate should be recorded for each field. A field on which extractors repeatedly disagree is usually poorly worded, and revising it is more efficient than reconciling it fifty times.
  6. Figures should be digitised with software, twice, with the calibration points recorded. Every figure-derived value should be flagged in the dataset so that it can be examined in sensitivity analyses.
  7. Every source document should be logged for each study. Sources include the article, abstract, registry record, protocol and regulatory file, and the protocol should state which takes precedence when they conflict.
  8. The extraction dataset should be deposited with the paper. It is the record that enables readers to verify the numbers rather than accept them on trust.

Frequently asked questions

For each outcome, in each arm of each study, Cochrane asks for all five elements of the outcome: the events or summary statistic, the number of participants on which the result is based, the measure of variability or precision, and sufficient detail to identify the outcome construct and the time point unambiguously. In addition, the form should capture study design and unit of randomisation, eligibility criteria as stated, intervention detail sufficient to judge whether two studies delivered the same intervention (the TIDieR items provide a practical prompt), the analysis population for each result, how missing data were handled, funding and conflicts, and risk-of-bias judgements with a supporting quotation for each.

Yes, and Cochrane’s MECIR standard for data collection states it as a minimum: at least two people working independently. Buscemi and colleagues, in the Journal of Clinical Epidemiology, found that single data extraction generated more errors than double extraction. The crucial requirement is independence: two people extracting blind and then reconciling detect misread values that a sequential check does not, because a checker queries only the entries they happen to doubt. Jones and colleagues found a high prevalence of extraction and reporting errors even in Cochrane reviews, which had already been through methodological editing.

The Cochrane Handbook scales it to review size in Table 5.4.a: a small team of 2 to 3 data extractors for small-scale reviews of fewer than 10 included studies, 4 to 6 for small-to-medium reviews of 10 to 20 studies, and more than 6 for larger reviews. That is a team figure, separate from the MECIR requirement that any given study be extracted by at least two people independently. For a sole author planning a thirty-study review, the table indicates that the plan is not feasible as designed, not that the standard is optional.

Digitising software should be used rather than reading values from the axis. Jelicic Kadic and colleagues, in the Journal of Clinical Epidemiology, compared the two and found software extraction was both faster and more reliable between raters than manual extraction; Gross, Schirm and Scholz describe one such tool, Ycasd, in BMC Bioinformatics. The image should be calibrated against two known axis points, both extractors should digitise independently and compare results, and the software and calibration points should be recorded so that the procedure is reproducible. The dataset should flag that the value came from a figure, because that provenance affects how much weight it should carry and whether it belongs in a sensitivity analysis.

The statistic should be recorded exactly as the paper reported it, in a field labelled for that statistic, and the conversion question resolved at the analysis stage rather than in the extraction form. Two errors must be avoided: entering a median into an SD field because the form provides no other field, and converting during extraction so that the original reported value is lost. If a conversion or imputation is used, it belongs in the analysis plan with its method named, and the original statistic should remain in the dataset. Where a variability statistic is missing altogether, contacting the authors is the appropriate step before deciding to impute.

Often not. Mayo-Wilson, Li, Fusco and Dickersin’s guidance in Research Synthesis Methods, drawing on the MUDS study, shows that a single trial may be described across a journal article, conference abstract, registry record, protocol, statistical analysis plan and regulatory submission, and that these disagree. Chan and colleagues documented protocol-to-publication discrepancies in JAMA, and Turner and colleagues compared the published antidepressant literature with the regulatory record in the New England Journal of Medicine. The protocol should state which source takes precedence, every source consulted should be logged for each study, and discrepancies should be reported rather than silently resolved.

Only partly; the most reliable baseline is Jonnalagadda, Goyal and Huffman’s systematic review in Systematic Reviews: the tools they reviewed each extracted only a small number of data elements, ranging from one to seven, and well over half of the data elements needed for systematic reviews had not been explored at all. Tsafnat and colleagues’ map of systematic review automation shows traction mainly in searching and screening. A tool’s output should be treated as one extractor’s entry to be verified against the source, not as a reconciled value, and the review should report which fields it populated and how it was checked.

The sources are not fully in agreement, and the answer depends on where the error occurs. Gøtzsche and colleagues documented extraction errors in meta-analyses using standardised mean differences in JAMA. Jones and colleagues found errors common in Cochrane reviews but with low impact on conclusions in their sample. Inverse-variance weighting reconciles these findings: an error in a lightly weighted study is usually absorbed, while an error in the standard error of a heavily weighted study moves the pooled estimate directly. Because it cannot be known in advance which kind of error has been made, extraction should be performed twice.

References

  1. Chapter 5: Collecting data. Cochrane Handbook for Systematic Reviews of Interventions, version 6.5. Cochrane. 2024;6.5. read 9 October 2026. cochrane.org
  2. Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA (editors). Cochrane Handbook for Systematic Reviews of Interventions, version 6.5. Cochrane. 2024;6.5. read 9 October 2026. cochrane.org
  3. Jonnalagadda SR, Goyal P, Huffman MD. Automating data extraction in systematic reviews: a systematic review. Systematic Reviews. 2015;4:78. read 9 October 2026. cochrane.org
  4. Page MJ, McKenzie JE, Kirkham J, Dwan K, Kramer S, Green S, Forbes A. Bias due to selective inclusion and reporting of outcomes and analyses in systematic reviews of randomised trials of healthcare interventions. Cochrane Database of Systematic Reviews. 2014;10:MR000035. read 9 October 2026. cochrane.org
  5. Tsafnat G, Glasziou P, Choong MK, Dunn A, Galgani F, Coiera E. Systematic review automation technologies. Systematic Reviews. 2014;3(1):74. doi:10.1186/2046-4053-3-74 read 9 October 2026. doi.org/10.1186/2046-4053-3-74
  6. Ip S, Hadar N, Keefe S, Parkin C, Iovin R, Balk EM, Lau J. A web-based archive of systematic review data. Systematic Reviews. 2012;1:15. read 9 October 2026. cochrane.org
  7. The Nordic Cochrane Centre. Review Manager (RevMan) version 4.2 for Windows. The Cochrane Collaboration (software). 2003. read 9 October 2026. link.springer.com
  8. Mayo-Wilson E, Li T, Fusco N, Dickersin K. Practical guidance for using multiple data sources in systematic reviews and meta-analyses (with examples from the MUDS study). Research Synthesis Methods. 2018;9:2-12. read 9 October 2026. cochrane.org
  9. Page MJ, Altman DG, Shamseer L, McKenzie JE, Ahmadzai N, et al. Reproducible research practices are underused in systematic reviews of biomedical interventions. Journal of Clinical Epidemiology. 2018;94:8-18. doi:10.1016/j.jclinepi.2017.10.017 read 9 October 2026. doi.org/10.1016/j.jclinepi.2017.10.017
  10. Mayo-Wilson E, Li TJ, Fusco N, Bertizzolo L, Canner JK, Cowley T, et al. Cherry-picking by trialists and meta-analysts can drive conclusions about intervention efficacy. Journal of Clinical Epidemiology. 2017;91:95-110. read 9 October 2026. cochrane.org
  11. Jelicic Kadic A, Vucic K, Dosenovic S, Sapunar D, Puljak L. Extracting data from figures with software was faster, with higher interrater reliability than manual extraction. Journal of Clinical Epidemiology. 2016;74:119-123. read 9 October 2026. cochrane.org
  12. Hoffmann TC, Glasziou PP, Boutron I, Milne R, Perera R, Moher D, et al. Better reporting of interventions: template for intervention description and replication (TIDieR) checklist and guide. BMJ. 2014;348:g1687. read 9 October 2026. cochrane.org
  13. Stansfield C, Brunton G, Rees R. Search wide, dig deep: literature searching for qualitative research. Research Synthesis Methods. 2014;5(2):142-151. read 9 October 2026. frontiersin.org
  14. Gross A, Schirm S, Scholz M. Ycasd - a tool for capturing and scaling data from graphical representations. BMC Bioinformatics. 2014;15(1):219. doi:10.1186/1471-2105-15-219 read 9 October 2026. doi.org/10.1186/1471-2105-15-219
  15. Riley RD, Higgins JPT, Deeks JJ. Interpretation of random effects meta-analyses. BMJ. 2011;342:d549. doi:10.1136/bmj.d549 read 9 October 2026. doi.org/10.1136/bmj.d549
  16. Sterne JAC, Sutton AJ, Ioannidis JPA, Terrin N, Jones DR, Lau J, Carpenter J, Rucker G, Harbord RM, Schmid CH, Tetzlaff J, Deeks JJ, Peters J, Macaskill P, Schwarzer G, Duval S, Altman DG, Moher D, Higgins JPT. Recommendations for examining and interpreting funnel plot asymmetry in meta-analyses of randomised controlled trials. BMJ. 2011;343:d4002. doi:10.1136/bmj.d4002 read 9 October 2026. doi.org/10.1136/bmj.d4002
  17. Viechtbauer W. Conducting meta-analyses in R with the metafor package. Journal of Statistical Software. 2010;36(3):1-48. doi:10.18637/jss.v036.i03 read 9 October 2026. doi.org/10.18637/jss.v036.i03
  18. Viechtbauer W, Cheung MWL. Outlier and influence diagnostics for meta-analysis. Research Synthesis Methods. 2010;1(2):112-125. doi:10.1002/jrsm.11 read 9 October 2026. doi.org/10.1002/jrsm.11
  19. Simera I, Moher D, Hirst A, Hoey J, Schulz KF, Altman DG. Transparent and accurate reporting increases reliability, utility, and impact of your research: reporting guidelines and the EQUATOR Network. BMC Medicine. 2010;8:24. read 9 October 2026. trialsjournal.biomedcentral.com
  20. Borenstein M, Hedges LV, Higgins JPT, Rothstein HR. Introduction to Meta-Analysis. 2009. read 9 October 2026. medcalc.org
  21. Wood L, Egger M, Gluud LL, Schulz KF, Juni P, Altman DG, et al. BMJ. 2008;336:601-605. read 9 October 2026. lshtm-eprints-repo-prod.lshtm-eprints.cdl.cosector.com
  22. Bax L, Yu LM, Ikeda N, Tsuruta N, Moons KGM. A systematic comparison of software dedicated to meta-analysis of causal studies. BMC Medical Research Methodology. 2007;7:40. doi:10.1186/1471-2288-7-40 read 9 October 2026. doi.org/10.1186/1471-2288-7-40
  23. Gotzsche PC, Hrobjartsson A, Maric K, Tendal B. Data extraction errors in meta-analyses that use standardized mean differences. JAMA. 2007;298:430-437. read 9 October 2026. cochrane.org
  24. Bax L, Yu LM, Ikeda N, Tsuruta H, Moons KG. Development and validation of MIX: comprehensive free software for meta-analysis of causal research data. BMC Medical Research Methodology. 2006;6(1):50. doi:10.1186/1471-2288-6-50 read 9 October 2026. doi.org/10.1186/1471-2288-6-50
  25. Buscemi N, Hartling L, Vandermeer B, Tjosvold L, Klassen TP. Single data extraction generated more errors than double data extraction in systematic reviews. Journal of Clinical Epidemiology. 2006;59:697-703. read 9 October 2026. cochrane.org
  26. Jones AP, Remmington T, Williamson PR, Ashby D, Smyth RL. High prevalence but low impact of data extraction and reporting errors were found in Cochrane systematic reviews. Journal of Clinical Epidemiology. 2005;58:741-742. read 9 October 2026. cochrane.org
  27. Chan AW, Hrobjartsson A, Haahr MT, Gotzsche PC, Altman DG. JAMA. 2004;291:2457-2465. read 9 October 2026. lshtm-eprints-repo-prod.lshtm-eprints.cdl.cosector.com
  28. Shaw RL, Booth A, Sutton AJ, Miller T, Smith JA, Young B, et al. Finding qualitative research: an evaluation of search strategies. BMC Medical Research Methodology. 2004;4:5. read 9 October 2026. frontiersin.org
  29. Glasziou P, Irwig L, Bain C, Colditz G. Systematic reviews in health care: a practical guide. Cambridge University Press (book). 2001. read 9 October 2026. link.springer.com
  30. Egger M, Davey Smith G, Altman DG. Systematic reviews in health care: meta-analysis in context. BMJ Books (book). 2001. read 9 October 2026. link.springer.com
  31. Sutton AJ, Abrams KR, Jones DR, Sheldon TA, Song F. Methods for meta-analysis in medical research. Wiley (book). 2000. read 9 October 2026. link.springer.com
  32. DerSimonian R, Laird N. Meta-analysis in clinical trials. Controlled Clinical Trials. 1986;7(3):177-188. doi:10.1016/0197-2456(86)90046-2 read 9 October 2026. doi.org/10.1016/0197-2456(86)90046-2
  33. Hedges LV, Olkin I. Statistical methods for meta-analysis. 1985. read 9 October 2026. shiny.lapclin-aids.ini.fiocruz.br
  34. Fleiss JL. Measuring nominal scale agreement among many raters. Psychological Bulletin. 1971;76:378-382. read 9 October 2026. trialsjournal.biomedcentral.com
  35. Mantel N, Haenszel W. Journal of the National Cancer Institute. 1959;22:719-748. read 9 October 2026. medcalc.org
  36. Veroniki AA, Jackson D, Bender R, Kuss O, Langan D, Higgins JPT, Knapp G, Salanti G. Methods to calculate uncertainty in the estimated overall effect size from a random-effects meta-analysis. doi:10.1002/jrsm.1319 read 9 October 2026. doi.org/10.1002/jrsm.1319

Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.

How to cite this article

Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.

Pubrica Academy. (2026, October 9). Extracting Quantitative Data for Systematic Review and Meta-Analysis: Preserving the Denominator. Pubrica. https://pubrica.com/academy/latest-topics/how-to-extract-quantitative-data-for-systematic-review-and-meta-analysis/

This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.

Extraction form fields you can paste into your spreadsheet

The form uses one row per outcome per arm, not one row per study. It should be piloted on the three most complex included studies before extraction begins, and the provenance and reconciliation columns should be retained, because they make the dataset auditable when it is deposited.

QUANTITATIVE EXTRACTION FORM — FIELD LIST

A. STUDY IDENTIFICATION
   Study ID (first author + year)
   All source documents retrieved  (article / abstract / registry record /
     protocol / SAP / regulatory file / correspondence)
   Source precedence applied, per protocol rule
   Registration number
   Funding source; author conflicts of interest

B. DESIGN AND POPULATION
   Design; unit of randomisation; unit of analysis
   Country / setting / care level
   Eligibility criteria AS STATED
   N randomised, total and per arm
   Baseline characteristics needed for pre-specified subgroups

C. INTERVENTION AND COMPARATOR  (TIDieR prompts)
   What (materials, procedures)
   Who provided; how delivered; where
   How much, how often, over what period
   Tailoring or modifications during the study
   Comparator: placebo / active / usual care — describe

D. OUTCOME  — one row per outcome per arm; ALL FIVE ELEMENTS
   1. Outcome construct, as defined in the paper
   2. Time point of measurement
   3. Events, OR summary statistic (state which)
   4. Analysed n   |   Randomised n   (both, separately)
   5. Variability / precision statistic AND its type:
        [ ] SD   [ ] SE   [ ] 95% CI   [ ] IQR   [ ] other: ______
   Scale / instrument, and direction of a favourable score
   Analysis population as stated (ITT / per protocol / completers)
   Missing-data handling as stated

E. PROVENANCE  (per value)
   Location in source (table / figure / text / supplement, with number)
   Value read from a FIGURE?  [ ] no  [ ] yes — software: __________
        calibration points recorded: __________
   Value imputed or converted?  [ ] no  [ ] yes — method: __________
   Author contact made?  [ ] no  [ ] yes — date and outcome: _______

F. RISK OF BIAS
   Domain judgement + VERBATIM supporting quote, per domain

G. DOUBLE EXTRACTION AND RECONCILIATION
   Extractor A initials; Extractor B initials; both extracted blind? [ ]
   Fields disagreeing: ______________________
   Arbitrated by: __________   Resolution and reason: ______________
   Discrepancies BETWEEN SOURCES recorded (not silently resolved): [ ]

Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.

How Pubrica can help

Pubrica develops and pilots the extraction form against the protocol’s synthesis groups, so that every field required for the analysis exists before extraction begins, and provides a second independent extractor where the review team cannot. Pubrica digitises figure-only data with recorded calibration, reconciles articles with registry and protocol records, and returns a deposit-ready dataset with provenance for every value.

Find something else

Search the Pubrica Academy

Access is free and requires no account. Enter a topic, or open the full library to filter by category, sub-topic or resource type.

Pubrica returns a piloted extraction form developed from the protocol and three included papers.

A scoping assessment is free and takes two business days. It provides a written assessment of what the work requires and which specialist would undertake it.

Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.