Contents
On this page
- Limitations of standardised effect sizes
- Approaches to presenting a continuous pooled effect
- Significance, magnitude and precision as distinct findings
- Effect of heterogeneity on claims about magnitude
- Value of the GRADE summary-of-findings table
- Interpretation in the absence of an established anchor
- Formulating conclusions for guideline panels
- Recurrent errors in interpreting effect size
- Key points
- Application to your study
- Frequently asked questions
Academy · Research Data Analytics & AI · Interpreting effect size
Making Sense of Effect Size in Meta-Analysis: Presenting a Pooled Estimate for Clinical Use
A pooled standardised effect of 0.389 is a correct answer to a question that clinicians do not ask. Magnitude becomes meaningful only when it is expressed on a scale that can inform action, and the published guidance describes four approaches to achieving this, none of which is the conventional small, moderate or large label.
The principal difficulty in a meta-analysis is usually not the computation of the summary estimate but the interpretation of its meaning. Standardised units are convenient for pooling but of limited use in clinical practice; statistical significance conveys no information about magnitude; and the conventional verbal labels for effect size were not derived from clinical consequences. This article addresses the stage that follows model fitting: the selection of a presentation scale, the anchoring of magnitude to a measurable quantity, and the formulation of a conclusion that a guideline panel can use.
Learning objectives
- The reasons a pooled standardised effect cannot be interpreted directly, and the published alternatives
- The presentation of a continuous outcome in minimal important difference units, in the units of the most interpretable instrument, as a ratio of means, or dichotomised at a threshold
- The reasons statistical significance and clinical importance are separate findings requiring separate statements
- The effect of heterogeneity on permissible claims about magnitude, and the reasons the prediction interval belongs in the conclusion
- The way in which the GRADE summary-of-findings format requires interpretive work that a forest plot does not
- The approach when no anchor exists, a common situation that constitutes a reportable limitation rather than a justification for applying a convention
Common misconceptions
- “An SMD of approximately 0.4 represents a moderate effect.” The conventional small, moderate and large labels are statistical conventions rather than clinical thresholds. Their cut-points are deliberately not reproduced here, because no source in the evidence base of this article establishes them as clinically meaningful.
- “A significant pooled result indicates that the intervention is effective.” It indicates that the data are inconsistent with the absence of an average effect. Magnitude, precision and the variation across settings are three further questions, each requiring a separate statement.
- “Standardised units are the scientifically appropriate means of reporting a continuous outcome.” Guyatt and colleagues, in the GRADE guidance in the Journal of Clinical Epidemiology, describe the standardised approach as the most established but note that it is sensitive to differences in heterogeneity across populations and may be difficult to interpret.
- “The confidence interval indicates what a clinician should expect.” Under a random-effects model, the confidence interval describes the mean of a distribution of effects. The prediction interval is the quantity relevant to the next patient, and it is usually considerably wider.
- “If the magnitude cannot be interpreted, the number should be reported and its interpretation left to readers.” Readers are likely to substitute a convention. An explicit statement that no anchor exists is more informative and more transparent.
Limitations of standardised effect sizes
DerSimonian and Laird’s worked examples in Controlled Clinical Trials provide an instructive example. For the Winship review the weighted non-iterative overall effect was 0.389 with a standard error of 0.058, a precisely estimated summary on a standardised scale 32. The difficulty becomes apparent when its clinical application is considered. It is not a change in a score, a number of symptom-free days or a risk; it is a fraction of a standard deviation, and that standard deviation is specific to whichever populations happened to be enrolled 3.
This is not a criticism of standardisation, which addresses a genuine problem: the pooling of trials that used different instruments to measure the same construct. The methods for doing so correctly (the choice of denominator and estimator, and whether a small-sample correction was applied) are described in standardised mean difference estimation. The present argument is that even a correctly computed standardised estimate is an intermediate quantity, and treating it as the finding leaves the interpretive work incomplete.
Guyatt and colleagues describe this difficulty explicitly in the GRADE guidance on continuous outcomes in the Journal of Clinical Epidemiology: when studies measure the same construct with different instruments, the most established approach divides each study’s mean difference by its standard deviation, but this is sensitive to differences in heterogeneity across populations and may be difficult to interpret 12. They then list alternatives: dichotomisation at a threshold, reporting in the units of the most interpretable instrument, the ratio of means, and minimal important difference units.
A precisely estimated summary on a standardised scale, illustrating that precision and interpretability are distinct properties. Source: DerSimonian R, Laird N. Meta-analysis in clinical trials. Controlled Clinical Trials.
Approaches to presenting a continuous pooled effect
Each of the four alternatives involves a trade-off, and the choice should be made deliberately rather than by default.
The units of the most interpretable instrument. The pooled standardised estimate is converted back to the scale of the instrument most familiar to the intended audience by multiplying it by that instrument’s standard deviation 12. The result is expressed as a number of points on a familiar scale. The limitation is that the conversion depends on the standard deviation selected; the figure should therefore be presented with that standard deviation stated and, ideally, with a range reflecting plausible alternatives.
Minimal important difference units. The estimate is divided by the smallest change on the instrument that patients perceive as beneficial. Johnston, Thorlund, Schünemann and colleagues developed this approach in Health and Quality of Life Outcomes and extended it in the Journal of Clinical Epidemiology, and its advantage is that the resulting value indicates directly whether the average effect exceeds the smallest difference that patients notice 14,19. The limitation is that a credible minimal important difference must exist for the instrument concerned, and for many outcomes none does.
Ratio of means. The effect is expressed as a proportional rather than an absolute or standardised change. Because no standard deviation appears in the denominator, the measure is comparable across instruments and avoids the problem of differing population variability, although it is more difficult to combine with absolute-risk reasoning.
Dichotomisation at a threshold. The continuous outcome is converted into the proportion of patients achieving a defined response, which yields a risk difference or number needed to treat. This is the most directly actionable presentation but also the one that discards the most information, and it depends entirely on the defensibility of the threshold.
Five presentations of a pooled continuous effect and their respective limitations
| Presentation | What the reader gets | What it requires | What it costs |
|---|---|---|---|
| Standardised mean difference | A dimensionless estimate that pools across instruments | Nothing beyond the pooled analysis itself | Not directly interpretable; sensitive to differing population variability across studies |
| Units of the most interpretable instrument | A number of points on a scale the audience knows | A standard deviation for that instrument, stated explicitly | The figure depends on which standard deviation was used for the conversion |
| Minimal important difference units | Whether the average effect exceeds the smallest change patients notice | An established minimal important difference for the instrument | Often unavailable; where values differ between sources, the choice must be justified |
| Ratio of means | A proportional change, comparable across instruments | Means on a ratio scale with a meaningful zero | Harder to combine with absolute-risk reasoning or with a number needed to treat |
| Dichotomised at a threshold | A proportion achieving response, convertible to risk difference or number needed to treat | A defensible response threshold | Discards information; the whole result depends on where the threshold is set |
The four alternatives to the standardised mean difference are those listed by Guyatt GH, Thorlund K, Oxman AD, Walter SD, et al., GRADE guidelines 13: preparing summary of findings tables — continuous outcomes, Journal of Clinical Epidemiology; the minimal important difference approach is developed in Johnston and colleagues (Health and Quality of Life Outcomes; Journal of Clinical Epidemiology). The cost column is editorial guidance drawn from those sources.
Significance, magnitude and precision as distinct findings
A conclusion stating that the intervention was effective because the pooled result was significant answers one question and implies answers to two others. Most over-claiming can be prevented by a simple procedural rule: a separate sentence should be written for each question.
The first states whether the data are consistent with no average effect. The second states the magnitude on an interpretable scale, with the anchor named. The third states the precision — the confidence interval, and whether the lower bound is still clinically worthwhile. A fourth, where a random-effects model was used, states the spread across settings via the prediction interval.
The fourth sentence is the one most frequently omitted, although Riley, Higgins and Deeks argued in BMJ that it is the sentence most relevant to clinicians: the prediction interval describes the effect to be expected in a new comparable setting, and it is typically substantially wider than the confidence interval around the mean 16. A pooled effect whose confidence interval excludes no effect but whose prediction interval does not is a substantively different finding from one in which both exclude it, and the distinction should be reported in the abstract rather than in an appendix. The choice of model itself is addressed in fixed-effect or random-effects.
Effect of heterogeneity on claims about magnitude
Magnitude claims are claims about a central tendency, and a central tendency is informative only if the distribution around it is narrow. Higgins and Thompson proposed three statistics in Statistics in Medicine — H, I-squared and R — to describe heterogeneity independently of the effect metric and the number of studies, illustrated on five datasets, and Higgins, Thompson, Deeks and Altman set out the applied interpretation of inconsistency in BMJ 26,27. Both papers are about quantifying dispersion rather than classifying it, and neither supplies a threshold above which a pooled magnitude becomes uninterpretable.
A practical drafting rule nevertheless applies. If the individual study estimates lie on both sides of the threshold of clinical importance, the position of the average relative to that threshold is not the finding; the distribution across the threshold is. The range of study estimates should be reported alongside the summary, with a statement of which side of the anchor each estimate falls on. This is more informative than any heterogeneity statistic and is immediately interpretable by a clinical reader.
In analyses of continuous outcomes a further consideration arises: part of the dispersion may result from the standardisation itself, because each study’s estimate is scaled by the variability of its own population 3,4. Where this is plausible, it should be stated, and a same-instrument sensitivity analysis should be considered, as described in standardised mean difference estimation.
Value of the GRADE summary-of-findings table
The GRADE summary-of-findings format, set out by Guyatt and colleagues in the Journal of Clinical Epidemiology and introduced in BMJ, is useful here for a reason that has little to do with certainty ratings. Its structure requires an absolute effect for each outcome, expressed in a way a reader can act on, next to the relative effect and the certainty judgement 12,15. The row cannot be completed without undertaking the interpretive work, whereas a forest plot alone allows authors to avoid it.
Completing the table early rather than at the end tends to expose two problems while they can still be corrected. The first is that the chosen outcome has no interpretable absolute presentation, which is an argument for changing the outcome or identifying an anchor. The second is a tension between the certainty rating and the magnitude claim: a large effect reported with low certainty requires different wording in the abstract from a large effect reported with high certainty, and the table makes this discrepancy visible.
Where pooling is not possible at all, the obligation to interpret the evidence remains. Campbell and colleagues’ SWiM guideline in BMJ sets out how to report synthesis without meta-analysis, and Higgins and colleagues, in BMJ Global Health, address synthesising quantitative evidence for complex interventions where a single pooled number would mislead. Both expect a statement about magnitude and direction; neither accepts a table of p-values as a synthesis 6,8.
Interpretation in the absence of an established anchor
This situation is common and warrants explicit handling. Many instruments have no established minimal important difference, no accepted response threshold and no version that a general clinical audience interprets readily. Reverting to the conventional verbal labels should be avoided: those conventions originate in behavioural-science rules of thumb and were not derived from clinical consequences for any particular outcome 12,19. Their cut-points are deliberately not reproduced on this page, because no source in the evidence base of this article establishes them as clinically meaningful.
Authors should instead provide a brief, specific statement of the pooled effect on the standardised scale, the range of the instrument, the absence of an established anchor for it, and the fact that the clinical importance of an effect of this size on this instrument has not been established. Such a statement is usable by a guideline panel. “A moderate effect” is not, because the panel cannot determine whether the label reflects a measurement or a convention.
Two partial substitutes are sometimes available. Where an included trial reported both the instrument and a patient-reported global rating of change, that trial can provide an informal anchor for the whole review, provided that it is identified as such. Where a related instrument has an established minimal important difference and a published correlation with the instrument in question, the conversion can be reported as approximate, with its source 14. Neither approach is ideal, but both are preferable to a convention.
Formulating conclusions for guideline panels
These considerations culminate in three or four sentences in the abstract and a corresponding passage in the discussion. A weak conclusion reports a standardised value, attaches a conventional label and asserts effectiveness. A strong conclusion converts the estimate, names the anchor, separates magnitude from precision and describes the variation across settings.
The same pooled result reported with two different conclusions (a constructed example; the figures in both versions are illustrative placeholders, not findings from any review)
Over-claims
The intervention significantly improved symptom scores compared with usual care (SMD 0.41, 95% CI 0.19 to 0.63, p<0.001), representing a moderate treatment effect. The intervention should be considered for routine use in this population.
What does 0.41 mean to a patient? The estimate is not converted, so its magnitude cannot be assessed.
Where does “moderate” come from? A statistical convention, not a clinical threshold for this instrument.
Is the lower bound still worthwhile? The interval is reported but is not interpreted against any reference value.
What should a clinician expect in their setting? No prediction interval is reported, so the variation across settings is not shown.
Does significance justify a practice recommendation? It does not, yet the final sentence makes such a recommendation.
Usable by a panel
The intervention improved symptom scores relative to usual care; the pooled standardised mean difference was 0.41 (95% CI 0.19 to 0.63). Converted to the instrument most widely used in practice, whose control-arm standard deviation in these trials was 9.4 points on a 0–60 scale, this corresponds to about 3.9 points (95% CI 1.8 to 5.9). The published minimal important difference for that instrument is 5 points, so the average effect falls below the smallest change patients report noticing, and the confidence interval includes values well below it. Individual trial estimates ranged from 1.1 to 7.2 points, straddling that threshold, and the 95% prediction interval for a new comparable trial was −0.8 to 8.6 points. On this evidence the average benefit is unlikely to be perceptible to most patients, although a minority of settings may see effects above the threshold; certainty was rated moderate, limited by inconsistency.
The estimate is converted into points on a named instrument, with the standard deviation used for the conversion stated.
Magnitude is anchored to a published minimal important difference rather than a verbal convention.
The confidence interval is interpreted relative to that anchor, not merely quoted.
The range of individual estimates and the prediction interval show that trials straddle the threshold.
The conclusion matches what the evidence supports, and the certainty rating qualifies it.
The second version reaches a weaker conclusion from the same statistically significant result. This is intended: the significance was never in question and was never the finding. Anchoring the magnitude changed the recommendation, and a guideline panel will perform this step whether or not the authors do so first.
Recurrent errors in interpreting effect size
Reporting only the relative effect for a binary outcome. A risk ratio without a baseline risk cannot be converted into a quantity that a patient can evaluate. The absolute effect should be reported at a stated baseline risk and, where baseline risk varies substantially across the evidence, at more than one baseline risk.
Interpreting the summary when the studies straddle the threshold. If some trials lie above and others below the minimal important difference, the average is a poor description of the evidence, however precisely it is estimated.
Allowing multiple outcomes to inflate a conclusion. López-López, Page, Lipsey and Higgins set out the handling of effect-size multiplicity in Research Synthesis Methods, and Mavridis and Salanti cover the multivariate approach in Statistical Methods in Medical Research 9,13. Reporting whichever outcome produced the largest estimate constitutes reporting bias, and the interpretation section is where reviewers tend to detect it.
Overinterpreting precision in a small or asymmetric evidence base. Sterne and colleagues’ BMJ recommendations on funnel plot asymmetry explain why small-study effects inflate apparent magnitude; a confident magnitude claim based on a small number of small trials is the pattern most often overturned subsequently 17. McKenzie, Beller and Forbes give a compact overview of the surrounding methodology in Respirology, and PRISMA 2020, published by Page and colleagues in PLOS Medicine, requires the certainty assessment that makes these caveats visible rather than optional 5,11. Our meta-analysis and medical writing teams treat the interpretation paragraph as a deliverable in its own right, and the relevant reporting requirements are collected in reporting guidelines.
Key points
Key points
A pooled standardised effect is an intermediate quantity rather than a finding. It should be converted to an anchored scale (points on an instrument familiar to the audience, minimal important difference units, a ratio of means or the proportion achieving a defined response), and the anchor used should be named. Significance, magnitude, precision and variation across settings should be reported in four separate sentences, and the prediction interval should be included when a random-effects model has been fitted. Where the individual trial estimates lie on both sides of the threshold of clinical importance, that distribution constitutes the result, and the average summarises it poorly. Where no anchor exists, this should be stated explicitly rather than replaced with a conventional label.
Application to your study
Application to your study
- Determine the presentation scale at the protocol stage rather than after the analysis. A standardised scale is appropriate for pooling, but the scale on which results will be presented should be selected and recorded in advance, because it determines the data that must be extracted.
- Identify the anchor in advance. A published minimal important difference or an accepted response threshold for the primary instrument should be sought, and its source and the population from which it was derived recorded.
- Convert the pooled estimate and both interval bounds, not only the point estimate. The standard deviation used for the conversion should be stated so that readers can verify it.
- Report the findings in four separate sentences. These address consistency with no effect, magnitude relative to the anchor, precision and whether the lower bound remains clinically worthwhile, and variation across settings as indicated by the prediction interval.
- Report the range of individual study estimates on the presentation scale. If the estimates lie on both sides of the clinical threshold, this should be stated in the abstract, because it is the finding a guideline panel requires.
- Report an absolute effect at a stated baseline risk for every binary outcome. If baseline risk varies widely across the evidence, the absolute effect should be given at two plausible values rather than one.
- Complete the GRADE summary-of-findings row at an early stage. An inability to complete the absolute-effect cell indicates an interpretive problem that is more easily resolved at this stage than during revision.
- Where no anchor exists, state this explicitly in a sentence. The sentence should report the pooled effect, the instrument range, the absence of an established minimal important difference and the fact that the clinical importance of this magnitude has not been established, without applying a conventional label.
Frequently asked questions
The estimate should be converted to a scale with a known meaning, and the conversion should be named. Guyatt and colleagues, in the GRADE guidance on continuous outcomes in the Journal of Clinical Epidemiology, list four approaches: dichotomisation at a defined threshold to give the proportion achieving response, reporting in the units of the most interpretable instrument, use of the ratio of means, and expression of the result in minimal important difference units. Each requires an input that may not be available (a defensible threshold, a standard deviation, a ratio-scaled outcome or an established minimal important difference), so the choice depends on the outcome. Reporting the standardised value alone with a verbal label attached is not an adequate alternative.
That framing derives from conventional statistical rules of thumb rather than from clinical evidence about any particular outcome, and the conventional cut-points are deliberately not reproduced on this page because no source in the evidence base of this article establishes them as clinically meaningful. The same standardised value can be trivial on one instrument and important on another, depending on the range of the instrument and on the smallest change that patients notice. The defensible approach is to compare the pooled effect with a published minimal important difference for the instrument. If none exists, this should be reported, together with a statement that the clinical importance of an effect of this size on this instrument has not been established.
It is the smallest change in an instrument’s score that patients perceive as beneficial, usually established by anchoring score changes to patients’ own global ratings of change. Johnston, Thorlund, Schünemann and colleagues set out its use in meta-analysis in Health and Quality of Life Outcomes and extended the methods in the Journal of Clinical Epidemiology: dividing the pooled effect by the minimal important difference expresses the result in units of the smallest noticeable change, so a value below one indicates an average effect that patients would not perceive. The minimal important difference used, its source and the population from which it was derived should be reported, since published values for a single instrument often differ.
Only after dichotomisation at a threshold, and only if that threshold is defensible. Converting a continuous outcome into the proportion of patients achieving a defined response yields a risk difference, from which a number needed to treat follows; this is the most directly actionable presentation available, which is why GRADE lists it among the alternatives to a standardised estimate. Its limitations are that it discards information and that the entire result depends on the position of the threshold. The threshold and its source should be reported, ideally with the result at a second plausible threshold, so that readers can assess the sensitivity of the conclusion.
Both findings should be stated in separate sentences. The first states that the data are inconsistent with no average effect. The second gives the magnitude on an anchored scale and states whether it reaches the smallest difference that patients notice. A statistically significant effect that falls below the minimal important difference is a legitimate and useful finding, and stating it explicitly is considerably more defensible than implying effectiveness from the p-value. If the confidence interval includes values above the threshold, this should also be stated: the appropriate conclusion may be that a worthwhile effect has not been excluded, rather than that one has been demonstrated.
Not automatically, but it changes what the magnitude means. Higgins and Thompson introduced H, I-squared and R in Statistics in Medicine to describe heterogeneity independently of the effect metric and the number of studies, and the companion BMJ paper on measuring inconsistency states explicitly that the aim is to quantify rather than to classify heterogeneity; there is no threshold above which an average becomes invalid. The practical test is whether the individual study estimates lie on both sides of the threshold of clinical importance. If they do, the range should be reported and this distribution stated, because the position of the average relative to the threshold is then not an adequate description of the evidence.
Fitting a random-effects model assumes that true effects vary between settings, and the confidence interval then describes the mean of that variation rather than the effect to be expected in a new setting. Riley, Higgins and Deeks made this argument in BMJ. The prediction interval is typically much wider and frequently includes no effect where the confidence interval does not; although unwelcome, this is the correct implication of the model. Reporting it in the abstract rather than in the supplement also anticipates the most common criticism of a positive meta-analysis, namely that its conclusion is stronger than its evidence base supports.
This situation is common, and the appropriate response is a specific statement of the limitation rather than a convention. The pooled effect on the standardised scale and the range of the instrument should be reported, together with a statement that no established minimal important difference exists for it and that the clinical importance of an effect of this size on this instrument has not been established. Two partial substitutes are sometimes available: an included trial that reported both the instrument and a patient global rating of change can serve as an informal anchor if identified as such, and a related instrument with a published minimal important difference and a documented correlation can support an approximate conversion. Both are imperfect, but both are preferable to a label.
References
- Deeks JJ, Higgins JPT, Altman DG. Analysing data and undertaking meta-analyses. Cochrane Handbook for Systematic Reviews of Interventions. 2019:241-284. doi:10.1002/9781119536604.ch10 read 9 October 2026. doi.org/10.1002/9781119536604.ch10
- European Medicines Agency, Committee for Medicinal Products for Human Use. Guideline on the choice of the non-inferiority margin. European Medicines Agency (guideline). 2005. read 9 October 2026. trialsjournal.biomedcentral.com
- Jung J, Aloe AM. Standardized mean differences: not so standard after all. Campbell Systematic Reviews. 2025;21(3):e70056. doi:10.1002/cl2.70056 read 9 October 2026. doi.org/10.1002/cl2.70056
- Lin L, Aloe AM. Evaluation of various estimators for standardized mean difference in meta-analysis. Statistics in Medicine. 2021;40(2):403-426. doi:10.1002/sim.8781 read 9 October 2026. doi.org/10.1002/sim.8781
- Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, Shamseer L, Tetzlaff JM, Akl EA, Brennan SE, Chou R, Glanville J, Grimshaw JM, Hrobjartsson A, Lalu MM, Li T, Loder EW, Mayo-Wilson E, McDonald S, McGuinness LA, Stewart LA, Thomas J, Tricco AC, Welch VA, Whiting P, Moher D. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. PLOS Medicine. 2021;18(3):e1003583. doi:10.1371/journal.pmed.1003583 read 9 October 2026. doi.org/10.1371/journal.pmed.1003583
- Campbell M, McKenzie JE, Sowden A, Katikireddi SV, Brennan SE, Ellis S, Hartmann-Boyce J, Ryan R, Shepperd S, Thomas J, Welch V, Thomson H. Synthesis without meta-analysis (SWiM) in systematic reviews: reporting guideline. BMJ. 2020:l6890. doi:10.1136/bmj.l6890 read 9 October 2026. doi.org/10.1136/bmj.l6890
- Amrhein V, Greenland S, McShane B. Retire statistical significance. Nature. 2019;567(7748):305-307. doi:10.1038/d41586-019-00857-9 read 9 October 2026. doi.org/10.1038/d41586-019-00857-9
- Higgins JPT, Lopez-Lopez JA, Becker BJ, Davies SR, Dawson S, Grimshaw JM, McGuinness LA, Moore THM, Rehfuess EA, Thomas J, Caldwell DM. Synthesising quantitative evidence in systematic reviews of complex health interventions. BMJ Global Health. 2019;4(Suppl 1):e000858. doi:10.1136/bmjgh-2018-000858 read 9 October 2026. doi.org/10.1136/bmjgh-2018-000858
- Lopez-Lopez JA, Page MJ, Lipsey MW, Higgins JPT. Dealing with effect size multiplicity in systematic reviews and meta-analyses. Research Synthesis Methods. 2018;9(3):336-351. doi:10.1002/jrsm.1310 read 9 October 2026. doi.org/10.1002/jrsm.1310
- Ahlbom A. Significance testing: why does it prevail? European Journal of Epidemiology. 2017;32(1):1-2. doi:10.1007/s10654-016-0220-0 read 9 October 2026. doi.org/10.1007/s10654-016-0220-0
- McKenzie JE, Beller EM, Forbes AB. Introduction to systematic reviews and meta-analysis. Respirology. 2016;21(4):626-637. doi:10.1111/resp.12783 read 9 October 2026. doi.org/10.1111/resp.12783
- Guyatt GH, Thorlund K, Oxman AD, Walter SD, Patrick D, Furukawa TA, Johnston BC, Karanicolas P, Akl EA, Vist G, Kunz R, Brozek J, Kupper LL, Martin SL, Meerpohl JJ, Alonso-Coello P, Christensen R, Schunemann HJ. GRADE guidelines: 13. Preparing summary of findings tables and evidence profiles - continuous outcomes. Journal of Clinical Epidemiology. 2013;66(2):173-183. doi:10.1016/j.jclinepi.2012.08.001 read 9 October 2026. doi.org/10.1016/j.jclinepi.2012.08.001
- Mavridis D, Salanti G. A practical introduction to multivariate meta-analysis. Statistical Methods in Medical Research. 2012;22(2):133-158. doi:10.1177/0962280211432219 read 9 October 2026. doi.org/10.1177/0962280211432219
- Johnston BC, Thorlund K, da Costa BR, Furukawa TA, Guyatt GH. New methods can extend the use of minimal important difference units in meta-analyses of continuous outcome measures. Journal of Clinical Epidemiology. 2012;65(8):817-826. doi:10.1016/j.jclinepi.2012.02.008 read 9 October 2026. doi.org/10.1016/j.jclinepi.2012.02.008
- Guyatt G, Oxman AD, Akl EA, Kunz R, Vist G, Brozek J, Norris S, Falck-Ytter Y, Glasziou P, DeBeer H, Jaeschke R, Rind D, Meerpohl J, Dahm P, Schunemann HJ. GRADE guidelines: 1. Introduction - GRADE evidence profiles and summary of findings tables. Journal of Clinical Epidemiology. 2011;64:383-394. doi:10.1016/j.jclinepi.2010.04.026 read 9 October 2026. doi.org/10.1016/j.jclinepi.2010.04.026
- 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
- 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
- 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
- Johnston BC, Thorlund K, Schunemann HJ, Xie F, Murad MH, Montori VM, Guyatt GH. Improving the interpretation of quality of life evidence in meta-analyses: the application of minimal important difference units. Health and Quality of Life Outcomes. 2010;8:116. doi:10.1186/1477-7525-8-116 read 9 October 2026. doi.org/10.1186/1477-7525-8-116
- Borenstein M, Hedges LV, Higgins JPT, Rothstein HR. Introduction to Meta-Analysis. 2009. read 9 October 2026. medcalc.org
- Guyatt GH, Oxman AD, Vist GE, Kunz R, Falck-Ytter Y, Alonso-Coello P, Schunemann HJ. GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ. 2008;336:924-926. doi:10.1136/bmj.39489.470347.AD read 9 October 2026. doi.org/10.1136/bmj.39489.470347.AD
- 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
- Viechtbauer W. Bias and efficiency of meta-analytic variance estimators in the random-effects model. Journal of Educational and Behavioral Statistics. 2005;30(3):261-293. doi:10.3102/10769986030003261 read 9 October 2026. doi.org/10.3102/10769986030003261
- Borenstein M, Hedges L, Higgins J, Rothstein H. Comprehensive Meta-Analysis version 2. Biostat (software). 2005. read 9 October 2026. link.springer.com
- Leandro G. Meta-analysis in medical research. Blackwell Publishing / BMJ Books (book). 2005. read 9 October 2026. link.springer.com
- Higgins JPT, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ. 2003;327(7414):557-560. doi:10.1136/bmj.327.7414.557 read 9 October 2026. doi.org/10.1136/bmj.327.7414.557
- Higgins JPT, Thompson SG. Quantifying heterogeneity in a meta-analysis. Statistics in Medicine. 2002;21(11):1539-1558. doi:10.1002/sim.1186 read 9 October 2026. doi.org/10.1002/sim.1186
- 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
- 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
- Jones B, Jarvis P, Lewis JA, Ebbutt AF. Trials to assess equivalence: the importance of rigorous methods. BMJ. 1996;313:36-39. read 9 October 2026. trialsjournal.biomedcentral.com
- Eysenck HJ. Meta-analysis and its problems. BMJ. 1994;309:789-792. read 9 October 2026. link.springer.com
- 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
- Hedges LV, Olkin I. Statistical methods for meta-analysis. 1985. read 9 October 2026. shiny.lapclin-aids.ini.fiocruz.br
- Hedges LV. Distribution theory for Glass's estimator of effect size and related estimators. Journal of Educational Statistics. 1981;6(2):107-128. read 9 October 2026. faculty.ucmerced.edu
- Mantel N, Haenszel W. Journal of the National Cancer Institute. 1959;22:719-748. read 9 October 2026. medcalc.org
- Cochran WG. The combination of estimates from different experiments. Biometrics. 1954;10(1):101-129. doi:10.2307/3001666 read 9 October 2026. doi.org/10.2307/3001666
- 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). Making Sense of Effect Size in Meta-Analysis: Presenting a Pooled Estimate for Clinical Use. Pubrica. https://pubrica.com/academy/medical-writing/making-sense-of-effect-size-in-meta-analysis-based-for-medical-research/
Pubrica Academy. “Making Sense of Effect Size in Meta-Analysis: Presenting a Pooled Estimate for Clinical Use.” Pubrica, 9 Oct. 2026, pubrica.com/academy/medical-writing/making-sense-of-effect-size-in-meta-analysis-based-for-medical-research/.
Pubrica Academy (2026) Making Sense of Effect Size in Meta-Analysis: Presenting a Pooled Estimate for Clinical Use. Pubrica. Available at: https://pubrica.com/academy/medical-writing/making-sense-of-effect-size-in-meta-analysis-based-for-medical-research/ (Accessed: 15 October 2026).
Pubrica Academy. “Making Sense of Effect Size in Meta-Analysis: Presenting a Pooled Estimate for Clinical Use.” Pubrica. October 9, 2026. https://pubrica.com/academy/medical-writing/making-sense-of-effect-size-in-meta-analysis-based-for-medical-research/
Pubrica Academy, ‘Making Sense of Effect Size in Meta-Analysis: Presenting a Pooled Estimate for Clinical Use’ (Pubrica, 9 October 2026) <https://pubrica.com/academy/medical-writing/making-sense-of-effect-size-in-meta-analysis-based-for-medical-research/> accessed 15 October 2026.
Pubrica Academy. Making Sense of Effect Size in Meta-Analysis: Presenting a Pooled Estimate for Clinical Use. [Internet]. Pubrica; 2026 Oct 9 [cited 2026 Oct 15]. Available from: https://pubrica.com/academy/medical-writing/making-sense-of-effect-size-in-meta-analysis-based-for-medical-research/
Pubrica Academy, “Making Sense of Effect Size in Meta-Analysis: Presenting a Pooled Estimate for Clinical Use,” Pubrica, Oct. 9, 2026. [Online]. Available: https://pubrica.com/academy/medical-writing/making-sense-of-effect-size-in-meta-analysis-based-for-medical-research/
@misc{making_sense_of_effect_size_in_meta_an_2026,
author = {{Pubrica Academy}},
title = {Making Sense of Effect Size in Meta-Analysis: Presenting a Pooled Estimate for Clinical Use},
howpublished = {Pubrica},
year = {2026},
month = {oct},
url = {https://pubrica.com/academy/medical-writing/making-sense-of-effect-size-in-meta-analysis-based-for-medical-research/},
note = {Accessed: <span data-cite-acc="iso">2026-10-15</span>}
}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.
Interpretation worksheet you can copy
This worksheet should be completed once the pooled analysis is available. Sections 1 to 4 produce the numbers, and section 6 assembles them into the four-sentence conclusion. If section 2 cannot be completed, the user should proceed directly to section 5 and record the limitation instead; this is a legitimate output.
EFFECT SIZE INTERPRETATION WORKSHEET
Review: ______________________ Outcome: ______________________
Pooled estimate (analysis scale): ________ 95% CI ________ to ________
Model used: fixed effect / random effects Prediction interval: ______ to ______
I-squared ______% Range of individual study estimates: ______ to ______
1. PRESENTATION SCALE (choose one, record the reason)
[ ] units of the most interpretable instrument
instrument: ______________ SD used for conversion: ______ source: ________
[ ] minimal important difference units
[ ] ratio of means
[ ] dichotomised at a threshold; threshold ________ source: ______________
[ ] standardised only, because no anchor exists -> go to section 5
2. THE ANCHOR
Minimal important difference / response threshold: ________
Source: ______________________ Population it was derived in: ______________
Does a competing published value exist? Y / N If yes: ________
3. CONVERTED RESULT (convert all three, not just the point estimate)
Point estimate on presentation scale ...... ________
Lower confidence bound .................... ________
Upper confidence bound .................... ________
Prediction interval, converted ............ ______ to ______
Individual study estimates, converted ..... ______ to ______
4. POSITION RELATIVE TO THE ANCHOR
Point estimate: above / below the anchor
Lower confidence bound: above / below
Upper confidence bound: above / below
Do individual studies straddle the anchor? Y / N
[ ] if yes, the straddling is reported in the abstract as the finding
5. IF NO ANCHOR EXISTS (write this sentence verbatim and fill the blanks)
"The pooled standardised mean difference was ________ (95% CI ________ to
________) on the ______________, which ranges from ______ to ______. No
established minimal important difference exists for this instrument, so the
clinical importance of an effect of this size is not established."
6. THE FOUR-SENTENCE CONCLUSION
(1) Consistency with no effect:
____________________________________________________________
(2) Magnitude against the anchor:
____________________________________________________________
(3) Precision, and whether the lower bound is still worthwhile:
____________________________________________________________
(4) Spread across settings (prediction interval; random effects only):
____________________________________________________________
7. GRADE SUMMARY-OF-FINDINGS ROW (if you cannot fill this, stop and fix it)
Outcome ____________ Participants / studies ____________
Relative effect ____________ Absolute effect ____________
Baseline risk assumed (binary outcomes) ______% Second value ______%
Certainty ____________ Reason for any downgrade ____________________ 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’s medical writing and biostatistics teams identify a defensible anchor for the primary outcome, convert the pooled estimate and both interval bounds to a presentation scale with the conversion stated, and draft the four-sentence conclusion and the GRADE summary-of-findings row together so that the claim in the abstract and the certainty rating agree. Where no anchor exists for the instrument, Pubrica states the limitation explicitly rather than applying a conventional label.