Contents
On this page
- Importance of the interval, rather than the point estimate alone, for harm outcomes
- Weighing of benefits and harms in GRADE
- Unequal interpretive weight of benefit and harm intervals
- A documented gap in the reporting of harm outcomes
- A sequence for interpreting benefit and harm evidence together
- Expressing the harm interval in absolute, patient-relevant terms
- A worked example of harm and benefit
Academy · Research Data, Analytics & AI · CIs for Harm vs Benefit
Confidence Intervals in the Assessment of Harm and Benefit: GRADE and Absolute Effects
A confidence interval that spans both a meaningful benefit and a meaningful harm does not establish the direction of a treatment’s effect. For this reason, guidance on interpreting harm-benefit trade-offs treats the entire interval, rather than the point estimate, as the basis for the decision.
Assessing whether a treatment's benefits outweigh its harms requires looking separately at the confidence intervals around its estimated effects on desirable outcomes and on adverse outcomes, because a point estimate alone can suggest a favorable trade-off while the surrounding uncertainty in either direction tells a very different story. A confidence interval for a harm outcome that includes a substantially increased risk, even if the point estimate itself looks modest, means the data have not ruled out a meaningfully worse safety profile than the single estimate implies1. Methodological guidance for weighing harms against benefits in clinical evidence consistently treats both intervals, for benefit and for harm, as the basis for the judgment, not the point estimates in isolation.
Learning objectives
- Why the confidence interval around a harm estimate matters as much as the interval around a benefit estimate
- How GRADE guidance incorporates both benefit and harm imprecision into an overall judgment
- Why a wide harm interval that includes serious risk should not be treated the same as a narrow one showing a small, precisely estimated risk
Common misconceptions
- “If a treatment's average benefit looks larger than its average harm, the trade-off is favorable.” Comparing point estimates of average benefit and average harm in isolation ignores the uncertainty around each; a harm outcome's confidence interval can include a serious adverse effect even when its point estimate looks small, and GRADE and other evidence-grading frameworks explicitly require weighing the interval, not just the central estimate, for both benefit and harm before judging the trade-off.
Importance of the interval, rather than the point estimate alone, for harm outcomes
A confidence interval around an estimated adverse-event rate conveys the plausible range of that rate given the data collected, and when a trial or meta-analysis has limited numbers of adverse events, which is common since serious harms are often rarer than the primary benefit outcome being studied, that interval can be wide even when the point estimate itself appears reassuring1. The Cochrane Handbook’s guidance on interpreting results notes that harm outcomes frequently have less statistical power than benefit outcomes in the same study, precisely because trials are usually sized around the primary efficacy outcome rather than around rarer safety events, so that the harm interval requires separate, careful attention rather than being read with the same apparent confidence as the benefit interval from the same trial2.
Weighing of benefits and harms in GRADE
The GRADE framework for moving from evidence to a clinical recommendation explicitly separates the certainty assessment for benefit outcomes and harm outcomes, and factors the imprecision of each, driven by the width of its own confidence interval, into the overall judgment about whether a treatment's desirable effects clearly outweigh its undesirable ones3. This structure means a recommendation can be rated differently depending on whether the uncertainty sits mainly in the benefit estimate, the harm estimate, or both, and guidance explicitly warns against collapsing benefit and harm into a single combined point-estimate comparison that discards this separately tracked uncertainty3. A treatment whose benefit interval is narrow and clearly favorable, paired with a harm interval that is wide and includes a seriously elevated risk, warrants a more cautious recommendation than the same point estimates would suggest if read without their intervals.
Unequal interpretive weight of benefit and harm intervals
Because the consequences of underestimating a harm are often more severe, in terms of patient safety, than the consequences of underestimating a benefit, methodological guidance for benefit-harm assessment generally recommends applying a more cautious standard to a harm interval that includes a serious adverse outcome than to an equivalently wide benefit interval1. In practice, this means a wide confidence interval around a harm estimate that cannot rule out a clinically serious adverse effect should weigh more heavily against a favorable recommendation than a symmetrically wide confidence interval around a benefit estimate weighs in favor of one, an asymmetry in how the two intervals should be read that a simple side-by-side comparison of point estimates does not capture.
A documented gap in the reporting of harm outcomes
Zorzela and colleagues' systematic review of adverse-event reporting quality across published systematic reviews found widespread deficiencies in how harm outcomes, and the uncertainty around them, were presented compared with benefit outcomes drawn from the same bodies of evidence4. This has direct implications for the interpretation of a harm interval: if the underlying systematic review has not consistently or completely reported adverse events across its included trials, the resulting harm confidence interval is likely to understate the true uncertainty even further than the statistical calculation alone would suggest, because it is built on an already incomplete count of events4. A reader comparing benefit and harm intervals should therefore also check how systematically the harm outcome was ascertained and reported across the included studies, not only the width of the resulting interval.
A sequence for interpreting benefit and harm evidence together
A defensible practical sequence for weighing a harm-benefit trade-off starts with the benefit interval, establishing the plausible range and clinical importance of the intended effect, then moves to the harm interval, checking specifically whether it can rule out a seriously elevated risk rather than only looking at its point estimate, and finally considers how completely the harm outcome itself was measured and reported across the underlying studies24. Only after all three steps have been completed does a side-by-side comparison of the two point estimates become a reasonable final check, rather than the first and only step in the assessment, which is the order in which the comparison is most often made in practice and the order that GRADE’s evidence-to-recommendation process is designed to avoid.
Expressing the harm interval in absolute, patient-relevant terms
A relative-risk confidence interval for a harm outcome can be difficult for a patient or non-specialist clinician to translate into a practical sense of risk, which is why guidance recommends converting it into an absolute measure such as the number needed to harm (NNH), the number of patients who would need to be treated for one additional patient to experience the adverse outcome, alongside its own corresponding confidence interval1. Because NNH is a transformation of the same underlying risk estimate, it inherits a confidence interval exactly as wide, in relative terms, as the original risk measure’s interval, so a wide relative-risk interval for a harm outcome produces an equally wide, and sometimes clinically very different, range of NNH values; a point estimate of “1 in 200 patients harmed” can correspond to an interval running from a reassuring “1 in 2,000” to a concerning “1 in 20,” and both ends of that range should be stated rather than only the single central figure.
A worked example of harm and benefit
In a hypothetical example, a drug is shown, with a narrow and favorable confidence interval, to reduce the absolute risk of a serious complication by 4 percentage points, a clear, well-estimated benefit. The safety data from the same trial show a rare but serious adverse event occurring in 3 of 500 treated patients versus 1 of 500 controls, a relative risk with a 95% confidence interval spanning from a trivial increase to a more than fivefold increase in risk, because so few events were observed in either group2. Reporting only the point estimates, a clear 4-point benefit against a modest-looking tripling of a rare event, understates the uncertainty on the harm side of this trade-off; a decision-maker reading both full intervals, and not merely both point estimates, would reasonably seek more safety data before adopting the drug widely, even though the benefit alone appears well established. Presenting the intervals for both outcomes together, on a comparable absolute scale wherever possible, allows a clinician or guideline panel to make this kind of judgment transparently, rather than having it made implicitly, and invisibly, by whichever outcome’s point estimate was emphasized more prominently in a trial’s abstract.
Expressing this worked example as a number needed to harm clarifies the implications further for a clinician explaining the trade-off to a patient. A relative risk of roughly 3, with a point estimate corresponding to 3 events in 500 treated patients, translates to an absolute risk increase of about 0.4 percentage points, or an NNH of roughly 250; but because the confidence interval around that relative risk spans from a trivial increase to more than fivefold, the corresponding NNH interval spans from a reassuring value in the low thousands down to a concerning value well under 1001. A patient told only “roughly 1 in 250 patients like you experience this side effect” is being given a single number that could, with equal statistical justification given the current data, be closer to 1 in 2,000 or closer to 1 in 70, and a shared decision-making conversation that accounts for both ends of that range differs considerably from one anchored to the single central figure alone2. For this reason, Cochrane and GRADE guidance both encourage reviewers to carry the full interval of a harm outcome, not only its point estimate, through to any plain-language summary intended for patients and clinicians, rather than allowing the translation into absolute, everyday terms to obscure the uncertainty that the statistical analysis was designed to quantify3.
Key points
Key points
Assessment of a treatment’s harm-benefit trade-off requires interpretation of the confidence intervals around both outcomes, not merely a comparison of point estimates. A harm interval can include a serious adverse effect despite a reassuring point estimate, harm outcomes are frequently reported less completely than benefit outcomes in the same evidence base, and GRADE guidance treats the width of the interval, for both benefit and harm, as central to the overall recommendation.
Practical application
Practical application
- Confidence intervals for harm outcomes should be reported with the same prominence as those for benefit outcomes. The interval for a harm outcome should not be given less prominence than the interval for the benefit outcome in the same report.
- A harm interval should be checked for inclusion of a seriously elevated risk, even when the point estimate appears small. A wide interval on a harm outcome that includes serious risk should weigh against a favorable recommendation even when the point estimate alone appears acceptable.
- Certainty assessments for benefit and harm outcomes should be explicitly separated. The GRADE structure of assessing benefit and harm imprecision separately should be followed, rather than combining both into a single judgment that discards the individual intervals.
Frequently asked questions
Trials are typically sized (powered) around the primary efficacy outcome, and serious adverse events are often rarer, so fewer events accumulate for the harm analysis, which necessarily produces a wider confidence interval around the harm estimate than around the better-powered benefit estimate.
Guidance generally recommends weighting them asymmetrically when the potential harm is serious, since the clinical cost of underestimating a serious risk is usually higher than the cost of underestimating a comparably sized benefit; this asymmetry is a judgment frameworks like GRADE build into the overall recommendation process rather than treating benefit and harm as interchangeable.
References
- Deeks JJ, Higgins JPT, Altman DG (editors). Chapter 10: Analysing data and undertaking meta-analyses. In: Cochrane Handbook for Systematic Reviews of Interventions. Cochrane, 2024. read 10 October 2026. View on Cochrane Training
- Guyatt GH, Oxman AD, Kunz R, et al. GRADE guidelines 6. Rating the quality of evidence - imprecision. Journal of Clinical Epidemiology. 2011;64(12):1283-1293. read 10 October 2026. View on Journal of Clinical Epidemiology
- Guyatt GH, Oxman AD, Vist G, et al. GRADE guidelines: 4. Rating the quality of evidence - study limitations (risk of bias). Journal of Clinical Epidemiology. 2011;64(4):407-415. read 10 October 2026. View on Journal of Clinical Epidemiology
- Zorzela L, Golder S, Liu Y, Pilkington K, Hartling L, Joffe A, Loke Y, Vohra S. Quality of reporting in systematic reviews of adverse events: systematic review. BMJ. 2014;348:f7668. read 10 October 2026. View on The BMJ
- Gardner MJ, Altman DG. Confidence intervals rather than P values: estimation rather than hypothesis testing. BMJ. 1986;292(6522):746-750. read 10 October 2026. View on The BMJ
- Greenland S, Senn SJ, Rothman KJ, Carlin JB, Poole C, Goodman SN, Altman DG. Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. European Journal of Epidemiology. 2016;31(4):337-350. read 10 October 2026. View on European Journal of Epidemiology
- Schulz KF, Grimes DA. Sample size calculations in randomised trials: mandatory and mystical. The Lancet. 2005;365(9467):1348-1353. read 10 October 2026. View on The Lancet
- du Prel JB, Hommel G, Rohrig B, Blettner M. Confidence interval or p-value? Part 4 of a series on evaluation of scientific publications. Deutsches Arzteblatt International. 2009;106(19):335-339. read 10 October 2026. View on Deutsches Arzteblatt International
- Riley RD, Higgins JPT, Deeks JJ. Interpretation of random effects meta-analyses. BMJ. 2011;342:d549. read 10 October 2026. View on The BMJ
- Wasserstein RL, Lazar NA. The ASA Statement on p-Values: Context, Process, and Purpose. The American Statistician. 2016;70(2):129-133. read 10 October 2026. View on Taylor & Francis Online
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 10). Confidence Intervals in the Assessment of Harm and Benefit: GRADE and Absolute Effects. Pubrica. https://pubrica.com/academy/confidence-intervals-in-the-analysis-of-harm-and-benefits/
Pubrica Academy. “Confidence Intervals in the Assessment of Harm and Benefit: GRADE and Absolute Effects.” Pubrica, 10 Oct. 2026, pubrica.com/academy/confidence-intervals-in-the-analysis-of-harm-and-benefits/.
Pubrica Academy (2026) Confidence Intervals in the Assessment of Harm and Benefit: GRADE and Absolute Effects. Pubrica. Available at: https://pubrica.com/academy/confidence-intervals-in-the-analysis-of-harm-and-benefits/ (Accessed: 15 October 2026).
Pubrica Academy. “Confidence Intervals in the Assessment of Harm and Benefit: GRADE and Absolute Effects.” Pubrica. October 10, 2026. https://pubrica.com/academy/confidence-intervals-in-the-analysis-of-harm-and-benefits/
Pubrica Academy, ‘Confidence Intervals in the Assessment of Harm and Benefit: GRADE and Absolute Effects’ (Pubrica, 10 October 2026) <https://pubrica.com/academy/confidence-intervals-in-the-analysis-of-harm-and-benefits/> accessed 15 October 2026.
Pubrica Academy. Confidence Intervals in the Assessment of Harm and Benefit: GRADE and Absolute Effects. [Internet]. Pubrica; 2026 Oct 10 [cited 2026 Oct 15]. Available from: https://pubrica.com/academy/confidence-intervals-in-the-analysis-of-harm-and-benefits/
Pubrica Academy, “Confidence Intervals in the Assessment of Harm and Benefit: GRADE and Absolute Effects,” Pubrica, Oct. 10, 2026. [Online]. Available: https://pubrica.com/academy/confidence-intervals-in-the-analysis-of-harm-and-benefits/
@misc{/la_confidence_intervals_harm_versus_benefit_2026,
author = {{Pubrica Academy}},
title = {Confidence Intervals in the Assessment of Harm and Benefit: GRADE and Absolute Effects},
howpublished = {Pubrica},
year = {2026},
month = {oct},
url = {https://pubrica.com/academy/confidence-intervals-in-the-analysis-of-harm-and-benefits/},
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.
Harm-Benefit Interval Comparison Template
This tool is intended for comparing benefit and harm outcomes using their full confidence intervals rather than point estimates alone.
[ ] Reported the confidence interval for every primary benefit outcome [ ] Reported the confidence interval for every primary harm outcome with equal prominence [ ] Checked whether the harm interval includes a seriously elevated risk, regardless of the point estimate [ ] Assessed benefit and harm imprecision separately rather than as one combined figure [ ] Applied a more cautious reading to a wide, serious-risk-including harm interval than to a similarly wide benefit interval
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 clinical evidence review team compares benefit and harm confidence intervals side by side, identifying cases in which a reassuring point estimate conceals a harm interval that cannot exclude serious risk.