What GRADE is and who maintains it
GRADE, Grading of Recommendations Assessment, Development and Evaluation, is a framework for rating the certainty of a body of evidence for a specific outcome and, separately, the strength of any recommendation based on it. It was developed by an international working group of clinicians, methodologists and guideline developers and introduced in a series of papers beginning in 2008 (Guyatt et al., 2008).
GRADE rates certainty on a four-level scale, from high to very low, starting from the study design (trials start high, observational studies start low) and then considering factors that can move the rating up or down: risk of bias, inconsistency across studies, indirectness of the evidence to the question being asked, imprecision, and publication bias on the downgrading side, and a large effect, a dose-response relationship, or plausible confounding in the opposite direction on the upgrading side.
Each level has a plain-language meaning that GRADE asks authors to use consistently: high certainty means further research is very unlikely to change confidence in the estimate; moderate certainty means further research may change it; low certainty means further research is likely to change it; and very low certainty means any estimate is highly uncertain. Using these fixed phrases, rather than inventing new wording each time, is part of what makes GRADE ratings comparable across different reviews and guidelines.
Why it exists
GRADE also makes a point of insisting that certainty is assessed per outcome rather than per study or per review as a whole, because the same body of studies can support very different levels of confidence depending on which outcome is being asked about. A set of trials might give high-certainty evidence that a drug lowers blood pressure, measured directly and precisely in every trial, while giving only low-certainty evidence about whether it reduces strokes, an outcome measured less often and with wider confidence intervals.
Before GRADE, different guideline panels and systematic reviews used different, often unexplained, systems for labelling evidence "strong" or "weak", which made it hard for a reader to compare how much confidence two different recommendations actually deserved, or to understand why one body of evidence was rated lower than another. A recommendation labelled "strongly recommended" by one group could rest on evidence a GRADE assessment would rate as low certainty, without a reader having any way to tell.
GRADE exists to make the reasoning behind a certainty rating explicit and consistent, so that a reader can see exactly why the evidence for one outcome was rated differently from another, and so that disagreements about a rating can be resolved by discussing a specific, named factor rather than a vague overall impression. In this respect GRADE shares its motivation with the EQUATOR Network's wider project of coordinating reporting standards, which was set up because inconsistent and incomplete reporting across many kinds of health research made evidence harder to judge and compare (Moher et al., 2008).
Which tasks it fits
GRADE fits any task that needs to summarise how much confidence to place in evidence for a specific outcome, most commonly within a systematic review or a clinical practice guideline, but also within a doctoral thesis that needs to characterise the strength of the literature supporting a claim.
| Task | How GRADE applies |
|---|---|
| Systematic review (with or without meta-analysis) | Rate certainty separately for each outcome reported, usually summarised in a Summary of Findings table, alongside PRISMA reporting of the review process itself (Page et al., 2021) |
| Clinical practice guideline | Combine certainty ratings with values, preferences and resource considerations to grade the strength of each recommendation |
| Doctoral literature review | Characterise how certain the existing evidence is for the specific claim the thesis relies on, rather than describing the literature only descriptively |
| Health technology assessment | Rate certainty for each outcome considered in a cost-effectiveness or policy analysis |
Where GRADE does not apply
GRADE is a rating framework, not a reporting guideline, and it does not replace PRISMA, STROBE or other guidelines for describing how a review or study was conducted; a review needs to report its methods fully under PRISMA (Page et al., 2021) and can then, separately, rate the certainty of its findings under GRADE. Using GRADE is not a substitute for reporting a complete search strategy or risk of bias assessment.
GRADE also does not generate statistical estimates on its own; it is applied after the effect estimate and its precision are already known, drawing on methods such as those behind diagnostic accuracy estimates reported under STARD (Bossuyt et al., 2015) or observational associations reported under STROBE (von Elm et al., 2007), and on trial estimates reported under CONSORT (Hopewell et al., 2025). It is not a replacement for a formal risk-of-bias tool applied to individual studies, which feeds into, rather than being replaced by, the GRADE rating.
Why it matters
A conclusion presented without any indication of how certain the underlying evidence is invites a reader to treat every finding as equally solid, which is rarely true. GRADE's explicit, factor-by-factor structure means a reader can see, for example, that a finding is rated low certainty because the two available trials disagreed with each other (inconsistency) rather than because the question was poorly studied overall, which has very different implications for what further research is needed. This kind of explicit reasoning is part of why GRADE has been adopted by a large number of international guideline organisations and systematic review groups since 2008.
How it is applied in practice
A useful habit for a doctoral researcher is to build the Summary of Findings table early, even in draft form, as soon as the key outcomes and their effect estimates are known, rather than treating the certainty rating as a final step tacked on after the discussion is already written.
In practice, GRADE is applied outcome by outcome, usually after the effect estimates for each outcome have been calculated, by considering each of the five downgrading factors and the three upgrading factors against the available studies and documenting the reasoning for any change from the starting point. The result is typically presented in a Summary of Findings table, showing the effect estimate, its precision, the certainty rating and a plain-language statement of what the rating means for interpreting the finding.
A common point of confusion worth addressing directly: GRADE's starting point for trial evidence is high certainty, not automatic certainty. A single small trial with a wide confidence interval and no blinding can still end up rated low or very low certainty once the relevant downgrading factors are applied, despite being a randomised design; the starting point only sets where the rating begins, not where it ends.
Guideline-development organisations worldwide, and the Cochrane Collaboration for its systematic reviews, use GRADE as their standard approach to rating evidence certainty, which means a doctoral researcher summarising the literature in a field where Cochrane or a major guideline body has already published a GRADE-rated review can draw directly on that published rating rather than repeating the assessment from scratch, citing it as the source of the certainty judgement.
An illustrative example
The passage below is a hypothetical illustration, not a quotation from a published review.
| Element | Without GRADE | With GRADE |
|---|---|---|
| Overall claim | "The evidence shows the intervention reduces symptoms." | "Moderate-certainty evidence shows the intervention probably reduces symptom severity (standardised mean difference 0.45, 95% CI 0.20-0.70); certainty was downgraded one level for risk of bias in two of the four included trials." |
| Secondary outcome | "Side effects were uncommon." | "Low-certainty evidence suggests the intervention may increase mild gastrointestinal side effects (risk ratio 1.8, 95% CI 0.9-3.5); certainty was downgraded for imprecision (wide confidence interval crossing no effect) and risk of bias." |
| Reasoning | Not given | "Reasons for each downgrade are listed in a footnote to the Summary of Findings table, naming the specific studies and the specific concern." |
| Recommendation | "This treatment is strongly recommended." | "A conditional recommendation is made, reflecting moderate-certainty evidence of benefit balanced against low-certainty evidence of harm and variation in how patients are likely to value the trade-off." |
What to check before you finalise a rating
- Have you rated certainty separately for each outcome, rather than giving one rating for the whole body of evidence?
- Have you considered and documented each of the five downgrading factors (risk of bias, inconsistency, indirectness, imprecision, publication bias)?
- For observational evidence, have you considered whether any upgrading factor (large effect, dose-response, opposing plausible confounding) applies?
- Is the reasoning for each change from the starting certainty level stated, not just the final rating?
- Have you kept certainty of evidence and strength of recommendation clearly separate?
- Official source: GRADE Working Group (gradeworkinggroup.org) .
Key takeaways
- GRADE rates the certainty of evidence for a specific outcome, and separately the strength of any recommendation (Guyatt et al., 2008).
- It exists because unexplained, inconsistent certainty labels made it hard to compare how much confidence different recommendations deserved.
- Trial evidence starts high and can be downgraded; observational evidence starts low and can sometimes be upgraded.
- GRADE is a rating framework applied after a review's methods are reported under a guideline such as PRISMA (Page et al., 2021) or STROBE (von Elm et al., 2007), not a substitute for them.
- A GRADE rating always names its reasoning, so a reader can see exactly why a certainty level changed.
