GRADE Summary of Findings Table: Building and Interpreting One

A GRADE “Summary of findings” (SoF) table is the single page a guideline panel, journal reader, or policymaker actually reads to decide whether to trust a systematic review’s results. It packages the effect estimate for each key outcome, how many participants and studies it rests on, and a certainty rating — high, moderate, low, or very low — into one standardized layout, instead of leaving a reader to reconstruct that picture from a results section and a separate risk-of-bias assessment. Cochrane Handbook Chapter 14, “Completing ‘Summary of findings’ tables and grading the certainty of the evidence,” is the reference specification; CASRAI’s Handbook chapter guide covers where this fits among the Handbook’s other 25 chapters.

What the table actually contains

A standard SoF table opens with a header block naming the population, setting, intervention, and comparison the whole table applies to, then lists, per outcome, six components:

• Outcomes — the critical and important outcomes identified in the review’s protocol, listed in order of importance, capped at a maximum of seven per table. A review that tracked twelve outcomes still only tables the seven that matter most for the decision at hand; the rest stay in the full results.
• Illustrative comparative risks — the assumed risk in the comparator group alongside the corresponding risk with the intervention, usually expressed per 1,000 people, so a reader sees an absolute difference rather than only a ratio.
• Relative effect (95% CI) — typically a risk ratio, odds ratio, or hazard ratio pooled from the underlying meta-analysis.
• Number of participants (studies) — how many participants were assessed and how many contributing studies fed that specific outcome row; this can differ outcome-to-outcome within the same review when not every included study reported every outcome.
• Certainty of the evidence (GRADE) — the four-level rating (high, moderate, low, very low), assessed separately per outcome across five domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias. See CASRAI’s GRADE dictionary entry for how each domain triggers a downgrade.
• Comments — a short field for flagging anything a bare number would hide: a wide confidence interval crossing the line of no effect, an outcome measured only in a subgroup, or a caveat about applicability.

Below the table itself sits an explanations section: numbered footnotes that justify every certainty downgrade shown in the ratings column, so a reader can trace exactly why an outcome landed at “low” rather than “high” instead of taking the rating on faith.

Building one with GRADEpro GDT

GRADEpro GDT (Guideline Development Tool) is the software the GRADE Working Group and Cochrane both point authors to for this — it is not something built by hand in a word processor once a review has more than a handful of outcomes. Its role in the workflow: you enter, per outcome, the comparator group’s baseline (assumed) risk and the pooled relative effect from your meta-analysis, and it calculates the corresponding absolute risk with the intervention automatically — the arithmetic that turns “risk ratio 0.72” into “27 fewer per 1,000” is exactly what the tool exists to get right consistently. It then walks the assessment through the five GRADE domains per outcome, converts the resulting judgments into the certainty column, and exports a publication-ready table (or, for Cochrane reviews specifically, one that links directly into RevMan). Three practical points worth knowing before starting one:

• Pick outcomes before opening the tool, not while inside it. The seven-outcome cap is a decision about what matters to the reader, not a technical limit — deciding it mid-build tends to produce a table that reflects what data happened to be easiest to enter rather than what a decision-maker actually needs to see.
• A downgrade needs a stated reason. GRADEpro will let a certainty rating be entered without a supporting footnote, but Handbook guidance treats an unexplained downgrade as incomplete — every rating below “high” should trace to a specific, named concern in one of the five domains.
• Newer table variants exist alongside the original seven-column layout: a v2 that reorders columns and drops the comments field, a v3 that adds an explicit “difference” column for the absolute effect size, and a v4 that further splits the outcome and participant-count columns and reports numeric data in absolute rather than percentage terms. Which variant a given journal or guideline body expects varies — check target formatting before building rather than after.

Reading one as a consumer, not just building one

Most people who touch a SoF table are reading one someone else built, not authoring it — a research administrator screening evidence for an institutional policy, a clinician deciding how much weight to put on a recommendation, an editor checking that a submitted review’s claims match its own evidence table. Three reading habits catch most of what a rushed read misses:

• Check the comparator’s baseline risk before trusting an absolute-difference number. A relative-risk reduction of 30% means very different things in absolute terms depending on whether the baseline risk is 2 per 1,000 or 200 per 1,000 — the illustrative-risks column exists specifically so this doesn’t have to be reconstructed by hand.
• Read the certainty column before the effect estimate, not after. A “very low certainty” rating on a striking effect size is a signal to treat the number as provisional, not a footnote to skim past once the headline number has already anchored the reader’s impression.
• Open the explanations section for any outcome the table is being used to justify a real decision on. The footnote will usually say which GRADE domain drove the downgrade — a downgrade for imprecision (a wide, uncertain confidence interval) supports a very different response than one for risk of bias (flawed included studies).

How this differs from an Evidence-to-Decision framework

A SoF table answers “how much should we trust this effect estimate,” full stop — it does not, by itself, tell a panel what to recommend. CASRAI’s guide to the GRADE Evidence-to-Decision (EtD) framework covers the separate, later step where a guideline panel takes the certainty ratings from a SoF table as one input among several (alongside values, balance of effects, resource use, feasibility) to reach an actual recommendation. Confusing the two is a common mistake: a high-certainty SoF table does not automatically produce a strong recommendation, since a panel can still weigh costs, feasibility, or patient-value variability heavily enough to recommend conditionally even against strong evidence.

Frequently asked questions

Is a Summary of Findings table required for every systematic review?

Cochrane requires one for every Cochrane Intervention review with at least one included study; outside Cochrane, requirements vary by journal and by whether the review is being used to inform a formal guideline, where GRADE-rated evidence tables are close to universal expectation.

Can a SoF table be built without GRADEpro GDT?

Yes — the layout and rating logic are a published methodology, not software-locked — but GRADEpro is the tool the Cochrane Handbook and the GRADE Working Group both point authors toward, and doing the absolute-risk calculations and downgrade bookkeeping by hand across seven outcomes is exactly the kind of arithmetic GRADEpro exists to make less error-prone.

Why does a SoF table cap at seven outcomes when a review measured more?

The cap forces the review team to decide, explicitly and in advance, which outcomes actually matter for the decision the table serves — a design choice, not a database limitation. The remaining outcomes stay reported in the review’s full results; they are simply not promoted to the one-page summary.

What does it mean when the “relative effect” and “illustrative risks” columns seem to disagree?

They should not disagree once the arithmetic is followed through correctly — the illustrative absolute risks are calculated directly from the relative effect applied to the stated baseline risk. An apparent mismatch is usually a sign the baseline (comparator) risk used differs from what a reader assumed, which is exactly why that number is stated explicitly rather than left implicit.