Outcomes are not interchangeable, and which ones a recommendation rests on is a decision made before the evidence is read, then revisited after. GRADE rates every candidate outcome into three ordinal categories: critical and important outcomes both bear on the recommendation, while only critical ones set the overall certainty rating. (Schünemann et al., n.d.)
The three categories
- critical — the primary factors influencing a recommendation
- important but not critical
- of limited importance — “in most situations” will not bear on the recommendation
An optional 1-9 numeric aid maps 7-9 critical, 4-6 important, 1-3 limited — used “to distinguish between importance categories,” not as a cardinal weight. Only outcomes rated critical determine the overall certainty of evidence supporting a recommendation; critical and important outcomes both appear in the evidence tables. (Schünemann et al., n.d.)
Mechanism — importance is chosen by what matters, NOT by what was measured
The load-bearing rule, and the one most often violated in practice:
“Guideline developers must base the choice of outcomes on what is important, not on what outcomes are measured and for which evidence is available. If evidence is lacking for an important outcome, this should be acknowledged, rather than ignoring the outcome.” (Schünemann et al., n.d.)
Operationalized by the empty row: an outcome the panel selected but for which no evidence exists still gets a row in the evidence profile, and “an empty row in an evidence profile can be informative in that it identifies research gaps.” (Schünemann et al., n.d.)
This is an explicit anti-streetlight device — it prevents the outcome set from being silently redefined as whatever the literature happened to measure, which is the mechanism by which a body of evidence comes to look more complete than it is.
Perspective decides the weighting
Importance “is likely to vary within and across cultures or when considered from the perspective of the target population, clinicians or policy-makers,” so a panel must declare whose perspective it is taking; where the audience is clinicians and their patients, “the perspective would generally be that of the patient.” (Schünemann et al., n.d.) Where evidence on values and preferences is absent, “panel members should use their prior experiences with the target population to assume the relevant values and preferences” (Schünemann et al., n.d.) — a fallback the handbook elsewhere concedes is likely to be uncertain (Schünemann et al., n.d.).
Declaring perspective before rating importance is the same discipline as declaring role, stage, materials and objective before an appraisal (Standpoint (Role Stage Materials Objective)).
Importance is re-rated after the evidence arrives
The first classification happens at protocol stage; a reassessment follows once evidence is in, both to add outcomes the reviews surfaced and to re-weigh existing ones. GRADE’s worked case: screening for abdominal aortic aneurysm, where all-cause mortality starts as critical, the evidence establishes a cause-specific reduction without a demonstrable all-cause reduction, and all-cause mortality «ceases to be a critical outcome» because the cause-specific finding is judged sufficiently compelling. (Schünemann et al., n.d.)
- This is a live abuse surface, and GRADE’s guard is an invariance test. Demoting an outcome the evidence failed to move is how a disappointing result becomes a reframed success. The handbook’s own trigger for the demotion: “An outcome turns out to be not necessary if, across the range of possible effects of the intervention on that outcome, the recommendation and its strength would remain unchanged” — and such judgments “require careful consideration and are probably rare.” (Schünemann et al., n.d.) Note the subject: it is the OUTCOME that becomes unnecessary, not the re-rating. The sentence says when an outcome may be dropped from the critical set — it is not a licence to skip re-rating. Read the other way it becomes permissive rather than restrictive, which is the reverse of its role. That constrains motivated demotion without excluding it — the test asks whether the outcome could have mattered, not whether the demoter wanted it to. (inferred from Schünemann et al., n.d.)
A cause-specific benefit without a mortality benefit is not a null (2026-08-01)
The AAA case above has a frequent mirror: an intervention moves a non-fatal morbidity outcome while all-cause mortality stays null — SFA reduction cuts combined CV events (RR 0.83) with mortality null (RR 0.96) -> Does Reducing Saturated Fat Reduce Cardiovascular Events. This is not a null, and the events are not irrelevant: a non-fatal MI or stroke is itself a critical/important patient outcome (disability, lost function — surviving with impairment), so the lever buys healthspan (fewer disabling events) even where it does not extend lifespan — a distinct good, not a failure. Reading events down, mortality flat as does nothing collapses the outcome menu onto mortality alone, the exact error the three-category rating exists to prevent (critical AND important outcomes both bear on a recommendation). How much a disabling-but-non-fatal event weighs against length of life is then the person’s layer-3 call.
A composite person-centred outcome is the anti-streetlight device made an endpoint (2026-08-28)
The empty-row rule keeps an unmeasured important outcome visible; a composite person-centred functional score is the same defence built into the chosen endpoint. Zheng’s healthspan SR deliberately rejects the disease-centred definition (multimorbidity) for intrinsic capacity (a WHO 5-domain objective composite) plus quality of life (subjective), because both share «a person-centered focus on functioning rather than diagnoses», and argues «a composite IC score would serve as a better indicator… instead of single domains of IC (e.g., locomotion or cognitive domain alone)». (Zheng et al., 2026)
- Why it belongs here: choosing a composite function outcome is the outcome-selection act this page governs — it resists the streetlight collapse of “health” onto whatever single well-lit endpoint (a biomarker, one cognitive test, a disease event) a trial measured. The catch is that IC/QoL are also the outcomes measured worst (self-report, unblindable interventions, heterogeneous scales), so the composite buys menu-completeness at the cost of measurement noise — more honest uncertainty, not more confident advice. The construct and its hedged evidence live on Intrinsic Capacity and Multidimensional Healthspan. (inferred from Zheng et al., 2026)
Decision relevance
- Ordinal is enough. GRADE never converts importance into a cardinal utility, and its recommendations do not require one — which is what makes an elicited outcome menu workable without inventing a scalar maximand.
- Ask of any recommendation: which outcomes were rated critical, and by whose perspective? Two panels reading identical evidence can differ legitimately on both, and the difference is the source of the disagreement more often than the evidence is.
- A missing outcome is a finding. If harms, burden or quality-of-life do not appear in the outcome set, the recommendation was scoped to exclude them — GRADE notes that “many, if not most, systematic reviews fail to address some key outcomes, particularly harms.” (Schünemann et al., n.d.)
Importance is also a MAGNITUDE threshold, not only an outcome ranking (2026-07-26)
This page ranks which outcomes count. A second importance judgement sits downstream: how big a change on a chosen outcome counts as meaningful. Naude’s Cochrane review states its bar for each outcome and then reads every estimate against it — weight “about 4 to 6 kg… would start to become clinically meaningful”, DBP “greater than 2 mmHg”, LDL “greater than 0.26 mmol/L”, HbA1c “greater than 0.5%”. (Naude et al., 2022)
Why this matters for reading any review: without a stated threshold, “no significant difference” and no meaningful difference are indistinguishable, and so are “significant” and “worth doing.” With one, a small point estimate whose CI excludes the threshold in both directions becomes a positive finding of no meaningful effect — the telos’s third evidence state, properly evidenced rather than inferred from a null.
The audit question to carry: were the thresholds pre-specified in the protocol, or stated in the discussion after the estimates were in view? In this review the answer is split, and the split is the honest reading: the weight threshold is pre-stated — Background gives “weight loss of 5% and more… may result in clinically meaningful improvements”, and the Methods rationale for the 12-week minimum calls five-to-ten per cent of initial body weight “(clinically meaningful)”, so the Discussion’s “about 4 to 6 kg” is a translation of a criterion already on the record. The DBP, LDL and HbA1c thresholds appear only in the Discussion. Disclosed either way, and still not the same instrument as pre-specification.
When the values evidence actually exists — a worked instance (2026-08-28)
GRADE’s fallback when values-and-preferences evidence is missing is panel experience (§3.3), which the handbook concedes is likely to be uncertain. The falls-prevention NMA for the Canadian Task Force is a case where that evidence was systematically reviewed rather than assumed: a dedicated review (44 studies, mostly EQ-5D, mostly from people who experienced the event) put disutilities on the 0-to-1 (death) scale and produced an empirical outcome ranking for the falls cluster (Pillay et al., 2024).
- The finding: «Based on the much higher disutility, fracture (of any type) is probably more important than either falls (0.09 over 12 months) or functional status (0.12 for impairment in at least 1 ADL)» (MODERATE certainty), with long-term-care admission the most important state of all — «80% of participants said they would rather be dead» (Pillay et al., 2024).
- Why it matters for this page: it is the rare instance where which outcome is critical is settled by measured patient values instead of panel assertion — and the ranking is non-obvious (the fall, the measured endpoint of most trials, is the least important state; the fracture and lost independence it can trigger are what patients weight). This both instantiates the perspective rule (values are the patient’s, and here they were elicited) and doubles as a streetlight warning: a trial that counts falls is measuring the low-disutility link in the chain, not the high-disutility outcome the person cares about. The full worked chain lives on Exercise for Preventing Falls in Older Adults. (inferred from Pillay et al., 2024)
Limits
- The 1-9 scale invites false precision; GRADE uses it only to sort into three bins, and the bins are the operative object.
- The re-rating step’s only guard is the invariance test above, which is insensitive to motive.
- Source currency: §3 is flagged in-source as rewritten in the 2024 GRADE Book. (Schünemann et al., n.d.)