GRADE Certainty in Nursing Evidence: Read the Outcome, Footnote, and Effect Together

Read the rating as part of an outcome, not as a badge for the paper

A review can provide moderate-certainty evidence for one outcome and very uncertain or absent evidence for another. Calling the whole paper 'moderate quality' loses that distinction. Start with the exact outcome, comparison, population, and follow-up attached to the rating.

Cochrane's GRADE guidance describes four certainty levels and assesses a body of evidence for a specified outcome. The GRADE Working Group also separates certainty from the strength of a recommendation. A certainty label does not by itself tell a nurse, student, or guideline panel what action to take.

This guide is about reading an existing assessment accurately. It uses an invented review extract to show how to move from an outcome row and its footnotes to a defensible assignment paragraph. The effects, study counts, and supplied ratings below are fictional teaching material. They are not an actual review, a new GRADE assessment, or evidence that an intervention works.

Collect the table header before interpreting a single row

The fictional review asks whether a structured discharge explanation with teach-back, compared with the ward's usual explanation, changes outcomes for adults discharged from general medical wards. Its readmission outcome is all-cause hospital readmission within 30 days. The extract is intentionally specific: it does not concern children, specialist surgical follow-up, or a telephone-only intervention.

The mock review supplies a risk ratio of 0.80 with a 95% confidence interval from 0.55 to 1.16 for readmission, based on three randomized trials and 600 participants. It labels the evidence moderate certainty and links that row to footnote a. Treat these values as facts supplied by the teaching extract, not as results derived here from unprovided trial data.

The review also presents recall of discharge instructions at 48 hours, and states that caregiver burden was not reported. Those are separate outcomes. Better recall cannot be substituted for fewer readmissions, and the absence of reported burden cannot be presented as proof that the intervention creates no burden.

In a real review, read the table title, explanatory notes, relevant results, and evidence profile together. Record the review version and the exact table location. If the abstract and detailed table appear inconsistent, resolve the discrepancy before choosing the more favorable wording.

Follow the footnote back to the reason for uncertainty

In this teaching extract, footnote a says the review authors downgraded readmission evidence by one level for imprecision. They considered a reduction of 50 readmissions per 1,000 important for their stated assessment context; the interval spans that benefit, no difference, and increased readmissions. The extract reports no additional downgrading for the other assessed domains. This is a supplied explanation to interpret, not a shortcut for assigning the same rating to any similar-looking interval.

Footnote b says the recall evidence was downgraded for risk of bias in outcome measurement and for imprecision relative to the review's stated threshold of a 2-point improvement. Assessors knew the allocated group and used a judgment-dependent scoring process; the interval includes values below and above that threshold. The extract supplies two one-level downgrades and a low-certainty label. Statistical exclusion of zero alone does not erase those concerns.

Read the reason, not just the number of downgrades. Risk of bias concerns how the results may be distorted; imprecision concerns how much the estimate leaves unresolved. Inconsistency, indirectness, and publication bias address other questions. A short student summary should retain the important reason given by the review without claiming to have independently completed every domain assessment.

If a real table provides a rating but no usable explanation, say that the rationale could not be verified from the material available. Locate the evidence profile, supplement, or full methods. Do not invent a footnote, use journal prestige to fill the gap, or mechanically downgrade because a review includes only a few studies.

Translate the relative estimate without losing its baseline or interval

The fictional readmission row uses an illustrative comparator risk of 200 per 1,000 over 30 days. Applying the supplied risk ratio gives 200 × 0.80 = 160 per 1,000 with the intervention. The point estimate is therefore 40 fewer readmissions per 1,000, or a 4-percentage-point reduction, for that assumed baseline.

Apply the same calculation to the interval limits: 200 × 0.55 = 110 and 200 × 1.16 = 232 per 1,000. Compared with 200 per 1,000, this ranges from 90 fewer to 32 more. That range explains why 'the intervention reduces readmission by 20%' is incomplete: it gives the relative point estimate while hiding the baseline, uncertainty, and possibility of an increase.

These absolute values are calculated illustrations, not additional observed patient counts. They hold the comparator risk fixed and do not incorporate uncertainty in that assumed risk. If the relevant population has a different baseline, the corresponding absolute difference may change; applicability and transport of the relative effect also require consideration.

Do not turn the reciprocal of this uncertain point-estimate difference into a confident treatment promise. The immediate reading task is to preserve the outcome, period, effect, interval, and certainty reason. A clinical recommendation requires a separate decision process using the relevant benefits, harms, preferences, feasibility, and context.

Keep certainty, effect size, and recommendation separate

The readmission row has a point estimate favoring the intervention, a wide interval, and a moderate-certainty label supplied by the mock review. Each item answers a different part of the reading task. The label does not shrink the confidence interval, and the favorable point estimate does not remove the reason for downgrading.

The recall row has a positive mean difference and low certainty. It would be wrong to write either 'teach-back definitely improves outcomes' or 'low certainty means there is no improvement.' The supplied evidence suggests a difference in this measured recall score, with the stated limitations. It does not establish the effect on all patient-important outcomes.

The GRADE Working Group's criteria for recommendations include explicit judgments about the balance of effects and other decision factors. Thus, high certainty is not a synonym for a strong recommendation, and low certainty is not a synonym for a recommendation against an intervention. When an assignment asks for the guideline recommendation, retrieve and cite that recommendation rather than manufacture one from a review's certainty column.

If you need to perform a new GRADE assessment, that is a different task from this reading exercise. It requires the full body of evidence, an appropriate systematic process, explicit outcome-level judgments, and documented reasoning. A checklist completed from an abstract is not a substitute. State clearly whether you are reporting the authors' rating or making a supported assessment yourself.

Write the evidence paragraph with its limits intact

A weak paragraph would say: 'Three trials prove that teach-back reduces readmissions by 20%. The review is moderate quality, so hospitals should use it.' That sentence changes a supplied estimate into proof, assigns the outcome rating to the whole review, removes the interval, and invents a practice recommendation.

A more faithful paragraph for the fictional extract is: 'For adults discharged from general medical wards, the mock review compared a structured explanation with teach-back against usual explanation. For all-cause readmission within 30 days, it reported RR 0.80 (95% CI 0.55–1.16) from three randomized trials with 600 participants. At its illustrative comparator risk of 200 per 1,000, this corresponds to 40 fewer readmissions per 1,000, with an interval from 90 fewer to 32 more. The supplied moderate-certainty rating reflects one-level downgrading for imprecision. The estimate therefore leaves uncertainty about the size and direction of the effect; it does not independently establish a practice recommendation.'

A separate sentence can address the other outcome: 'The extract reported a 2-point higher 48-hour recall score on a 0–10 scale, but rated that evidence low certainty because of outcome-measurement bias concerns and imprecision relative to its importance threshold. Caregiver burden was not reported.' Keeping these sentences separate stops a favorable educational measure from taking over the readmission conclusion.

For a real assignment, replace the mock details with the actual review's data and cite the table and footnotes you used. Preserve the authors' rating and rationale accurately, then identify any unresolved issue separately. Do not present this worked paragraph as evidence about a real intervention.

Use a reading worksheet before drafting your own summary

The worksheet below can be used without assigning a new certainty rating. Fill it from the source, then write a paragraph that keeps the important fields together. When a field is unavailable, mark it unavailable and identify the material needed to resolve it.

A finished worksheet for the readmission example would identify the general-medical-ward population, teach-back comparison, 30-day readmission outcome, three trials and 600 participants, RR 0.80 with its interval, the assumed 200-per-1,000 baseline, moderate certainty, and the imprecision footnote. The unresolved decision would be whether the evidence and wider context support a specific practice recommendation; the rating alone does not answer it.

Before submitting, ask whether a reader could locate the same row and reconstruct your statement. If you have changed the outcome, follow-up, scale, direction, or certainty rationale, return to the source. Accurate evidence writing depends on keeping those connections visible, especially when the result is uncertain.