High, moderate, low or very low: reading certainty in a quit-smoking review

01

What the four labels say

In the current GRADE overview, high means strong confidence that the true effect lies in the specified range or on the specified side of a threshold. Moderate means moderate confidence, with a possible departure into another range. Low means limited confidence, with a departure into another range likely. Very low means very little confidence, with a substantial departure from the assessed range probable. The review must make clear what conclusion that range or threshold concerns; the reader is not asked to invent a medical cut-off.

These categories describe uncertainty in the evidence, not four success percentages. A small difference, little or no important difference, or an unwanted effect can have high-certainty evidence. A promising-looking large estimate can have very low certainty. Read direction and magnitude in their own columns.

[2]

02

Try one invented findings row

The table below is wholly fictional, with no real study or product behind it. It compares a leaflet plus optional reminders with the same leaflet alone in adults, using a defined six-month abstinence outcome. The imaginary review supplies the moderate rating; this page does not calculate it from the numbers.

The central estimate is 30 more per 1,000, but the displayed range runs from 10 fewer to 70 more. Its invented footnote says the range leaves important uncertainty about the difference and was a reason for downgrading. Preserve the range and explanation, rather than rewriting the row as ‘reminders work with moderate certainty’.

A separate row about unwanted effects would need its own data and certainty judgment. Nothing in this quitting row answers that question or predicts a reader's result.

03

Follow the explanation, not a homemade score

Cochrane describes five reasons that may lower certainty: bias in study conduct or reporting; unexplained differences across results; evidence not directly matching the question; an estimate too imprecise for the relevant conclusion; and missing studies or results that may distort the evidence. The review should explain which concerns mattered for this outcome.

These are judgment domains, not a reader's five-point subtraction exercise. Severity, the body of evidence and the question matter; related concerns should not simply be counted twice. If a footnote is absent, record that the reason is not explained in the material read, rather than guessing it. A study-design label alone does not settle the final rating.

[2][3]

04

A certainty label does not issue a recommendation

GRADE separates evidence certainty from recommendation strength. Recommendations also weigh benefits against harms and burdens, preferences, resources and context. Even a clearly estimated effect does not by itself tell one person which support or treatment to use.

For a real review, keep the comparison, outcome definition, follow-up, estimate and range, certainty, its explanations and evidence-search date together. Discuss personal decisions with a qualified local clinician or cessation professional; in England, NHS quit-smoking information provides support routes. Those service arrangements are not universal to English-speaking countries.

[2][4]

05

What to keep in mind

• Keep the label attached to its specific outcome and comparison.
• Read the estimate and its uncertainty separately from certainty.
• Keep the review's explanations instead of inventing a grade.

06

Common questions