WHY ENGLISH?
A Summary of Findings table is short because the judgement has been organised, not because uncertainty has disappeared.
This guide helps students and evidence teams connect outcomes, effect estimates and GRADE certainty judgements to plain language that decision-makers can understand and verify.
GRADEpro supports evidence tables and certainty assessments used in systematic reviews and guidelines. A Summary of Findings table brings outcomes, effects, participant numbers and certainty into a compact view. English matters because every short label rests on a chain of judgement. Terms such as high, moderate, low and very low certainty describe confidence in an effect estimate, not whether an intervention is morally good or a study received a school grade.
People searching for GRADEpro, GRADE certainty of evidence or Summary of Findings tables often need help moving between statistics and decisions. Relative effects, absolute risks and confidence intervals answer different questions. Footnotes must explain why certainty changed. Plain-language summaries should remain faithful to magnitude, direction and uncertainty.
This is practical English for health, education and policy evidence. It does not provide clinical advice, replace the GRADE Handbook or turn software into a methodological authority. Current GRADEpro help checked on 4 November 2026 describes certainty domains including risk of bias, inconsistency, indirectness, imprecision and publication bias, and supports several evidence-table formats including Summary of Findings and interactive versions.
Start With One Decision Question and Its Outcomes
An evidence table is organised around a question, not around whichever statistics are easiest to copy.
Define population, intervention, comparator and outcomes. Select outcomes important to decision-makers, including benefits and harms. Keep outcome definitions, time points and measurement scales explicit.
Outcome example: Start With One Decision Question and Its Outcomes
For a school reading programme, comprehension at twelve weeks and withdrawal due to burden may both matter. Satisfaction alone cannot stand in for learning.
Outcome selection after seeing favourable results can bias the table.
Table move: Start With One Decision Question and Its Outcomes
List outcomes and decision importance before entering results; record any later change.
Know What Certainty Refers To
Certainty is assessed for a body of evidence and an outcome-specific effect estimate.
The GRADEpro certainty guide describes four levels from high to very low. The object is confidence that the true effect lies near the estimate, not confidence in software, authors or popularity.
Certainty example: Know What Certainty Refers To
One review can have moderate certainty for reading accuracy and very low certainty for long-term motivation because the evidence bodies differ.
Do not assign one blanket certainty grade to an entire intervention or review unless a method explicitly justifies that synthesis.
Table move: Know What Certainty Refers To
Complete the phrase ‘We have __ certainty in the estimate for __ at __ time point’ before writing any conclusion.
Separate Study Design From Final Certainty
Study design influences the starting framework but does not finish the judgement.
Randomised and non-randomised evidence can face risk of bias, inconsistency, indirectness, imprecision and publication bias. Some factors may increase certainty under defined circumstances. Follow current GRADE guidance rather than an oversimplified design hierarchy.
Design example: Separate Study Design From Final Certainty
Several randomised trials have serious attrition and inconsistent results. The final certainty can be lower than the design label suggests.
Do not call every observational result weak or every trial strong without domain assessment.
Table move: Separate Study Design From Final Certainty
Record the evidence base and each domain judgement with a reason tied to the outcome.
Explain Risk of Bias as Effect-Distorting Concern
Risk of bias asks whether features of included studies could systematically distort an effect estimate.
Use the appropriate appraisal tool, then translate study-level concerns into a body-of-evidence judgement. Explain which bias, its direction if known and why it matters to this outcome.
Bias example: Explain Risk of Bias as Effect-Distorting Concern
Unblinded outcome assessors rate a subjective writing rubric. The team downgrades and explains the plausible direction instead of writing only ‘serious risk of bias.’
A long list of study flaws is not automatically a reason to downgrade twice. Apply the framework consistently.
Table move: Explain Risk of Bias as Effect-Distorting Concern
Write a footnote that names the concern, affected evidence and consequence.
Describe Inconsistency Beyond the Word Heterogeneity
Inconsistency concerns unexplained differences in results across studies.
Examine effect direction, magnitude, confidence intervals and relevant statistical measures. Look for prespecified explanations such as population, delivery, dose or measurement differences.
Inconsistency example: Describe Inconsistency Beyond the Word Heterogeneity
Three studies show benefit and two show little effect. Differences align with programme duration, but subgroup evidence is sparse. The footnote states both pattern and residual uncertainty.
A high statistical heterogeneity value is not self-interpreting; few studies can also make estimates unstable.
Table move: Describe Inconsistency Beyond the Word Heterogeneity
State what varies, whether the variation is explained and how it affects confidence.
Use Indirectness to Name the Decision Gap
Evidence is indirect when it does not closely match the target question or depends on a surrogate link.
Compare population, intervention, comparator, outcome and setting. Explain the path from available evidence to the actual decision and where assumptions enter.
Indirectness example: Use Indirectness to Name the Decision Gap
University studies evaluate an app intended for lower-secondary pupils. The mechanism may transfer, but age, curriculum and supervision differ. The table identifies that gap.
Geographic difference alone is not automatically indirect; relevance depends on the mechanism and decision context.
Table move: Use Indirectness to Name the Decision Gap
Write ‘Evidence is indirect because __ differs from the target __, which may change __.’
Read Imprecision Through Decision-Relevant Bounds
Imprecision concerns how much the true effect could vary within the estimate’s uncertainty.
Examine confidence intervals, event numbers, sample size and thresholds that would change a decision. A point estimate can look impressive while compatible with trivial benefit or harm.
Imprecision example: Read Imprecision Through Decision-Relevant Bounds
An intervention appears to improve pass rates, but the interval includes almost no difference and a meaningful benefit. The plain-language conclusion emphasises uncertainty.
Do not interpret ‘not statistically significant’ as proof of no effect.
Table move: Read Imprecision Through Decision-Relevant Bounds
State the plausible range and whether it crosses a decision threshold.
Discuss Publication Bias as Missing-Evidence Risk
Publication bias concerns selective availability or reporting that can distort a body of evidence.
Look for small-study patterns, missing protocols, industry patterns, unpublished results and selective outcomes. GRADEpro lists publication bias among other considerations that can lower certainty.
Publication-bias example: Discuss Publication Bias as Missing-Evidence Risk
Several small positive studies are published, while registry entries show completed trials without results. The team explains why missing evidence could exaggerate benefit.
Funnel plots and suspicion are not automatic proof, especially with few studies.
Table move: Discuss Publication Bias as Missing-Evidence Risk
Record the signals, alternative explanations and reason for the final judgement.
Translate Relative Effects Into Absolute Effects
Decision-makers often understand absolute differences more readily than ratios.
Use an appropriate baseline risk and show how the relative effect changes expected events. Keep the time horizon and population visible. Explain that absolute effects change when baseline risk changes.
Absolute-effect example: Translate Relative Effects Into Absolute Effects
A relative risk reduction becomes ‘about 30 fewer events per 1,000 over one year’ under the stated baseline. Another setting with lower baseline risk would have a smaller absolute difference.
Never manufacture an absolute effect without a defensible baseline and interval.
Table move: Translate Relative Effects Into Absolute Effects
Display baseline, treated risk, absolute difference, interval and time frame together.
Write Footnotes as Compact Reasoning Records
A certainty symbol without an explanatory footnote hides the judgement.
A good footnote names evidence, problem, seriousness and consequence. It should let another reader understand why the rating changed without searching meeting minutes.
Footnote example: Write Footnotes as Compact Reasoning Records
‘Downgraded one level for serious imprecision: 95% CI includes no important effect and the prespecified benefit threshold’ is stronger than ‘imprecise.’
Footnotes should not become miniature essays or copy generic stock phrases.
Table move: Write Footnotes as Compact Reasoning Records
Use one reason per clause and link it to the relevant outcome and threshold.
Keep High and Low Certainty Separate From Large and Small Effect
Certainty and effect magnitude are different dimensions.
A small effect can have high certainty; a large estimate can have very low certainty. Decision importance also depends on harms, burden, cost, feasibility and values.
Magnitude example: Keep High and Low Certainty Separate From Large and Small Effect
Many precise trials show a small average improvement. A single tiny study shows a dramatic gain. The table does not let drama outrank confidence.
Do not call low-certainty evidence ‘no evidence’. It is evidence with limited confidence.
Table move: Keep High and Low Certainty Separate From Large and Small Effect
Write two sentences: estimated effect first, certainty second.
Use Plain Language Without Erasing Uncertainty
Plain language should make the evidence easier to understand while preserving the judgement.
Choose verbs matched to certainty: ‘probably’ for moderate certainty and ‘may’ for low certainty are common conventions, but follow the current framework and context. State direction, magnitude and uncertainty.
Plain-language example: Use Plain Language Without Erasing Uncertainty
‘The programme may slightly improve comprehension, but the evidence is uncertain’ is clearer than ‘positive outcomes were observed.’
Do not use certainty words mechanically or imply guaranteed individual outcomes.
Table move: Use Plain Language Without Erasing Uncertainty
Ask a non-specialist to paraphrase the conclusion; correct any interpretation stronger than the table.
A Worked Summary of Findings Row
Imagine five studies of peer feedback for secondary writing.
The outcome is rubric score at one term. The pooled estimate suggests a small benefit. Concerns include unblinded rating, variation in programme design and a confidence interval spanning trivial to moderate benefit.
End-to-end outcome: A Worked Summary of Findings Row
The row reports participants and studies, relative or mean effect as appropriate, absolute interpretation where possible, certainty and footnotes. The plain-language line says the intervention may slightly improve scores, with limited confidence because of bias, inconsistency and imprecision.
The example does not recommend adopting the programme; recommendation requires broader decision criteria.
Table move: A Worked Summary of Findings Row
Draft the row, then trace every number and judgement back to an analysis or meeting record.
Connect Summary of Findings to Recommendations Carefully
A Summary of Findings table informs a recommendation but does not generate one automatically.
GRADEpro also supports Evidence to Decision frameworks that consider benefits, harms, values, resources, equity, acceptability and feasibility. Evidence certainty is one input.
Decision example: Connect Summary of Findings to Recommendations Carefully
A modestly beneficial programme may still be impractical in one school and feasible in another. The recommendation explains context and conditions.
Do not convert high certainty into ‘must adopt’ or low certainty into ‘must reject’ without the decision framework.
Table move: Connect Summary of Findings to Recommendations Carefully
Separate evidence conclusion, contextual judgement and final recommendation in the document.
Handle Multiple Time Points and Scales Honestly
Outcomes measured at different times or with different scales may not belong in one simple row.
Define the time point and direction of better scores. If standardised effects are used, explain what they mean and provide a familiar interpretation where possible.
Measurement example: Handle Multiple Time Points and Scales Honestly
Immediate writing quality improves, but six-month retention is uncertain. Two rows keep the temporal difference visible.
Combining unlike scales can obscure clinical or educational meaning even when statistically possible.
Table move: Handle Multiple Time Points and Scales Honestly
Label scale, range, direction, follow-up and minimally important threshold where available.
Document Team Judgement and Disagreement
GRADE assessments involve structured judgement, so the record should show who decided and how disagreements were resolved.
Use independent assessment where required, consensus meetings and recorded reasons. Update footnotes when evidence changes. Distinguish methodological guidance from local thresholds.
Team-judgement example: Document Team Judgement and Disagreement
Two reviewers disagree about indirectness. They compare the target population, document the difference and agree on one-level downgrading with a precise rationale.
Consensus should not erase minority concerns that reveal unresolved uncertainty.
Table move: Document Team Judgement and Disagreement
Keep a decision log linked to each outcome and table version.
Parents and Students Can Learn From Evidence Tables
A well-built table teaches that a confident answer can coexist with uncertainty about size or transfer.
Ask the student to explain one outcome, one effect measure, one certainty level and one footnote in ordinary language. Then ask what decision the table cannot make alone.
Learning example: Parents and Students Can Learn From Evidence Tables
A student sees ‘low certainty’ and initially says the intervention fails. After reading, the student revises: the estimate suggests benefit, but confidence is limited.
Evidence tables about health or policy are not personal advice.
Table move: Parents and Students Can Learn From Evidence Tables
Use a four-question routine: what happened, how much, how sure and what else matters?
Transfer Certainty Language to Education and Careers
The habit of separating estimate, confidence and decision is useful in data science, policy, business and everyday information.
Forecasts, audits and evaluations all need magnitude, uncertainty, assumptions and consequences. Plain-language summaries should remain traceable to technical analysis.
Workplace example: Transfer Certainty Language to Education and Careers
A workplace report says a training course ‘probably reduces errors slightly’ and names sample and measurement limits instead of promising productivity gains.
GRADE terminology should not be borrowed casually when a formal GRADE assessment was not performed.
Table move: Transfer Certainty Language to Education and Careers
Use generic language such as confidence, uncertainty and limitations unless the actual framework was applied.
Finish With a Table-to-Text Audit
Before publication, every narrative sentence should match the table and its footnotes.
Check outcome, population, comparator, direction, magnitude, interval, time point, certainty and wording. Ensure the abstract does not sound stronger than the Summary of Findings.
Final audit example: Finish With a Table-to-Text Audit
The table says low certainty and small possible benefit; the abstract says ‘effective’. The team revises the abstract to preserve uncertainty.
A clean table cannot repair an incorrect meta-analysis or flawed included studies.
Table move: Finish With a Table-to-Text Audit
Read numbers, footnotes and plain-language conclusion as one unit and sign off the final version.
Choose the Correct Evidence Table Type
GRADEpro offers several evidence-table formats for different audiences and purposes.
The official table-type guidance distinguishes evidence profiles, Summary of Findings versions and interactive formats. Choose based on whether readers need full domain judgements or a concise decision summary.
Format example: Choose the Correct Evidence Table Type
A technical methods team uses an evidence profile; a school leader receives a Summary of Findings table with plain-language notes.
A prettier format is not automatically more appropriate.
Table move: Choose the Correct Evidence Table Type
Name the audience and decision before choosing the table.
Define the Comparator Precisely
Effects have meaning only relative to a clear comparator.
Name usual practice, placebo, waitlist, alternative programme or no intervention and describe important co-interventions.
Comparator example: Define the Comparator Precisely
‘Peer feedback versus teacher feedback’ differs from ‘peer feedback plus teacher feedback versus teacher feedback alone.’
Ambiguous comparators make absolute effects and recommendations hard to interpret.
Table move: Define the Comparator Precisely
Repeat the comparator in every key conclusion.
Keep Direction of Benefit Visible
Scales can run in opposite directions: higher may mean better performance or worse symptoms.
Record scale range, direction and minimally important difference where available. Reverse or transform scores transparently.
Direction example: Keep Direction of Benefit Visible
A lower error score indicates improvement, while a higher comprehension score indicates improvement. The table labels both.
A sign error can reverse a conclusion.
Table move: Keep Direction of Benefit Visible
Add ‘higher is better’ or ‘lower is better’ beside each continuous outcome.
Explain Numbers of Participants and Studies
Participant totals and study counts describe the evidence base but do not prove precision or diversity.
Report the analysed population relevant to the outcome, not merely everyone enrolled across all studies. Explain clustered or repeated measures where relevant.
Denominator example: Explain Numbers of Participants and Studies
Ten studies contribute to the review, but only four and 620 participants contribute to the long-term outcome row.
Large totals can hide that one study dominates the estimate.
Table move: Explain Numbers of Participants and Studies
Match each row’s N and study count to its analysis.
Report Harms With the Same Discipline as Benefits
Harms and burdens can be rare, poorly measured or omitted.
Define adverse outcomes, ascertainment and time horizon. Apply certainty domains separately. Avoid ‘safe’ when evidence only shows no detected difference.
Harms example: Report Harms With the Same Discipline as Benefits
No study reports serious harm, but dropout due to workload is inconsistently measured. The summary says harms are uncertain.
Absence of reported harm is not evidence of absence.
Table move: Report Harms With the Same Discipline as Benefits
Include a harms row or explicitly explain why it cannot be assessed.
Avoid Double Downgrading the Same Problem
One methodological issue can influence several domains, but teams should not mechanically punish it twice.
Identify the causal pathway from limitation to uncertainty. Decide where it is best represented and explain any related judgements.
Domain-interaction example: Avoid Double Downgrading the Same Problem
Small studies produce wide intervals and variable effects. The team distinguishes imprecision from inconsistency rather than applying duplicate vague penalties.
Domain interactions require expert judgement.
Table move: Avoid Double Downgrading the Same Problem
Review all downgrade reasons together before finalising the rating.
Use Thresholds That Matter to the Decision
Statistical thresholds and decision thresholds are not the same.
Define what difference would be important enough to change action, with stakeholder and domain input. Use that threshold when judging imprecision and plain-language magnitude.
Threshold example: Use Thresholds That Matter to the Decision
A two-point score change is statistically detectable but below the agreed meaningful threshold. The conclusion says little or no important difference.
Thresholds can vary across populations and should not be invented after seeing results.
Table move: Use Thresholds That Matter to the Decision
Document the threshold source and date before analysis.
Distinguish No Important Effect From No Evidence
These phrases answer different questions.
‘Little or no important effect’ reflects an estimate sufficiently bounded around a trivial range. ‘Evidence is very uncertain’ reflects inability to locate the effect reliably.
Interpretation example: Distinguish No Important Effect From No Evidence
A narrow interval around zero supports the first phrase; a wide interval spanning harm and benefit supports the second.
Both may produce a non-significant p-value, so p-values alone cannot choose the wording.
Table move: Distinguish No Important Effect From No Evidence
Base language on magnitude, interval, threshold and certainty.
Version the Table When Evidence Updates
Evidence tables are living products when searches and analyses update.
Record table version, search date, changed studies, revised estimates and judgement changes. Preserve prior versions for audit.
Update example: Version the Table When Evidence Updates
A new large trial narrows imprecision and raises certainty from low to moderate. The update note explains exactly why.
Silent replacement obscures decision history.
Table move: Version the Table When Evidence Updates
Publish a short change log with every material update.
Check Accessibility of Interactive Tables
Interactive Summary of Findings can present outcomes visually, but readers need accessible text and stable links.
Provide captions, plain-language alternatives and exportable versions. Test mobile display and keyboard access where possible.
Accessibility example: Check Accessibility of Interactive Tables
A chart tooltip contains the only confidence interval. The team adds the value to visible text so it survives printing and screen readers.
Interactivity should not hide evidence from users with limited bandwidth or assistive technology.
Table move: Check Accessibility of Interactive Tables
Ensure every visual claim has a textual equivalent.
Explain Certainty Changes Across Updates
When a rating changes, readers need the reason rather than only the new symbol.
Compare evidence base, domain judgements and thresholds between versions. A new study may improve precision while introducing inconsistency.
Change example: Explain Certainty Changes Across Updates
Certainty rises from low to moderate because a large rigorous trial narrows the interval, while indirectness remains.
Changing guidance methods can also alter ratings and should be disclosed.
Table move: Explain Certainty Changes Across Updates
Publish an old-to-new domain table with one sentence per change.
Avoid Mixing Certainty With Recommendation Strength
Strong and conditional recommendations are not synonyms for high and low certainty.
Recommendations incorporate balance of effects and contextual criteria. Occasionally a strong recommendation may arise despite lower certainty under specific circumstances, and high-certainty small effects may still yield conditional action.
Recommendation example: Avoid Mixing Certainty With Recommendation Strength
A costly programme has high-certainty tiny benefit, so adoption remains context-dependent.
Do not infer recommendation strength from the certainty column alone.
Table move: Avoid Mixing Certainty With Recommendation Strength
Read the Evidence to Decision rationale beside the table.
Preserve Units and Time Horizons
An effect without units or time is incomplete.
Write points on a named scale, events per denominator and follow-up period. Convert only with transparent assumptions.
Unit example: Preserve Units and Time Horizons
‘Five fewer absences per 100 students over one term’ is more useful than ‘reduced absence by 5.’
Mixing weeks, terms and years can create false comparisons.
Table move: Preserve Units and Time Horizons
Add unit, denominator and time to every effect cell and narrative sentence.
Use Sensitivity Analyses as Context
Sensitivity analyses show whether conclusions depend on reasonable analytic choices.
Report whether excluding high-risk studies, changing models or using alternative assumptions changes magnitude or certainty. Do not choose only the analysis that supports a preferred claim.
Sensitivity example: Use Sensitivity Analyses as Context
The average benefit disappears when high-risk studies are removed, strengthening a risk-of-bias concern.
A sensitivity analysis is informative only when its choice is justified.
Table move: Use Sensitivity Analyses as Context
Summarise the main and decision-relevant sensitivity results together.
Create a Reader Test Before Publication
A table has succeeded when intended readers interpret it at the correct strength.
Ask several readers to explain the comparator, effect, absolute difference, certainty and decision limit without coaching. Revise labels and footnotes where interpretations diverge.
Reader-test example: Create a Reader Test Before Publication
Readers mistake ‘very low certainty’ for ‘very small effect’. The team rewrites the heading note to separate confidence from magnitude.
Readability testing does not validate the underlying statistics.
Table move: Create a Reader Test Before Publication
Record reader errors and the wording changes they prompted.
Keep Statistical Models Behind the Table Traceable
A concise row should still point back to the analysis that produced it.
Record effect measure, model, adjustment, clustering, missing-data handling and software output. The Summary of Findings presents the result; the analysis file preserves its derivation.
Traceability example: Keep Statistical Models Behind the Table Traceable
A cluster-randomised school trial requires an adjusted effective sample. The row uses the corrected analysis rather than raw pupil counts.
Copying a forest-plot number without its model and population can create mismatches.
Table move: Keep Statistical Models Behind the Table Traceable
Link every row to a named analysis object and verification note.
Write Conclusions That Survive Headline Compression
Abstracts, slides and media summaries often compress evidence more aggressively than tables.
Prepare a one-sentence conclusion containing direction, approximate magnitude, certainty and key population. Test shorter versions against it.
Communication example: Write Conclusions That Survive Headline Compression
‘The programme may slightly improve short-term comprehension in secondary pupils; certainty is low’ survives better than ‘Programme works.’
A headline cannot carry every footnote, so it must avoid stronger verbs than the full evidence permits.
Table move: Write Conclusions That Survive Headline Compression
Approve the headline and table together and prohibit wording that drops the certainty signal.
Frequently Asked Questions
What does GRADE certainty describe?
Confidence in the effect estimate for a specific outcome and evidence body, not a score for the intervention, authors or software.
Is high certainty the same as a large benefit?
No. A small effect can have high certainty, while a large estimate can have very low certainty.
Does low certainty mean no evidence?
No. It means confidence is limited and the true effect may differ substantially from the estimate.
Why show absolute effects?
They translate relative effects into expected differences under a stated baseline risk, which often supports clearer decisions.
Can a confidence interval prove no effect?
No. It shows a range compatible with the estimate under assumptions. Interpret whether it includes important benefit, harm or trivial effect.
What belongs in a certainty footnote?
The domain concern, evidence affected, seriousness and why that concern changes confidence in the outcome estimate.
Does GRADEpro make the judgement automatically?
No. It structures data and judgements; trained reviewers remain responsible for the evidence and rationale.
Can one review have different certainty levels?
Yes. Certainty is usually outcome-specific because evidence bodies and limitations differ.
Does a Summary of Findings table make a recommendation?
No. Recommendations also consider benefits, harms, values, resources, equity, acceptability and feasibility.
What is the best wording check?
Confirm that the text matches the table’s direction, magnitude, interval, time point and certainty without adding promises.
Why should the table show both benefit and burden?
A decision depends on desirable and undesirable effects, not one preferred outcome. Reporting workload, withdrawals, harms and implementation burden beside benefits prevents a favourable score from becoming an incomplete recommendation. Readers can then see which outcomes were measured, which remain uncertain and which values still require local discussion.
Useful Next Reading
- GRADEpro certainty assessment guide — verify the current domains and workflow.
- GRADEpro evidence-table types — compare Summary of Findings and evidence profiles.
- How systematic reviews work — connect certainty assessment to the complete evidence pipeline.
- Why English matters when explaining scientific evidence — move from technical result to careful explanation.
Closing thought: A good Summary of Findings table does not flatten evidence. It concentrates the reasoning. English keeps the estimate, confidence and decision separate enough to be honest—and close enough to be useful.