ACADEMIC WRITING GUIDE

How to Use GRADE and Write a Summary of Findings Table for a Dissertation (With Worked Examples)

By Acadelyra Editorial TeamPublished

A systematic review can find several studies pointing in the same direction and still leave an important question unanswered: how certain should readers be about the overall finding? A dissertation that reports only a pooled effect or a list of study conclusions risks confusing the size of a result with confidence in it. GRADE provides a structured way to explain that distinction.

This guide shows how to assess certainty of evidence for important outcomes, construct a Summary of Findings (SoF) table, explain downgrading decisions and write an appropriately cautious dissertation Results and Discussion section. It includes a fully fictional worked example involving a university feedback programme, an absolute-risk calculation and a model table you can adapt to your own verified data.

Teaching example disclaimer: Every study count, effect estimate, confidence interval, certainty judgment and participant number in the worked example is invented. It is not a real systematic review, evidence supporting an educational intervention, or a completed GRADE assessment. Apply the actual current GRADE guidance and your supervisor's approved review methods before reporting real findings.

What is GRADE, and what does a Summary of Findings table show?

GRADE stands for Grading of Recommendations Assessment, Development and Evaluation. In intervention reviews, it provides a transparent framework for assessing certainty in the body of evidence for each important outcome. Certainty concerns how confident reviewers are that the true effect is sufficiently close to the estimated effect to inform a conclusion. It is not a grade awarded to an individual paper, a score measuring how impressive the dissertation is, or a substitute for statistical analysis.

A Summary of Findings table presents a review's most decision-relevant outcomes in one place, usually identifying the population, intervention and comparison, outcome and follow-up, number of studies and participants, estimated effects, certainty rating, and reasons for important judgments. Where appropriate, it reports both relative effects (such as a risk ratio) and absolute effects (such as fewer events per 1,000 people). These help readers see not just whether an intervention might work, but roughly how much difference it might make and how much uncertainty surrounds that difference.

The authoritative starting point is Cochrane Handbook Chapter 14, which explains SoF tables, GRADE's assessment domains and effect presentation. The GRADE Working Group's evolving GRADE Book provides additional methodological guidance.

Not every dissertation must use GRADE. It is especially relevant to systematic reviews addressing intervention effects and other review questions for which an appropriate GRADE framework exists. A narrative literature review, scoping review or qualitative synthesis should not automatically copy an intervention-effect SoF table. Agree the review type, guidance and reporting expectations with your supervisor and protocol first.

Why study-level risk of bias is not the same as certainty of evidence

Our guide to assessing risk of bias in a systematic review addresses whether an individual study result may have been systematically distorted. GRADE asks a broader question: taken together, how confidently can we interpret the evidence about a particular outcome?

A set of trials may have reasonably controlled bias yet produce very wide confidence intervals. Another body of evidence may be precise but based on a population unlike the one relevant to the dissertation question. A single study can also contribute to several outcomes with different certainty ratings because the amount, precision and quality of evidence differ by outcome.

Appraisal tasks and units of assessment in a systematic review
Appraisal taskUnit of assessmentMain outputWhat it cannot establish alone
Study-result risk of biasSpecific result within a studySupported domain and overall judgments under the chosen toolCertainty across all studies for an outcome
Evidence synthesisStudies or results addressing a common questionEstimate, patterns or justified narrative synthesisConfidence that all limitations have been accounted for
GRADE certainty assessmentBody of evidence for a given outcomeHigh, moderate, low or very low certainty, with explanationsA clinical or policy recommendation by itself
Summary of Findings tableImportant outcomes for a specified comparisonEffect estimates, absolute context, numbers and certainty side by sideReplacement for the Methods chapter, evidence profile or full analysis

For a broader discussion of appraising primary studies, see our critical-appraisal guide. The two activities should connect, not be duplicated under different labels.

The four GRADE certainty levels

GRADE generally communicates four levels. Wording must still describe the particular question and evidence—not imply more precision than the data warrant.

Four GRADE certainty levels and example dissertation wording
Certainty levelWhat it communicates in plain languageExample wording for a dissertation
HighVery confident that the true effect lies close to the estimate“The evidence indicates that the intervention reduces the outcome.”
ModerateModerately confident; the true effect is likely close, but could be substantially different“The intervention probably reduces the outcome.”
LowLimited confidence; the true effect may be substantially different“The intervention may reduce the outcome.”
Very lowVery little confidence; the true effect is highly uncertain“The effect is very uncertain.”

These are communication conventions, not a mechanical phrase generator. A moderate-certainty estimate with a confidence interval spanning no effect and important benefit still requires a conclusion faithful to that interval. A high-certainty finding about a small, narrow population is not automatically applicable to everyone.

For conventional GRADE assessments of intervention effects, evidence from randomised trials generally starts at high certainty and may be rated down, while non-randomised evidence has traditionally started lower and may sometimes be rated up. Do not apply these starting points blindly: contemporary approaches and the use of tools such as ROBINS-I require attention to the relevant GRADE guidance and rating pathway. The example below deliberately uses randomised trials to keep its teaching assumptions clear.

Step 1: Define one review comparison and the outcomes that matter

Start with a focused PICO-style question: population, intervention, comparator and outcomes. Decide in advance which outcomes are critical or important to your readers. In Cochrane-style intervention reviews, the main SoF table generally includes no more than seven important outcomes for a comparison; it should not list every secondary variable merely because a paper reported it.

For our fictional dissertation, the question is:

Among university students receiving usual teaching, does a structured weekly feedback programme, compared with usual teaching alone, reduce the proportion who fail an end-of-term assessment?

The review authors have planned three important outcomes: assessment failure (critical), assessment score (important), and adverse workload effects (important). The relevant follow-up is one academic term, except where specified. A different comparison—such as feedback programme versus peer mentoring—would ordinarily need a separate SoF presentation.

Why choose outcomes before seeing the results? Selecting only the most statistically favourable outcomes produces a misleading picture and can introduce selection bias into the review. Your planned outcome set should reflect the dissertation question and what matters to people making decisions. If an important outcome has no usable studies, report the absence rather than quietly removing it.

For upstream methods, our systematic literature-search guide explains how to document the evidence base. Our PRISMA 2020 flow-diagram guide explains study selection; neither a search diagram nor a flow diagram is itself a certainty assessment.

Step 2: Prepare an outcome-level evidence record

Before choosing certainty ratings, extract the actual findings for the same comparison and outcome. Check that outcome definitions, measurement scales, follow-up times, populations and effect measures are sufficiently compatible for the intended synthesis.

For each outcome, keep a traceable worksheet containing:

  • The prespecified outcome, its measurement and follow-up period.

  • Exact contributing study identifiers and participant totals; verify whether the totals differ across outcomes.

  • Study designs, existing risk-of-bias judgments and their documentary basis.

  • Effect estimate and confidence interval, or a transparent narrative synthesis where pooling is not appropriate.

  • Comparator or baseline risk for binary outcomes, with a defensible source and population context.

  • Any prespecified threshold for a difference that matters to people making decisions.

  • Domain-level GRADE judgments, reasons and locations in the underlying evidence.

A pooled estimate is not obligatory simply because several studies exist. If data cannot responsibly be combined, explain the narrative method used; use the applicable SoF format for narrative evidence rather than inventing a pooled risk ratio. The GRADEpro guidance on evidence table types distinguishes quantitative and narrative presentation modes.

Step 3: Evaluate the five main reasons for rating certainty down

For intervention evidence, GRADE asks reviewers to consider five core domains. A downgrade is a reasoned judgment, not an automatic penalty for spotting a buzzword in an article.

Risk of bias

Could limitations in the included studies systematically change the estimated effect? Use the actual result-specific assessments, consider how much evidence comes from studies at higher risk and explain why the problem is consequential. Do not downgrade solely because the reporting is imperfect if the uncertainty does not materially affect the result being judged. Equally, do not claim “low risk” merely because the trials were randomised.

Inconsistency

Do the study results disagree in ways that remain unexplained? Look at effect directions, sizes, confidence intervals, populations, interventions and plausible sources of heterogeneity. A high I² statistic can be a signal to examine, but I² alone is not a GRADE downgrade rule. Nor does a low I² automatically establish consistency when there are very few studies or very imprecise estimates.

Indirectness

Do the actual studies answer the dissertation's intended question? Evidence about graduate students in one type of institution may be less directly relevant to undergraduate students elsewhere. Outcomes may also be surrogates for what readers truly care about: improved satisfaction is not interchangeable with better academic performance. The point is to identify meaningful differences in population, intervention, comparator or outcome.

Imprecision

Is the evidence too uncertain to distinguish effects that would lead to different conclusions? Confidence intervals, sample size, event counts, decision thresholds and the range of plausible benefit or harm all matter. An interval spanning an unimportant improvement and a substantial improvement may merit a downgrade even if it excludes exactly zero; conversely, statistical non-significance does not automatically establish imprecision without examining the question and thresholds.

Publication bias

Could eligible studies or results be missing in a way that distorts the body of evidence? Consider trial registrations, protocols, selective availability of results, small-study patterns and the completeness of the search where relevant. A funnel plot is not a reliable standalone verdict, especially with few studies. Missing-results bias should not be confused with a poorly documented PRISMA screening count.

Record a short, evidence-based explanation for each important judgment. A defensible note might read: “Rated down one level for serious imprecision because the confidence interval spans effects smaller and larger than the prespecified decision-relevant threshold.” Avoid notes such as “low sample quality” or “papers not good” that cannot be audited.

Step 4: Calculate an absolute effect from a risk ratio — fictional example

Suppose a fictional synthesis of four randomised trials involving 920 students reports an end-of-term assessment-failure risk ratio (RR) of 0.75, with a 95% confidence interval of 0.60 to 0.95. In this teaching example, a plausible baseline risk under usual teaching is 200 failures per 1,000 students.

A risk ratio compares the event risk with the intervention to the risk with the comparator. To express the result on an absolute scale, apply the RR to a justified baseline risk:

Formula: Intervention risk = baseline risk × risk ratio (RR).

At the point estimate: 200 per 1,000 × 0.75 = 150 failures per 1,000. The absolute difference is 150 − 200 = 50 fewer failures per 1,000.

At the confidence limits: 200 × 0.60 = 120 per 1,000 and 200 × 0.95 = 190 per 1,000. Relative to the baseline, those correspond to 80 fewer to 10 fewer failures per 1,000. Written in conventional increasing-order effects, this is 50 fewer per 1,000 (95% CI: 80 fewer to 10 fewer).

Conversion of a fictional risk ratio into corresponding and absolute risks
Fictional quantityCalculationResult
Comparator riskGiven hypothetical baseline200 failures per 1,000
Estimated intervention risk200 × 0.75150 failures per 1,000
Lower corresponding risk200 × 0.60120 failures per 1,000
Upper corresponding risk200 × 0.95190 failures per 1,000
Absolute difference150 − 20050 fewer failures per 1,000
Absolute difference interval120 − 200; 190 − 20080 fewer to 10 fewer per 1,000

Table 1. Conversion of a hypothetical pooled risk ratio into corresponding and absolute risks for an explicitly assumed baseline.

This calculation uses the risk ratio, not an odds ratio, for direct multiplication. Odds ratios require a different conversion. The baseline risk is also not a universal constant: if the typical comparison-group risk were 100 per 1,000 instead of 200, the absolute difference implied by RR 0.75 would be smaller. Always identify where the baseline comes from, state the follow-up period and use suitable rounding.

The numerical estimate is a separate matter from certainty. Even a correctly calculated interval does not establish whether the underlying studies are at risk of bias, whether the evidence is indirect or whether studies are missing. For broader advice about explaining effect measures, see our dissertation effect-size guide.

Step 5: Make transparent GRADE judgments for the example

For the fictional assessment-failure outcome, imagine the reviewers found that most contributing trial results were at low risk of bias; effect directions were broadly compatible; the students, programmes and outcome were directly relevant; and there was no specific evidence suggesting missing studies. However, the protocol defined 30 fewer failures per 1,000 as a threshold for an educationally meaningful benefit. The hypothetical interval spans 10 fewer (less than that threshold) through 80 fewer (more than that threshold). On that rationale, the reviewers judge serious imprecision and downgrade one level: high → moderate certainty.

That is an illustration of a reasoned judgment, not a rule that every interval spanning 30 per 1,000 must be downgraded. Real ratings require reviewing the full body of evidence and using the current methodology appropriate to the question.

For the second fictional outcome, imagine assessment score on a 0–100 scale, pooled across two trials involving 410 students. The estimated mean difference (MD) is 4.2 points higher with the programme (95% CI: 1.0 to 7.4 higher). In this invented example, the reviewers judge serious risk of bias because important score outcomes were incompletely measured, and serious imprecision because the interval spans differences both below and above a prespecified 5-point educational-importance threshold. Two warranted one-level downgrades take certainty from high → low. This does not mean the programme has a proven 4.2-point benefit; it means the estimate is uncertain and the evidence warrants a cautious statement.

The third planned outcome, adverse workload effects, was not measured by any included trial. It cannot simply receive a fabricated “very low” rating attached to an invented pooled estimate. State no eligible outcome data, explain the evidence gap, and follow the applicable guidance for presenting an empty or unmeasured important outcome.

Step 6: Build a Summary of Findings table

The table below is deliberately compact for readability on mobile devices; formal reviews may use a more detailed accepted GRADEpro layout. It shows how the numbers, outcomes and certainty explanations must remain connected.

Comparison: Structured weekly feedback programme versus usual teaching alone. Population: University students. Setting: Hypothetical higher-education institutions. Follow-up: One academic term. All data are fictional.

Fictional GRADE Summary of Findings for weekly feedback versus usual teaching
Important outcomeStudies / studentsEstimated resultGRADE certainty
Assessment failure4 trials / 920 studentsBaseline 200/1,000; intervention 150/1,000; 50 fewer/1,000 (95% CI: 80 fewer to 10 fewer); RR 0.75 (95% CI 0.60–0.95)Moderate¹
Assessment score (0–100)2 trials / 410 studentsMD 4.2 points higher (95% CI: 1.0–7.4 higher)Low²
Adverse workload effectsNo eligible trial dataNot estimable from the included evidenceNot rated³

Table 2. Fictional GRADE-style Summary of Findings illustration for a dissertation. It is an instructional simplification, not an official completed GRADEpro assessment.

Explanations: ¹ Downgraded one level for serious imprecision: the absolute-effect interval crosses the example's prespecified threshold of 30 fewer failures per 1,000. ² Downgraded one level for serious risk of bias and one for serious imprecision, based on the hypothetical evidence described above. ³ The important outcome was prespecified but not measured; do not invent an estimate or certainty rating.

A good table also makes the direction of benefit and harm unmistakable. For assessment failure, a lower event risk is favourable. For assessment score, a higher mean is favourable. Many dissertation tables become confusing because they use “positive effect” without stating which direction of the underlying measure is desirable.

For binary outcomes, include an appropriate comparator or assumed risk alongside an absolute effect and its interval. For continuous outcomes, give the measurement scale and meaning; if studies used different scales, document how the standardised effect was interpreted. Include exact units, follow-up periods and study denominators. Do not conceal missing outcomes or extrapolate the 920-student denominator to outcomes measured in only 410 students.

Step 7: Use GRADEpro without letting software make the judgments for you

GRADEpro GDT is the GRADE Working Group's software for preparing SoF tables and related evidence profiles; Cochrane's GRADEpro resource describes its role and integration with review tools. It can structure the table and help calculate corresponding risks, but it cannot verify that your extracted data are correct or replace reasoned decisions about indirectness and imprecision.

A practical workflow is to enter the review question and comparison, define the prespecified important outcomes, add effect data and study counts, document the five-domain judgments with evidence-backed explanations, review any calculated absolute effects against the baseline risk, and export the resulting table into the dissertation's Results chapter. Check the software's current help and your institution's access arrangements rather than assuming a particular button name or import function will remain unchanged.

Keep the underlying evidence profile or assessor notes, including who made each judgment and how disagreements were resolved where applicable. A table with four coloured certainty labels but no explanation is not reproducible. Likewise, a generated table should be checked for copy-and-paste errors, mislabelled outcomes and missing confidence intervals.

Step 8: Write the Methods paragraph accurately

Your Methods chapter should explain what you actually did, not claim adherence to a protocol that was never followed. The following is an adaptable fictional model, not text to paste unchanged into a real dissertation:

We assessed certainty of evidence separately for the prespecified critical and important outcomes using the GRADE approach. For each outcome, we considered risk of bias, inconsistency, indirectness, imprecision and publication bias. We documented reasons for any rating change and presented key comparisons in a Summary of Findings table. We used the outcome-specific effect estimate and its confidence interval, together with an identified comparator risk where appropriate, to describe absolute effects. [Insert the actual assessor arrangement, guidance version, decision thresholds and discrepancy-resolution process used.]

If you worked alone, state that; do not imply a second independent reviewer. If your evidence synthesis is narrative rather than quantitative, describe how you assessed certainty for that presentation method. If your review does not use GRADE, do not insert this paragraph simply to sound methodologically sophisticated.

For help matching methods to the question, see our dissertation methodology chapter guide.

Step 9: Write Results and Discussion without overstating certainty

A Results paragraph should state the direction and scale of the finding, the uncertainty interval, the body of evidence and the certainty rating together. Our fictional example might be reported as follows:

Four hypothetical randomised trials (920 students) suggested that a structured weekly feedback programme probably reduced end-of-term assessment failure compared with usual teaching (RR 0.75, 95% CI 0.60–0.95; approximately 50 fewer failures per 1,000 students at an assumed baseline of 200 per 1,000; 95% CI 80 fewer to 10 fewer). We rated the certainty moderate, downgrading for imprecision because the absolute-effect interval included improvements smaller than the prespecified educational-importance threshold. Two hypothetical trials (410 students) reported higher assessment scores (MD 4.2 points, 95% CI 1.0–7.4 higher), but certainty was low owing to risk of bias and imprecision. No eligible trial measured adverse workload effects.

In the Discussion, interpret what the findings could mean for the particular population without turning the fictional example into a real practice recommendation. The assessment-failure estimate suggests a benefit, but the magnitude important to the decision remains uncertain. The score outcome is less secure. The absence of evidence on workload prevents a balanced assessment of benefits and burdens.

Avoid three common leaps: “the pooled effect is statistically significant, therefore the evidence is high certainty”; “the table says moderate, therefore the intervention is definitely effective”; and “no studies measured harms, therefore there are no harms.” None follows logically. When you move from evidence to recommendations, consider feasibility, values, resources and competing consequences—questions beyond the certainty grade alone. For appropriately qualified recommendations, see our guide to writing dissertation recommendations.

A repeatable GRADE decision log

For a real project, keep this short audit log behind each published table. It is a practical aid; it does not replace the formal tool's own worksheets.

Outcome-level GRADE decision log and supporting evidence
Decision recordExample of what belongs there
Question and comparisonUniversity feedback programme versus usual teaching; one-term assessment failure
Outcome importance and thresholdCritical outcome; meaningful benefit defined before seeing the results
Contributing evidenceStudy IDs, trial designs, participant totals, data locations
Effect dataPooled RR and interval; justified baseline risk; absolute effects
Bias judgmentDomain findings from the relevant study-result tool, with supporting pages
Consistency / directnessWhy study variation matters or does not; how closely studies match PICO
PrecisionWhether intervals span meaningfully different decisions and why
Missing evidenceRegistration, protocols, missing outcomes, suspected reporting patterns
Final certainty and explanationStarting point, each rating change, final level, reviewer and date

Table 3. Suggested internal evidence trail for an outcome-level GRADE decision.

The log matters most when a supervisor asks, “Why was this outcome rated moderate rather than high?” Instead of reconstructing your reasoning from memory, you can show the exact estimate, threshold, studies and rationale. That also makes updating the review easier if a newly included study changes the pooled result or bias profile.

Common mistakes when using GRADE in a dissertation

Giving one certainty grade to the entire dissertation. GRADE ratings usually apply to a body of evidence for a particular outcome, not the entire review, individual study or document.

Choosing outcomes after seeing which results are significant. An SoF table should reflect important prespecified outcomes, including outcomes with no usable evidence where appropriate.

Confusing a risk ratio with an absolute effect. A relative reduction of 25% does not mean 25 percentage points fewer events. The absolute reduction depends on baseline risk.

Treating the number of studies as certainty. Ten weak or highly inconsistent studies can support less confident conclusions than fewer well-conducted studies with appropriate precision.

Using statistical significance as the imprecision rule. Confidence intervals should be evaluated against outcome-relevant thresholds, not just against whether they cross a null value.

Rating down automatically for every difference across studies. Explain whether effect differences are important, whether they are already accounted for, and whether the underlying evidence supports a consistent conclusion.

Reporting certainty without explanations. Readers need to know which domain affected confidence and what evidence supports the judgment.

Pooling outcomes that are not comparable. The appearance of a tidy effect estimate does not justify mixing incompatible interventions, measurement scales or follow-up periods.

Copying a GRADEpro table without verification. Cross-check denominators, time horizons, effect direction, baseline risk and footnotes against the actual studies and synthesis.

Claiming that GRADE is mandatory for every review. Use the approach suited to the actual review question, protocol, programme and disciplinary guidance.

Frequently asked questions

Is GRADE required for every dissertation systematic review?

No universal dissertation rule applies. GRADE is widely used for suitable systematic reviews of intervention effects and other contexts with applicable guidance. Requirements depend on your research question, review type, programme and supervisor. Follow your approved protocol and clearly explain any alternative certainty-assessment approach.

Can I use GRADE if I do not perform a meta-analysis?

Potentially, yes. The key is to use guidance appropriate for the kind of narrative synthesis you performed, and to be transparent about what can and cannot be estimated. Never manufacture a pooled effect to fill a table. Narrative evidence tables and appropriate certainty assessments require methodological care.

Is GRADE the same as Cochrane RoB 2?

No. RoB 2 assesses the risk of bias in particular randomised-trial results. Risk of bias is one domain that can influence GRADE certainty across the body of evidence for an outcome. Other important considerations include imprecision, inconsistency, indirectness and publication bias.

Can an outcome have moderate certainty even if its confidence interval excludes no effect?

Yes. An interval can exclude the null but still span effects that differ in practical importance. Certainty also depends on bias, consistency, directness and missing evidence. The reasons for a rating must be explicit and grounded in the actual evidence.

How many outcomes go in a Summary of Findings table?

Cochrane-style intervention reviews typically choose up to seven of the most important outcomes for each principal comparison. Do not include seven by default; a smaller, carefully justified set is often better. Follow the requirements applicable to your review.

Can I copy the worked example into my dissertation?

Use it to understand the method, not as empirical evidence. Its trials, participants, effect estimates and ratings are fictional. A genuine dissertation must report its own extracted study data, justified rating decisions and accurately cited sources.

Final checklist before submitting your Summary of Findings table

  • [ ] I confirmed that GRADE is appropriate to my review question and approved methods.

  • [ ] My important outcomes and comparisons were chosen on defensible grounds.

  • [ ] Each outcome has the correct contributing studies and participants.

  • [ ] Effects, confidence intervals, scales and follow-up periods match the underlying synthesis.

  • [ ] Binary-outcome absolute effects use an explicit and justified comparator risk.

  • [ ] GRADE ratings are outcome-specific and reflect the appropriate methodological pathway.

  • [ ] I evaluated the five main domains rather than using a generic “study quality” score.

  • [ ] Each meaningful downgrade or upgrade is supported by a traceable explanation.

  • [ ] I did not invent effect estimates or certainty ratings for unmeasured outcomes.

  • [ ] The Methods, Results, Discussion and SoF table tell a consistent story.

  • [ ] My wording distinguishes estimated effects from confidence in those estimates.

  • [ ] All citations and internal links lead to the intended original guidance or Acadelyra resources.

Final takeaway

A strong Summary of Findings table is not impressive because it contains symbols or complex statistics. It is useful because a reader can identify the question, how much difference the evidence suggests, how uncertain that difference is and why the certainty rating is justified. Work from the actual studies, make your assumptions visible, and keep the wording as cautious as the evidence requires.

If you would value human-led help with the organisation, clarity, referencing or presentation of your systematic review, explore Acadelyra's academic support services or request academic help. Our role is to support your legitimate research and writing process, not to invent results or substitute for the research you must conduct yourself.

Authoritative references and further reading

  1. Schünemann, H. J., Higgins, J. P. T., Vist, G. E., Glasziou, P., Akl, E. A., Skoetz, N., & Guyatt, G. H. Chapter 14: Completing ‘Summary of findings’ tables and grading the certainty of the evidence. In the Cochrane Handbook for Systematic Reviews of Interventions. Official chapter.

  2. Cochrane. The GRADE approach. Methodological resource.

  3. GRADE Working Group. GRADE Book: Overview of the GRADE approach. Official evolving guidance.

  4. Cochrane Methods GRADEing. GRADEpro GDT. Official description of the tool.

  5. GRADEpro. Types of questions and evidence tables. Official help guidance.

  6. Cochrane. Methodological expectations for assessing the certainty of evidence and summarizing findings. MECIR conduct standards.