Guide · Core, Pathways and documents

From PubMed to the bedside: how guidelines, reviews and trials are prioritised

Written for
Hospital physicians, emergency doctors and residents
Reading time
7 minutes
Published
Updated

Written by the Sana Health team for clinicians. It summarises published guidance, is not a guideline and has not been peer reviewed. Check current local protocols before acting.

Key points

  • Study design matters because observational studies have repeatedly suggested benefits that randomised trials did not confirm.
  • The evidence hierarchy is a starting point: the best design depends on whether the question is about treatment, harm, diagnosis or prognosis.
  • GRADE rates certainty for each outcome, not for a study, and separates the certainty of evidence from the strength of a recommendation.
  • Sana Health searches guidelines, systematic reviews, meta-analyses and randomised trials first, reads the records from the databases when the question is asked, and excludes retracted papers.
  • Drug labels attached to answers are US product information; licensed doses and indications differ between countries, and local guidance takes precedence.
  • The reference list supports the clinician's own reading of the evidence; it does not replace it.

Why the order of evidence matters

Evidence-based medicine was defined by the McMaster group as the integration of the best current evidence with clinical expertise and patient values, practised in steps: ask an answerable question, search, appraise, apply and evaluate [1]. Appraisal starts with design. Well-designed and properly executed randomised trials are regarded as the most reliable evidence on the benefits of a treatment [2], because randomisation balances the known and unknown differences between groups that confound observational comparisons.

The history of medicine has examples. Observational studies suggested that vitamin E might prevent cardiovascular disease and cancer, yet a meta-analysis of randomised trials raised the possibility that high doses increase mortality [3]. Observational studies supported the benefits of a Mediterranean diet, but the randomised evidence is more limited [4]. Statin product labels list adverse effects drawn largely from non-randomised, unblinded studies, which a meta-analysis of double-blind trials set out to test [5]. Some reviews compare randomised and observational estimates for the same question directly [6].

The hierarchy, and where it bends

For questions about treatment, the usual order runs from guidelines built on systematic reviews, through systematic reviews and meta-analyses of randomised trials, individual randomised trials, cohort and case-control studies, case series and reports, to expert opinion. Systematic reviews sit near the top only when they are done and reported well, which is what the PRISMA 2020 guidance addresses [7].

QuestionDesign usually most informativeCaveat
Does a treatment work?Randomised trials and systematic reviews of themCheck the population and the comparator against your patient
Does it cause a rare or late harm?Large cohort and case-control studies, alongside trialsConfounding by indication
Is a test accurate?Cross-sectional studies against a reference standardSpectrum of patients studied
What is the prognosis?Inception cohort studiesLoss to follow-up
How is a rare disease managed?Registries, case series and expert consensusLow certainty is often unavoidable

Design is a starting point, not a verdict. A large, well-conducted cohort can be more informative than a small trial at high risk of bias, and a meta-analysis is only as good as the trials it pools.

GRADE in practical terms

The GRADE approach, now used by many guideline bodies including the Infectious Diseases Society of America [8] and the American Gastroenterological Association [9], rates the certainty of evidence for each outcome rather than for each study. Randomised evidence starts as high certainty and observational evidence as low. Certainty is rated down for risk of bias, imprecision, inconsistency, indirectness and publication bias, and observational evidence can occasionally be rated up, for example for a very large effect.

CertaintyWhat it meansTypical wording
HighFurther research is very unlikely to change confidence in the estimatereduces
ModerateThe true effect is probably close to the estimate, but may differprobably reduces
LowThe true effect may be substantially differentmay reduce
Very lowThe estimate is very uncertainthe effect is uncertain

Strength of recommendation is a separate judgement: it weighs certainty with the balance of benefit and harm, patients' values and resources, so a strong recommendation can occasionally rest on low-certainty evidence. Recommendations also change as evidence accumulates; EULAR's earlier fibromyalgia recommendations were mostly expert opinion because studies were few [10]. And a guideline can only be as applicable as its trials: restrictive eligibility criteria in stroke thrombectomy trials were sometimes carried word for word into guidelines, leaving open what to do for patients the trials excluded [11].

How Sana Health orders its references

  • References are read from PubMed, ClinicalTrials.gov and, when a question turns on named medicines, DailyMed, at the time the question is asked. The model does not write the list, so its links cannot be invented.
  • Guidelines, systematic reviews, meta-analyses and randomised trials are searched first, including a set of leading general and specialty journals.
  • Each answer ends with at least 40 PubMed records when the literature holds that many, together with registered trials from ClinicalTrials.gov.
  • Retracted papers are excluded; retracted articles can otherwise go on being cited as though they were sound [12].
  • Each article links to PubMed, to the journal's own page and, where one exists, to the free full text.
  • The answer cites the list by number, and any PubMed link the model writes itself is checked against PubMed.

Drug labels attached to answers are US product information from DailyMed. Licensed doses and indications differ between countries, and national formularies and local guidance take precedence.

Reading the list under an answer

  1. Start with the most recent guideline relevant to your country and setting, and note its date.
  2. Look for a systematic review published since that guideline, which may change the picture.
  3. Open the pivotal trials and check their populations, comparators and outcomes against your patient.
  4. Check that each numbered citation supports the sentence it is attached to.
  5. Read registered ongoing trials as a sign that the answer may change.

The reference list is there so that the clinician can check the reasoning. Sana Health supports clinical judgement and does not replace it; the clinician remains responsible for every decision, and local protocols take precedence.

How Sana Health helps

Evidence Review applies this hierarchy to a single question. It restates the question as population, intervention, comparator and outcomes, gives the position of current guidelines with their body, year and strength of recommendation, summarises systematic reviews before individual trials, appraises each key study, and rates the certainty of each main outcome in the manner of GRADE with its reasons. Ongoing trials from ClinicalTrials.gov are listed under gaps, and the numbered references are read from the source databases when the question is asked.

Questions

Is a systematic review always better evidence than a randomised trial?

No. A systematic review is only as reliable as the studies it includes and the way it was done. A meta-analysis of small trials at high risk of bias can be less trustworthy than one large, well-conducted trial. GRADE handles this by rating the whole body of evidence for each outcome.

What is the difference between certainty of evidence and strength of recommendation?

Certainty describes how confident we can be in an estimate of effect. Strength of recommendation is a panel's judgement about whether to advise for or against an intervention, weighing certainty together with benefits, harms, patients' values and resources.

Why are the drug labels in Sana Health answers from the United States?

They come from DailyMed, the official source of US product information. Licensed doses, indications and warnings differ between countries, so the national formulary and local guidance should be checked before prescribing.

References

  1. Evidence-based radiology: steps 1 and 2--asking answerable questions and searching for evidence (opens in a new tab)Radiology, 2007
  2. CONSORT 2025 Statement: Updated Guideline for Reporting Randomized Trials (opens in a new tab)JAMA, 2025
  3. Meta-analysis: high-dosage vitamin E supplementation may increase all-cause mortality (opens in a new tab)Ann Intern Med, 2005
  4. Mediterranean-style diet for the primary and secondary prevention of cardiovascular disease (opens in a new tab)Cochrane Database Syst Rev, 2019
  5. Assessment of adverse effects attributed to statin therapy in product labels: a meta-analysis of double-blind randomised controlled trials (opens in a new tab)Lancet, 2026
  6. Operative treatment versus nonoperative treatment of Achilles tendon ruptures: systematic review and meta-analysis (opens in a new tab)BMJ, 2019
  7. PRISMA 2020 explanation and elaboration: updated guidance and exemplars for reporting systematic reviews (opens in a new tab)BMJ, 2021
  8. Clinical Practice Guidelines by Infectious Diseases Society of America (IDSA): 2025 Guidelines on Management and Treatment of Complicated Urinary Tract Infections--Duration of Antibiotics for Complicated UTI (opens in a new tab)Clin Infect Dis, 2026
  9. AGA Living Clinical Practice Guideline on Pharmacological Management of Moderate-to-Severe Ulcerative Colitis (opens in a new tab)Gastroenterology, 2024
  10. EULAR revised recommendations for the management of fibromyalgia (opens in a new tab)Ann Rheum Dis, 2017
  11. The Problem of Restrictive Thrombectomy Trial Eligibility Criteria (opens in a new tab)Stroke, 2022
  12. Citations of microRNA Biomarker Articles That Were Retracted: A Systematic Review (opens in a new tab)JAMA Netw Open, 2024