NEW SYNTHESIS
Evidence you can verify
Findings & analysis materialsCertainty for each outcomeHuman review & release version
Every conclusion links to its sources
LIVE EVIDENCE · GLOBAL COLLABORATIONFrom global research,
From global research,
to evidence that matters to you.
Evidence infrastructure for clinical AGI
Evidence prepared for a person: multiple sources, rigorous checks and professional review, with a verifiable path behind every judgement.
Find my serviceGlobal research & collaboration
Literature & guidelines
Trials & registered studies
Authorised case research data
Case data enter research after authorisation, de-identification and study design.
THE EOS EVIDENCE PROCESS
EOSMedical annotation · Trusted assets
Reliable · Traceable · Auditable
Reasoned exclusionsGaps await research
Deliver for a specific question
Personalised evidence report
For this person: applicability, benefits, harms and uncertainty.
Personal context informs applicability, not eligibility.EOS CORE DATA ASSETS
Source → Passage → Annotation → Review → Analysis → Version
EVIDENCE CHAIN · PROCESS
Gather globally
Define the clinical question and search plan. Gather literature, registered studies and authorised research data, recording searches, dates and sources without filtering unfavourable findings by preference.
Records retained at this stage
- Question & protocol version
- Search strategy & sources
- Search dates & update plan
The illustration shows the target process; agree update frequency, appraisal methods and review arrangements for each project.
Evidence sources and appraisal methods
Explore journals, evidence syntheses, registries and clinical inputs, then inspect the scope of their applicability.
EVIDENCE PYRAMID
Evidence has levels. Decisions have context.
From global research to the individual patient: inspect how the evidence was produced and whether it answers the question.
A simplified guide for treatment-effect questions. Bias, consistency, precision and relevance determine confidence alongside design.
- Systematic reviews & meta-analyses
- Randomized controlled trials
- Observational studiesCohorts · registries · case–control
- Case reports & case series
- Expert opinion & mechanistic reasoning
Patient goalsCircumstancesIndividual relevance
UNDERSTANDING EVIDENCE SYNTHESIS
Different meta-analyses answer different questions.
The value is to turn scattered research into more reliable, useful information for patients. A suitable method matters more than a complex name.
Systematic review: find and appraise
Find, select and appraise research using predefined methods, including evidence that cannot be pooled.
Meta-analysis: combine appropriately
Statistically combine comparable results, usually weighting by precision, to estimate effects and uncertainty. A review need not include a meta-analysis.
Patient decision: does it fit me?
Consider absolute benefits, risks, comorbidity, goals and circumstances. Average benefit does not imply suitability for everyone.
TEACHING ASSUMPTION · NOT A STUDY RESULT
The same relative effect can mean different absolute benefits.
Assume a treatment reduces relative risk by 20% (RR = 0.8), over the same follow-up.
Without treatment
10 / 100
With treatment
8 / 100
Expected reduction per 100 people: 2 events; absolute risk reduction: 2 percentage points
The shared scale is 0–10 events per 100 people. Actual benefit requires a matching population, time window and reliable evidence; these numbers do not apply to a patient.
Start with the question, then choose the method.
The directory includes methods, data types, updating approaches and fields. They can be combined and are not a reliability ranking. Examples illustrate methods, not treatment findings.
Treatment effects & choices
Compare options and investigate differences in benefit.
Pairwise meta-analysisIs A better than B, and by how much?
- Example
- Combine comparisons of a preoperative intervention with usual care for infection and function.
- Required evidence
- Randomized trials, or comparative studies organized by design and bias.
- Conditions & limits
- Outcomes, comparisons and time windows must be sufficiently comparable. Random effects do not remove bias.
Network meta-analysisHow do several treatments compare?
- Example
- Use a common comparator when exercise, medication and injections lack every direct comparison.
- Required evidence
- Usually a connected network of randomized trials; other designs need specific justification.
- Conditions & limits
- Check transitivity and agreement of direct and indirect evidence. Ranking is not a treatment decision.
Individual participant data meta-analysis · IPDWhich patients may benefit more?
- Example
- Use participant-level trial data to examine whether diabetes or BMI modifies treatment effects.
- Required evidence
- Authorized participant-level data checked and harmonized across studies.
- Conditions & limits
- Prespecify interactions and retain trial structure. Unavailable studies matter; IPD is not automatically a personal prediction model.
Crossover-trial meta-analysisHow do interventions differ within the same person?
- Example
- Combine trials in which the same participants receive two short-acting, reversible symptom treatments in sequence.
- Required evidence
- Randomized crossover trials with paired results, period and washout information.
- Conditions & limits
- Account for pairing, period and carryover effects; observations are not independent groups.
Risks, doses & population characteristics
Distinguish description, association and treatment effects.
Single-arm meta-analysisWhat outcomes occur among people receiving a treatment?
- Example
- Summarize revision rates for an implant or response rates for a treatment.
- Required evidence
- Single-arm trials, case series or one treatment arm from a study.
- Conditions & limits
- Without a fair comparator, superiority is unproven. Rates from separate studies are not a randomized comparison.
Proportion & prevalence meta-analysisHow common is a condition in a population?
- Example
- Combine the prevalence of osteoporosis among people with diabetes across settings.
- Required evidence
- Population studies with defined numerators, denominators, diagnostic criteria and sampling.
- Conditions & limits
- Prevalence, incidence rates and post-treatment proportions differ. Check population, sampling and follow-up.
Meta-analysis of single-group meansWhat is the average level of a measure?
- Example
- Describe average bone density or function scores in a defined population.
- Required evidence
- Means, standard deviations, sample sizes and comparable units.
- Conditions & limits
- An average level is not a treatment effect. Scales, timing and populations may not be comparable.
Risk-factor meta-analysisIs a characteristic associated with a future outcome?
- Example
- Study the association between sarcopenia and postoperative complications.
- Required evidence
- Cohort and case-control studies, with attention to adjustment and timing.
- Conditions & limits
- Association is not causation. Adjustment sets may differ; cross-sectional studies cannot establish timing.
Correlation meta-analysisDo two measures vary together?
- Example
- Combine correlations between grip strength and function scores.
- Required evidence
- Correlation coefficients, sample sizes and compatible definitions and measurements.
- Conditions & limits
- Co-variation does not establish causation or directly predict an individual outcome.
Dose-response meta-analysisHow do benefits and risks change with dose or exposure?
- Example
- Examine walking duration and function for a gradient or plateau.
- Required evidence
- Dose or exposure levels, group sizes, effects and uncertainty.
- Conditions & limits
- Observational data retain confounding and reverse causation. A turning point is not automatically a recommended dose.
Diagnosis & prediction
Appraise tests and models, then consider clinical usefulness.
Diagnostic accuracy meta-analysisHow accurately does a test identify a condition?
- Example
- Appraise sensitivity and specificity of an imaging test for fractures.
- Required evidence
- Index test, reference standard, thresholds and diagnostic classification data.
- Conditions & limits
- Use appropriate hierarchical models and consider thresholds and selection. Accuracy does not establish patient benefit.
Machine-learning model review & meta-analysisDoes a model work in new patients and settings?
- Example
- Synthesize external validation of models predicting postoperative complications.
- Required evidence
- Model development and validation studies, distinguishing internal and external validation.
- Conditions & limits
- This is a field, not one statistical model. Check calibration, discrimination, leakage and usefulness, not just mean AUC.
Radiomics review & meta-analysisDo imaging-feature models offer reproducible diagnostic or predictive value?
- Example
- Appraise imaging models that distinguish bone-tumour types.
- Required evidence
- Imaging acquisition, segmentation, features and diagnostic or predictive validation data.
- Conditions & limits
- Choose diagnostic or prediction methods; check acquisition, segmentation, feature selection and independent validation.
Evidence accumulation & updating
Track changing findings or map existing reviews.
Cumulative meta-analysisWhen do accumulating studies change the findings?
- Example
- Add trials by publication date and track effects and intervals.
- Required evidence
- Study effects, uncertainty and a defined ordering rule.
- Conditions & limits
- Ordering affects interpretation and repeated significance testing needs care. Cumulative analysis does not remove bias.
Trial sequential analysisDo accumulating trials support a conclusion, or is it premature?
- Example
- Assess information size and sequential boundaries under prespecified assumptions in a planned series of trials.
- Required evidence
- Usually trial synthesis with assumptions about meaningful effects, error rates and information size.
- Conditions & limits
- Results depend on assumptions. Cochrane discourages their use for main analyses or conclusions in most standard updated meta-analyses. Crossing a boundary does not remove bias or establish general applicability.
Living systematic review & meta-analysisDo new studies require a change to the current judgment?
- Example
- Search for new trials on a schedule and update versions after selection and review.
- Required evidence
- Ongoing searching, eligibility rules, update triggers and review arrangements.
- Conditions & limits
- Living describes maintenance. Search cutoff, completed review and report release are distinct; zero delay is not guaranteed.
Prospective meta-analysisCan synthesis be planned before results are known?
- Example
- Teams agree outcomes and synthesis plans before combining trial results.
- Required evidence
- A collaboration protocol defined before results and eligible studies.
- Conditions & limits
- This limits result-driven selection but still requires appraisal of conduct, missing results and applicability.
Umbrella reviewWhat do existing reviews say across a field?
- Example
- Map reviews of exercise, diet and medication for bone health and their certainty.
- Required evidence
- Systematic reviews and meta-analyses rather than primary studies as the inclusion unit.
- Conditions & limits
- Do not simply average reviews. Address overlapping studies, differences in quality and outdated searches.
Complex data & specialist fields
Match the model to the data structure and question.
Multilevel & multivariate meta-analysisHow can correlated results from one study be used?
- Example
- A rehabilitation trial reports pain, function and multiple follow-up times.
- Required evidence
- Effect estimates, hierarchy and correlation information or explicit assumptions.
- Conditions & limits
- Correlated outcomes are not independent studies. Complexity does not ensure reliability.
Genetic, ecological & other specialist meta-analysesCan a defined quantity be combined across studies in a specialist field?
- Example
- Combine associations of the same genetic variant and outcome, or compare environmental detection methods.
- Required evidence
- Comparable target quantities, measurement methods and domain-specific data structures.
- Conditions & limits
- This is not one method. Group-level associations are not individual relationships, and genetic association is not automatically causal.
What makes a synthesis more trustworthy?
A high position in a simplified pyramid does not ensure high certainty. Appraise the question and each important outcome.
- Prespecified questions and analyses; traceable searching and selection.
- Compatible populations, comparisons, outcomes and timing; no duplicate counting.
- Appraise bias, heterogeneity and missing evidence; examine robustness.
- Report effect size, intervals, absolute benefits and certainty.
- Useful negative findings matter; nonsignificance does not establish no effect.
What if information is incomplete or cannot be pooled?
First seek additional data or reconstruct estimates from reported statistics. If pooling remains inappropriate, use prespecified synthesis of available effects, directions and gaps, stating what it can answer.
Do not turn a direction into an invented effect size or vote by significance. EOS research on mixed-information synthesis must evaluate added information, error and applicability; superiority to traditional meta-analysis cannot be presumed.
Read about other synthesis methodsInspect methodological sources
Cochrane · Meta-analysisCochrane · Network meta-analysisCochrane · Individual participant dataCochrane · Study designsCochrane · Non-randomized evidenceCochrane · Diagnostic test accuracyPROBAST+AI · BMJ 2025Cochrane · Accumulating evidenceCochrane · Overviews of reviewsGRADE · Certainty of evidenceCochrane · Other synthesis methods
Method-reference check: 2026-10-06. This is not a clinical topic’s search cutoff or completed-review date.
Have a clinical question to turn into verifiable evidence?
Explore research services & deliverablesGLOBAL RESEARCH · CLINICAL CONTEXT
Multiple evidence routes. One clinical question.
Journals, evidence syntheses, registries and clinical expertise complement one another.
1 / 6
NEJM
Clinical researchPeer-reviewed studies, reviews and clinical cases. Inspect the design, population, findings and relevance of each article.
Visit original sourceJAMA Network
General and specialty medicineClinical research, commentary and specialty journals provide traceable literature across medical questions.
Visit original sourceCochrane
Evidence synthesisSystematic reviews combine studies around focused questions, including benefits, harms and uncertainty.
Visit original sourceAccess to public material and full text depends on permission; expert contributions and project inputs require specific authorisation. Source names and marks do not imply institutional partnerships.
Database, guidance and appraisal-method directory
PubMed
NLM / NIH
Check papers, abstracts and original sourcesFind literature for a clinical question and inspect authors, publication dates, abstracts and full-text links.
Scope & access conditions
PubMed provides citations and abstracts, not journal full text. Access depends on the publisher or archive.
Read the official descriptionNCI · Cancer Treatment
National Cancer Institute / NIH
Cancer treatment options and clinical trialsExplore treatment types, treatment questions and clinical trials, then identify choices to discuss with the treating team.
Scope & access conditions
Public patient information cannot determine your best treatment. Pathology, stage, prior treatment and professional assessment remain necessary.
PubMed Central
NLM / NIH
Read freely available full textsInspect study methods, results tables, supplements and the relevant passages in the full text.
Scope & access conditions
Free reading does not mean unrestricted reuse. Check each article’s copyright and licence.
Read the official descriptionClinicalTrials.gov
NLM / NIH
Check studies and reported resultsReview study objectives, eligibility, locations, status and reported results, then identify questions for the clinical team.
Scope & access conditions
Sponsors or investigators submit the records. A listing is not government approval of safety or effectiveness, or a determination that you should participate.
Read the official descriptionCochrane
Cochrane
Read systematic reviews and plain-language summariesExplore a question’s research synthesis and the confidence researchers have in the findings.
Scope & access conditions
Plain-language summaries are public; full-review access varies. Applicability still depends on the population and question.
WHO Guidelines
World Health Organization
Inspect public-health and clinical guidanceFind WHO guidelines and review the recommendation questions, evidence base and intended settings.
Scope & access conditions
Global guidance is not evidence of local implementation outcomes. Check population, resources, policy and timing.
NICE · NG197
NICE · United Kingdom
Discuss options, benefits and risksShared-decision guidance addresses benefits, risks, consequences, uncertainty and personal preferences.
Scope & access conditions
This is UK decision-support guidance, not a treatment conclusion for an individual or validation of EOS service outcomes.
NICE · NG56
NICE · United Kingdom
Recheck applicability when conditions coexistSingle-condition evidence may come from populations with fewer coexisting conditions. Consider treatment burden, goals, benefits and harms.
Scope & access conditions
This supports the need to review applicability; it does not establish the effectiveness of a new treatment strategy.
PRISMA 2020
PRISMA Executive
Report the review transparentlyReport why the review was done, how studies were identified and selected, the methods and the results.
Scope & access conditions
PRISMA is a reporting guideline, not quality certification or a guarantee of positive findings or publication.
GRADE · Cochrane Handbook
Cochrane · Chapter 14
Show effect size and confidence separatelyPresent important outcomes, relative and absolute effects, available data and certainty, with reasons for downgrading.
Scope & access conditions
GRADE evaluates a body of evidence for a particular outcome. Database inclusion or one study does not automatically establish high certainty.
PRISMA-LSR
PRISMA · 2024 extension
Record how the evidence is updatedThe living-review reporting extension adds detail about the searching and updating process.
Scope & access conditions
It informs a living-review reporting scope; it does not establish that every EOS product already updates automatically.
FTC · Health Products
Federal Trade Commission · US
Does the research match the product claim?Scientific support for health-product claims should match the product and claim, rather than merely a similar ingredient.
Scope & access conditions
This is US guidance. Other markets require separate assessment; it is not an approval or compliance conclusion.
Civil Evidence Rules · China
Supreme People’s Court · China
Link the materials to the issues in the caseRead the civil-evidence rules and agree purpose, material format, deadlines and professional responsibility with the lawyer.
Scope & access conditions
The lawyer checks jurisdiction and the relevant legal date. Literature review is not statutory appraisal or a guarantee of admissibility or success.
How can a delivery be checked?
Locate original sources
Links, DOI or registry ID, with the supporting page or passage.
Check applicability
Population, comparator, outcome, time window and product version.
Separate findings and judgments
Label findings, appraisal basis, professional judgments and missing inputs.
Retain uncertainty
Retain bias, conflicting findings, gaps and unanswered questions.