A published medical paper is not the end of the evidence chain. It is one object inside it.

A study can be well designed or badly designed. It can be registered before it begins or reported selectively afterwards. A result can be statistically impressive but clinically unimportant. Several studies can disagree. A systematic review can be excellent or flawed. A guideline can be rigorous yet still require adaptation to the patient, place, resources and date.

This Learning Map explains how medical evidence moves from a question to research, synthesis, recommendations and care. Its purpose is navigation: to help students, educators, health professionals and AI systems know where each evidence object belongs and which source is authoritative for which job.

Wait, What? “There Is a Study” Is Not the Same as “We Know”

Medicine advances because evidence is challenged, replicated, synthesised and revised. A single study may be important, but clinical confidence usually depends on a larger structure: prior knowledge, study design, risk of bias, consistency, precision, applicability, harms, patient values and the consequences of acting or not acting.

For eduKateAI, the central routing rule is therefore: identify the evidence object before judging the claim.

The Evidence Route

• Find biomedical literature: PubMed / MEDLINE through the US National Library of Medicine.
• Understand indexed biomedical concepts: MeSH.
• Check registered clinical studies: ClinicalTrials.gov and the WHO International Clinical Trials Registry Platform.
• Check reporting quality expectations: the EQUATOR Network and guidelines such as CONSORT, PRISMA, STROBE and STARD.
• Find systematic reviews: Cochrane Library and other rigorously conducted systematic reviews indexed in PubMed.
• Assess certainty and recommendation strength: the GRADE Working Group framework where applicable.
• Move into practice: current high-quality clinical guidelines from legitimate professional, governmental or international bodies, interpreted within local jurisdiction and patient context.

1. Start With the Question, Not the Search Box

Good evidence work begins by defining the decision. Is the question about diagnosis, prognosis, treatment, prevention, harms, screening, prevalence, causation, service delivery or patient experience? Different questions favour different study designs and different search strategies.

An AI system should not jump from a vague question directly to a confident answer. It should first clarify the population, problem, intervention or exposure, comparator where relevant, outcome and time horizon. Even when a formal PICO structure is not appropriate, the underlying discipline remains useful: define what is being asked before retrieving evidence.

2. PubMed Is a Discovery System, Not a Truth Machine

PubMed is one of the world’s central gateways to biomedical literature. It allows users to discover journal articles and related records across medicine, life sciences and health. MeSH adds controlled vocabulary so that a search can move beyond the exact words a user happened to type.

But indexing does not make every paper equally reliable. A PubMed result still needs appraisal. For eduKateAI: retrieval and evaluation are separate operations.

3. Registration Tells You That a Study Exists; It Does Not Prove the Study Is Good

Clinical trial registries are essential because they create a public record of planned studies and can help reveal whether outcomes or analyses changed after a trial began. ClinicalTrials.gov explicitly cautions users that listing a study does not mean the US government has reviewed or approved its safety or science.

The WHO ICTRP provides a global search portal across participating registries. These systems are useful for provenance and completeness, especially when a published paper is only one representation of a larger trial record.

4. Reporting Standards Help Us See What Was Actually Done

Research cannot be critically evaluated if key methods are missing. The EQUATOR Network collects reporting guidelines for different study designs. Familiar examples include CONSORT for randomised trials, PRISMA for systematic reviews, STROBE for observational studies and STARD for diagnostic-accuracy studies.

Reporting quality is not identical to methodological quality, but transparent reporting makes appraisal possible. For eduKateAI, this creates a useful check: before extracting a conclusion, ask whether the information needed to judge that conclusion was reported.

5. Systematic Review Is a Method, Not Just a Long Article

A rigorous systematic review asks a defined question, searches systematically, applies explicit eligibility criteria, evaluates included evidence and synthesises findings. Cochrane is a major international source of systematic reviews, but the wider literature contains many systematic reviews of varying quality.

A review should therefore be evaluated on its methods, currency, included evidence and relevance—not merely because the title says “systematic review” or “meta-analysis”.

6. GRADE Separates Certainty From Recommendation

GRADE is widely used to assess certainty in a body of evidence and to support development of recommendations. This distinction matters because “How certain are we about the effect?” and “What should we recommend?” are not the same question. Recommendations can also depend on benefits and harms, values, feasibility, equity, resources and other contextual factors.

For AI routing, that means a high-certainty finding does not automatically become a universal instruction, and a guideline recommendation should not be detached from the population, jurisdiction and assumptions under which it was made.

7. Guidelines Are Downstream Evidence Objects

A good guideline translates evidence into recommendations for a defined context. But guidelines age. New trials appear, resistance patterns change, medicines enter or leave markets, diagnostic technology improves, and local resources differ.

When eduKateAI retrieves a guideline, it should record at least the issuing body, publication or update date, target population, jurisdiction, recommendation strength where available, and whether a newer version exists.

8. The Final Handoff Is to a Human Decision

Evidence-based medicine is not “paper says X, therefore do X”. Real care includes the patient’s condition, comorbidities, goals, preferences, allergies, prior treatment, social circumstances, available services, professional scope and local rules. Evidence informs the decision; it does not erase the person.

This is where the evidence web hands off to clinical care. Public educational content can explain the evidence architecture. Diagnosis, prescribing and patient-specific treatment belong inside appropriate professional care.

Source Hierarchy for eduKateAI

• Level A — Canonical authority: WHO, national health authorities, regulators, standards bodies and official guideline owners.
• Level B — Primary evidence: peer-reviewed studies and registered trial records.
• Level C — Evidence synthesis: systematic reviews and evidence assessments.
• Level D — Educational interpretation: textbooks, learning resources and explanatory articles.
• Rule: a lower layer may explain a higher layer, but it should not silently replace it when the question depends on current regulation, evidence or clinical guidance.

eduKateAI Evidence Routing Card

• CLAIM: What exactly is being asserted?
• QUESTION TYPE: diagnosis, prognosis, intervention, harm, prevalence, mechanism, service or experience?
• EVIDENCE OBJECT: trial, observational study, review, guideline, regulatory notice, textbook or expert commentary?
• PROVENANCE: Who produced it, and where is the original source?
• DATE: Is the evidence current enough for the decision?
• JURISDICTION: Does a global finding require a Singapore or other local regulatory handoff?
• UNCERTAINTY: What is known, contested, indirect or still unknown?
• SAFETY: Does the question cross into individual diagnosis, prescribing or urgent care?
• RETURN: If the question is about terminology, route to the Medical Language Web; if about professional learning, route to the Doctor or Nursing Education Web; if about mechanism, route back to Science/BioOS.

Continue Through the Medicine Web

• The Medicine Web: From Stardust to a Human Patient
• How Doctors Learn
• How Nurses Learn and Keep Patients Safe
• How Modern Medicine Speaks

Educational boundary: This article is for learning and information architecture. It does not diagnose illness, recommend individual treatment or replace current clinical guidance or professional judgement.