Evidence-Based Medicine

Introduction

Evidence-Based Medicine (EBM) is a structured approach to clinical decision-making. It combines the best available research evidence with clinical expertise, the values and preferences of the patient, and the practical circumstances of care. EBM does not mean following research mechanically. It means asking a focused question, finding trustworthy evidence, judging its strengths and limitations, applying it to the person or population in front of you, and evaluating the result.

For university students, EBM is both a research skill and a professional reasoning skill. You use it when deciding whether a treatment is effective, whether a diagnostic test is useful, whether an exposure is harmful, how accurate a prognosis is, or whether a guideline is appropriate for a particular setting.

The evidence hierarchy above is a useful orientation, but it is not a universal ranking for every question. A well-conducted randomized controlled trial is especially valuable for many intervention questions, while diagnostic accuracy, prognosis, harms, and qualitative questions may require other designs. A systematic review is only as reliable as its methods and the studies it includes.

Learning Objectives

By the end of this aiMOOC, you should be able to:

Clinical question: Formulate a focused, answerable question using a framework such as PICO.
Literature search: Select appropriate evidence sources and design a reproducible search strategy.
Critical appraisal: Evaluate internal validity, risk of bias, effect estimates, precision, and relevance.
Evidence synthesis: Interpret systematic reviews, meta-analyses, forest plots, and publication-bias signals.
Clinical decision-making: Integrate research evidence with clinical expertise, patient preferences, and context.
GRADE approach: Explain how certainty of evidence can be rated across important outcomes.
Shared decision-making: Communicate benefits, harms, uncertainty, and alternatives in understandable terms.

The EBM Process

A common practical sequence is Ask, Acquire, Appraise, Apply, and Assess. These steps are iterative rather than strictly linear. New information can lead you to reformulate the question, search again, or reconsider how evidence fits a patient.

Ask: Turn Uncertainty into an Answerable Question

A useful clinical question is specific enough to guide a search. For many intervention questions, the PICO framework helps you define:

Population: Who is the patient group or population?
Intervention: What treatment, exposure, test, or strategy is being considered?
Comparison: What is the relevant alternative?
Outcome: Which benefits, harms, or patient-important outcomes matter?

For example, instead of asking "Do exercise programs help older adults?", you could ask: "In community-dwelling adults aged 65 years and older, does a supervised balance-and-strength program, compared with usual activity, reduce falls over one year?" A focused question improves both searching and appraisal.

Other frameworks may be more suitable for different questions. Diagnostic questions often specify the index test and reference standard; qualitative questions may focus on population, phenomenon of interest, and context. The key principle is to make the uncertainty searchable and clinically meaningful.

Acquire: Find the Best Available Evidence

Efficient evidence searching usually starts with high-quality synthesized sources when they are appropriate. You may consult clinical practice guidelines, systematic reviews, evidence summaries, and then primary studies when needed. Common databases include PubMed/MEDLINE, the Cochrane Library, Embase, CINAHL, PsycINFO, and discipline-specific databases.

A strong search strategy identifies core concepts from the clinical question, selects synonyms and controlled vocabulary, combines terms with Boolean operators, and documents the search so another researcher could reproduce it. Search filters can help identify study designs, but overly restrictive filters may miss relevant evidence.

You should also distinguish finding evidence from selecting evidence. Relevance alone is not enough: the methods, recency, population, outcomes, and risk of bias matter.

Study Designs and the Hierarchy of Evidence

Different study designs answer different questions. Evidence quality depends not only on design label but also on how well the study was planned, conducted, analyzed, and reported.

Randomized Controlled Trials

In a Randomized controlled trial, participants are assigned to intervention groups by a random process. Randomization aims to create comparable groups and reduce confounding. Allocation concealment protects the randomization process before assignment, while blinding can reduce performance or measurement bias when feasible.

Datei:Randomization in clinical trials.webm

Important appraisal questions include:

Was the random sequence generated appropriately?
Was allocation concealed?
Were participants, clinicians, and outcome assessors blinded where possible?
Were groups treated similarly apart from the intervention?
Was follow-up sufficiently complete?
Were outcomes analyzed according to the assigned groups?
Were all prespecified important outcomes reported?

Randomization does not guarantee a trustworthy result. Poor allocation concealment, large loss to follow-up, selective reporting, protocol deviations, or inappropriate analysis can still bias an RCT.

Observational Studies

Cohort studies compare outcomes over time among groups with different exposures. Case-control studies compare prior exposures between people with and without an outcome. Cross-sectional studies measure exposure and outcome at a particular point or period.

Observational studies are often essential for prognosis, rare harms, long-term outcomes, and exposures that cannot ethically or practically be randomized. Their main challenge is confounding: differences between groups may influence the observed association. Good observational research therefore measures important confounders, uses appropriate adjustment, explores alternative explanations, and reports limitations transparently.

Diagnostic Accuracy Studies

Diagnostic evidence asks whether a test correctly distinguishes people with and without a target condition. Core measures include sensitivity, specificity, positive and negative predictive values, and likelihood ratios. Sensitivity and specificity describe test performance in the studied setting, while predictive values depend strongly on disease prevalence.

A high sensitivity can be useful when missing disease is especially costly, while high specificity can be useful when false positives are especially harmful. However, no single statistic should be interpreted in isolation. Spectrum effects, verification bias, imperfect reference standards, and differences between study populations can alter performance.

Appraise: Judge Trustworthiness and Importance

Critical appraisal asks three broad questions: Are the results valid? What do the results mean? Are the results applicable?

Internal Validity and Risk of Bias

Bias is a systematic error that can move an estimate away from the truth. Important forms include selection bias, performance bias, detection bias, attrition bias, selective reporting, and confounding. Risk-of-bias tools are structured aids, not substitutes for reasoning.

When appraising a study, examine the protocol or registration when available, methods of participant selection, assignment and masking, outcome measurement, missing data, analysis choices, selective reporting, funding, and conflicts of interest. A large sample cannot repair a fundamentally biased design.

Effect Measures

For binary outcomes, common effect measures include:

Risk ratio: The risk in the intervention group divided by the risk in the comparison group.
Odds ratio: The odds of an outcome in one group divided by the odds in another group.
Risk difference: The absolute difference in risk between groups.
Number needed to treat: The number of people who need the intervention for one additional beneficial outcome over a specified time.
Number needed to harm: The number of people exposed for one additional harmful outcome over a specified time.

Relative effects can appear impressive while absolute effects remain small. If an intervention reduces risk from 2% to 1%, the relative risk reduction is 50%, but the absolute risk reduction is 1 percentage point and the number needed to treat is 100 over the stated time period. Clear EBM communication reports both relative and absolute effects when possible.

Precision, Confidence Intervals, and Clinical Importance

A point estimate is incomplete without its uncertainty. A confidence interval gives a range of values compatible with the data under the statistical model. Wide intervals indicate imprecision. A result can be statistically compatible with little or no effect even when the point estimate appears favorable.

Statistical significance is not the same as clinical importance. A very small effect can be statistically significant in a large study, while an important effect can remain uncertain in a small study. You should compare the estimate and its confidence interval with a threshold of clinical importance, not only with a p-value.

Systematic Reviews and Meta-Analysis

A Systematic review uses explicit methods to identify, select, appraise, and synthesize evidence addressing a focused question. A Meta-analysis statistically combines compatible study results. Meta-analysis is not automatically appropriate: studies must be sufficiently comparable, and important clinical or methodological differences should be examined.

The PRISMA flow diagram documents how records moved through identification, screening, eligibility assessment, and inclusion. Transparent reporting helps readers understand how the evidence base was assembled.

Reading a Forest Plot

A forest plot displays effect estimates from individual studies and often a pooled estimate. Study markers show point estimates, horizontal lines show confidence intervals, and the pooled result is commonly shown as a diamond. The vertical line of no effect depends on the effect measure: it is 1 for ratios such as risk ratios and odds ratios, and 0 for differences such as risk differences and mean differences.

When reading a forest plot, ask whether effects point in a similar direction, how precise each study is, how much weight each study receives, whether confidence intervals overlap, whether the pooled estimate crosses the line of no effect, and whether heterogeneity has a plausible explanation. A pooled number should never be interpreted without considering study quality and clinical diversity.

Heterogeneity

Heterogeneity means variation among study results. Clinical heterogeneity arises from differences in participants, interventions, comparators, or outcomes. Methodological heterogeneity arises from differences in study design or risk of bias. Statistical heterogeneity is variation in effect estimates beyond what might be expected from sampling error alone.

Statistics such as I-squared can summarize inconsistency, but they do not explain it. Interpretation should consider effect sizes, confidence intervals, study characteristics, prespecified subgroup hypotheses, and whether pooling still makes sense.

Publication Bias and Selective Reporting

Positive or statistically significant results may be more likely to be published, published quickly, or highlighted. Selective outcome reporting can also distort the evidence base when unfavorable or null outcomes are omitted.

Funnel plots can help explore small-study effects when enough studies are available, but asymmetry is not proof of publication bias and symmetry is not proof that bias is absent. Trial registries, protocols, regulatory documents, and searches for unpublished studies can help reveal missing evidence.

Certainty of Evidence with GRADE

The GRADE approach rates the certainty of a body of evidence for each important outcome, rather than assigning one quality label to an entire review. The four certainty levels are high, moderate, low, and very low.

For intervention effects, five key reasons for rating certainty down are:

Risk of bias: Important limitations in study design or conduct.
Inconsistency: Unexplained differences in results across studies.
Indirectness: Evidence differs materially from the population, intervention, comparator, or outcome of interest.
Imprecision: The data are too uncertain to distinguish important benefit, little effect, or harm.
Publication bias: Available studies may be an unrepresentative sample of the evidence that exists.

Under appropriate circumstances, evidence from non-randomized studies may be rated up for factors such as a large effect, a dose-response gradient, or plausible residual confounding that would reduce an observed effect.

For advanced study, the following Cochrane Training webinar examines how thresholds can inform judgments about certainty and decision-making.

GRADE separates certainty of evidence from strength of recommendation. Recommendations also depend on the balance of benefits and harms, patient values and preferences, resource use, equity, acceptability, and feasibility.

Apply: Move from Evidence to a Decision

Evidence becomes useful only when it is interpreted for a real person, population, or policy context. Ask whether the participants, setting, intervention, comparator, and outcomes resemble the situation you face. Consider baseline risk, comorbidities, treatment burden, access, cost, feasibility, and patient goals.

A statistically convincing trial does not force one decision for every patient. A small absolute benefit may be worthwhile to one person and not to another. Likewise, a treatment with average benefit can be inappropriate if the patient has contraindications, values a competing outcome more strongly, or cannot realistically access the intervention.

Shared Decision-Making and Risk Communication

Shared decision-making combines clinical expertise and evidence with the patient's informed preferences. Good risk communication uses consistent denominators, meaningful time frames, absolute risks, balanced framing, and plain language.

For example, "10 out of 100 people experience the outcome without treatment, compared with 7 out of 100 with treatment" is usually more informative than saying only that treatment produces a "30% relative reduction." Visual aids can help, but they should not exaggerate certainty or hide harms.

Assess: Evaluate the Decision and the Process

The final EBM step asks whether your decision achieved the intended outcome and whether your reasoning process can improve. You may audit patient outcomes, adverse effects, adherence, costs, or implementation barriers. You can also evaluate whether your search was efficient, whether appraisal was accurate, and whether new evidence changes the conclusion.

EBM is therefore a cycle of learning. Clinical uncertainty generates questions, questions lead to evidence, evidence informs decisions, and outcomes create new questions.

Common Pitfalls and Misconceptions

EBM is not cookbook medicine. Evidence must be interpreted with expertise and patient values.

A systematic review is not automatically high certainty. A review can synthesize biased, inconsistent, indirect, or imprecise studies.

A p-value is not the probability that the hypothesis is true. It is a property of the observed data under a specified statistical model.

Statistical significance does not establish clinical importance. Magnitude, uncertainty, baseline risk, harms, burden, and patient priorities matter.

The evidence pyramid is not a substitute for question-specific reasoning. The most appropriate design depends on what you are trying to learn.

Absence of evidence is not always evidence of absence. An inconclusive study may simply be too imprecise to rule out an important effect.

Reporting Standards and Research Transparency

Reporting guidelines improve transparency but do not by themselves guarantee high methodological quality. CONSORT 2025 is the current core guideline for reporting randomized trials and supersedes CONSORT 2010. PRISMA 2020 supports transparent reporting of systematic reviews and meta-analyses. Other guidelines address observational studies, diagnostic accuracy studies, prediction models, qualitative research, and trial protocols.

Prospective registration, accessible protocols, predefined outcomes, data sharing where ethically appropriate, reproducible analysis code, and disclosure of conflicts of interest can reduce avoidable ambiguity and make research easier to scrutinize and reuse.

Key Standards and Resources

Cochrane Handbook: Detailed methods for systematic reviews of interventions and evidence synthesis.
PRISMA 2020: Reporting guidance, checklists, and flow diagrams for systematic reviews.
CONSORT 2025 and SPIRIT 2025: Current reporting guidance for randomized trial reports and protocols.
GRADE Working Group: Methods for judging certainty of evidence and strength of recommendations.
Oxford Centre for Evidence-Based Medicine tools: Practical resources for focused questions, searching, appraisal, effect measures, and decisions.

Interactive Tasks

Quiz: Test Your Knowledge

Which statement best describes evidence-based medicine? (Integration of research evidence clinical expertise and patient values) (!Following the newest study regardless of quality) (!Using only randomized trials for every clinical question) (!Replacing patient preferences with guideline recommendations)

What is the main purpose of PICO? (To structure an answerable clinical question) (!To calculate a p value) (!To grade publication bias) (!To replace a literature search)

Why is allocation concealment important in a randomized trial? (It prevents foreknowledge of upcoming assignments) (!It guarantees complete follow up) (!It eliminates all confounding after randomization) (!It makes every outcome clinically important)

Which measure expresses an absolute difference between two event risks? (Risk difference) (!Risk ratio) (!Odds ratio) (!Hazard ratio)

What does a wide confidence interval usually indicate? (Imprecision in the effect estimate) (!Proof of no treatment effect) (!Absence of bias) (!Perfect external validity)

What does the diamond in a typical forest plot represent? (The pooled effect estimate and its confidence interval) (!The study with the highest risk of bias) (!The number of excluded records) (!The threshold for publication)

Which is a GRADE domain that can lower certainty of evidence? (Indirectness) (!Popularity) (!Novelty) (!Journal impact factor)

Why should absolute effects be reported alongside relative effects? (They show the size of benefit or harm at the relevant baseline risk) (!They always produce larger numbers) (!They remove all statistical uncertainty) (!They make confidence intervals unnecessary)

What is a major limitation of interpreting funnel plot asymmetry? (It can have causes other than publication bias) (!It can only be used for qualitative studies) (!It proves that every small study is biased) (!It cannot display study precision)

What is the best next step after applying evidence to a clinical decision? (Assess outcomes and the quality of the decision process) (!Assume the decision will remain correct indefinitely) (!Ignore new evidence once treatment starts) (!Use the same decision for all future patients)

Memory Game

PICO	Framework for population intervention comparison and outcome
Randomization	Chance assignment intended to reduce confounding
Sensitivity	Proportion of people with the condition who test positive
Forestplot	Graphic showing study estimates and often a pooled result
GRADE	Framework for rating certainty of a body of evidence
PRISMA	Reporting guideline for systematic reviews

Drag and Drop

Match the correct terms.	Topic
Ask	Formulate a focused clinical question
Acquire	Search for the best available evidence
Appraise	Judge validity importance and applicability
Apply	Integrate evidence with expertise values and context
Assess	Evaluate outcomes and improve the process

...

Crossword Puzzle

Randomization	What process assigns trial participants by chance?
Confounding	What distortion occurs when another factor influences an observed association?
Sensitivity	What measure is the proportion of people with a condition who test positive?
Precision	What concept describes the degree of statistical uncertainty around an estimate?
Heterogeneity	What term describes variation among study results?
Applicability	What concept asks whether evidence fits the patient or setting?

LearningApps

Cloze Text

Evidence-based medicine integrates research evidence with clinical expertise and

. A focused intervention question can often be structured with

. Random assignment in a controlled trial is intended to reduce

. When appraising results, a confidence interval helps you judge

. Relative effects should usually be interpreted together with

. In a systematic review, a forest plot can display individual and pooled

. Unexplained differences among studies are called

. The GRADE framework evaluates certainty separately for each important

. Evidence may be rated down when it is indirect, imprecise, inconsistent, at risk of bias, or affected by

. Applying evidence requires attention to baseline risk, feasibility, clinical expertise, and

.

Open-Ended Tasks

Easy

PICO practice: Convert a broad health question into a PICO question and explain why each element is useful.
Risk communication: Rewrite a relative treatment effect as an absolute-risk explanation using a hypothetical baseline risk and a consistent denominator.
Study design identification: Find three health-research abstracts and identify the study design used in each, with one sentence explaining your choice.
Evidence diary: Keep a one-week log of clinical or health claims you encounter and classify what type of evidence would best test each claim.

Standard

Database search: Design and document a reproducible PubMed search for a focused clinical question, including synonyms, Boolean operators, and inclusion criteria.
Critical appraisal: Appraise one randomized controlled trial with a recognized appraisal framework and identify at least three potential threats to validity.
Forest plot interpretation: Select a published meta-analysis, annotate its forest plot, and explain the pooled estimate, confidence intervals, heterogeneity, and clinical meaning.
Patient decision aid: Create a one-page decision aid that presents benefits, harms, uncertainty, and alternatives for a real clinical choice in plain English.

Advanced

Systematic review protocol: Draft a short protocol defining eligibility criteria, search methods, outcomes, risk-of-bias assessment, and a synthesis plan for a systematic review.
GRADE assessment: Choose one outcome from a systematic review and justify a certainty rating by considering risk of bias, inconsistency, indirectness, imprecision, and publication bias.
Replication analysis: Recalculate an effect measure and confidence interval from published study data, compare your result with the paper, and explain any discrepancy.
Evidence implementation project: Interview clinicians or other health professionals about a gap between evidence and practice, analyze barriers, and propose an implementation strategy with measurable outcomes.

Learning Assessment

Clinical reasoning assessment: Given a patient scenario, formulate a focused question, select an appropriate evidence source, and justify how the evidence would influence the decision.
Bias analysis: Compare two studies addressing the same intervention and explain how differences in randomization, missing data, outcome measurement, or confounding could change confidence in their results.
Effect interpretation: Translate relative and absolute effects into a patient-centered explanation and discuss how baseline risk changes the practical meaning of the evidence.
Evidence synthesis critique: Evaluate a systematic review by examining its search strategy, inclusion criteria, risk-of-bias methods, heterogeneity, and certainty judgments.
Transfer to practice: Propose an evidence-informed recommendation for a setting with different resources or patient characteristics and explain what makes the evidence direct or indirect.
Uncertainty communication: Present a balanced recommendation when confidence intervals include both clinically important benefit and little or no effect.

Evidence of Learning

Evidence that you have learned the topic should include both knowledge and performance. Important evidence includes:

Knowledge: Accurate explanation of EBM principles, major study designs, bias, effect measures, confidence intervals, systematic reviews, and certainty of evidence.
Search skills: A focused question and reproducible search strategy that retrieves relevant evidence efficiently.
Appraisal skills: Structured judgments about validity, bias, precision, applicability, and clinical importance.
Quantitative reasoning: Correct interpretation of relative and absolute effects, diagnostic accuracy measures, and forest plots.
Products: A critical appraisal, decision aid, evidence summary, protocol, or evidence-based recommendation that is transparent about uncertainty.
Communication: Clear explanation of benefits, harms, and uncertainty using language appropriate for patients, professionals, or policy audiences.
Transfer: Ability to adapt evidence-informed reasoning to a new patient group, healthcare setting, research question, or resource constraint.
Reflection: Ability to identify limitations in your own evidence-searching and decision process and propose concrete improvements.

OERs on the Topic

Linked Learning Areas

Evidence-Based Medicine

Further Connections

Evidence-Based Medicine connects clinical care with Epidemiology, Statistics, Health sciences, Medical education, Nursing, Public health, Health economics, Research methodology, Bioethics, and Science communication. These connections matter because an evidence-informed decision is simultaneously a scientific judgment, a clinical judgment, and a value-sensitive choice made within a real healthcare system.

aiMOOC Projects

Diesen aiMOOC als Battle spielen: Deutsch, Klasse 5-6, G-Niveau

Inhalte werden geladen ...

Mediathek wird aus dem Wiki geladen ...

Frag mich zum aktuellen Lerninhalt oder wähle eine kurze Hilfe.

Textmodus bereit

Sprache