EBP-EAP cheat sheet

EBP is the integration of best available research evidence, clinical expertise, and patient values and preferences; omitting any one element is not EBP.

The classic five-step EBP process: formulate the question, search for evidence, appraise the evidence, apply it in combination with expertise and patient preferences, and evaluate the outcome.

The evidence hierarchy ranks study designs by risk of bias for questions of intervention effect: systematic reviews and meta-analyses of randomized trials at the top; single randomized trials; cohort; case-control; case series; expert opinion and mechanistic reasoning at the bottom.

The hierarchy is question-specific: for prognosis, inception cohort studies rank highly; for diagnostic accuracy, cross-sectional studies with consecutive recruitment and a valid reference standard rank highly; for qualitative questions, design-specific appraisal tools apply.

Levels-of-evidence schemes (for example Oxford CEBM or GRADE-based ratings) grade bodies of evidence, not individual studies alone; GRADE starts at high for randomized trials and low for observational studies, then upgrades or downgrades.

GRADE downgrades evidence for risk of bias, inconsistency, indirectness, imprecision, and publication bias; it upgrades observational evidence for large magnitude of effect, dose-response, or plausible confounding that would bias against the observed effect.

Internal validity concerns whether the study's results are correct for its sample; external validity concerns whether results apply to other patients, settings, and time periods.

Background questions ask about a condition, test, or concept (who, what, when, how); foreground questions ask about a specific decision in a specific population and drive the search.

Translational continuum: T1 moves basic discovery to humans, T2 moves efficacy to effectiveness in practice, T3 moves evidence into widespread implementation, T4 evaluates population-level outcomes.

PICO components: Population or Patient, Intervention (or exposure), Comparison, and Outcome; a well-built PICO question names a specific population, a specific action, and a measurable outcome.

For qualitative questions, PICo replaces Comparison and Outcome with a Phenomenon of Interest; for diagnostic questions, PIRD uses Population, Index test, Reference standard, Diagnosis.

Question types map to preferred designs: therapy to randomized trials and systematic reviews; diagnosis to cross-sectional accuracy studies; prognosis to cohort studies; etiology or harm to cohort and case-control studies; cost to economic evaluations.

Boolean operators: AND narrows by requiring both concepts; OR broadens by accepting synonyms; NOT excludes terms but risks discarding relevant records.

Search strategy structure: identify concepts from PICO, list synonyms and subject headings for each, combine synonyms within a concept with OR, then combine concepts with AND.

Controlled vocabulary (for example MeSH in MEDLINE or Emtree in Embase) captures indexed synonyms; free-text or keyword searching captures new terminology and unindexed concepts; a high-quality search uses both.

Filters and hedges (study-design filters, date limits, language limits) increase precision but reduce sensitivity; language and date limits can introduce bias and should be justified.

Report the search reproducibly: name each database and platform, give the full strategy, state dates of coverage and the date searched, and record limits and result counts.

Grey literature includes conference abstracts, dissertations, trial registries, regulatory documents, and organizational reports; searching it reduces publication bias.

Snowballing (citation chasing) has two directions: backward for references cited by a key paper and forward for papers that cite it.

Randomization with adequate allocation concealment prevents selection bias; blinding of participants, clinicians, and outcome assessors prevents performance and detection bias.

Intention-to-treat analysis includes all randomized participants in their assigned groups and preserves prognostic balance; per-protocol analysis can exaggerate treatment effects.

Attrition threatens validity when losses are large or differential; a common rule of thumb treats attrition above roughly 20 percent as serious, and differential attrition is more damaging than equal attrition.

Absolute risk reduction equals control event rate minus intervention event rate; number needed to treat equals 1 divided by absolute risk reduction; number needed to harm equals 1 divided by absolute risk increase.

Relative measures (relative risk, odds ratio, hazard ratio) describe association magnitude and do not by themselves convey clinical importance; absolute measures depend on baseline risk.

A confidence interval that excludes the null value (1 for ratios, 0 for differences) indicates statistical significance at the corresponding alpha; the interval width conveys precision.

P values indicate the probability of data at least as extreme under the null hypothesis; they do not indicate effect size or the probability that the null is true.

Validity threats in diagnostic studies: spectrum bias from unrepresentative patients, work-up bias from non-consecutive sampling, incorporation bias when the index test forms part of the reference standard, and review bias from unblinded interpretation.

Sensitivity equals true positives divided by all with disease; specificity equals true negatives divided by all without disease; positive and negative predictive values depend on prevalence.

A likelihood ratio is the change in odds of disease from a test result: positive likelihood ratio equals sensitivity divided by (1 minus specificity); negative likelihood ratio equals (1 minus sensitivity) divided by specificity.

Systematic review appraisal: check a registered protocol, a comprehensive reproducible search, duplicate independent screening and extraction, risk-of-bias assessment, and exploration of heterogeneity.

Heterogeneity statistics: Cochran Q tests for heterogeneity, I-squared estimates the proportion of variability due to between-study differences, and tau-squared estimates between-study variance.

Critical appraisal of qualitative research addresses credibility, transferability, dependability, and confirmability, plus reflexivity and adequacy of sampling.

Narrative synthesis organizes findings by question and outcome without statistical pooling; meta-analysis statistically pools effect estimates and requires clinical and methodological comparability.

Fixed-effect meta-analysis assumes one true effect shared by all studies; random-effects meta-analysis assumes a distribution of true effects and is preferred when heterogeneity is expected.

In a forest plot, each study's point estimate is marked with a confidence interval; square size reflects study weight; the diamond represents the pooled estimate and its confidence interval.

Funnel-plot asymmetry suggests small-study effects or publication bias; it is unreliable when fewer than about ten studies are pooled.

GRADE rates bodies of evidence as high, moderate, low, or very low, and separates the quality of evidence from the strength of a recommendation; strong recommendations can follow low-quality evidence when benefits clearly outweigh harms.

GRADE strong versus conditional (weak) recommendations hinge on balance of benefits and harms, certainty of evidence, patient values and preferences, resource use, and acceptability.

Evidence-to-decision frameworks (for example GRADE EtD) make judgments explicit across benefits and harms, values, resources, equity, acceptability, and feasibility.

Implementation science frameworks: the Consolidated Framework for Implementation Research organizes determinants across intervention characteristics, outer setting, inner setting, characteristics of individuals, and process.

Implementation outcomes are distinct from clinical outcomes and include acceptability, adoption, appropriateness, feasibility, fidelity, cost, penetration, and sustainability.

Audit and feedback, clinical reminders, and multifaceted strategies are more effective than passive dissemination of guidelines alone; passive distribution of educational materials rarely changes practice.

The Model for Improvement poses three questions (what are we trying to accomplish, how will we know a change is an improvement, what changes can we make) and tests them through Plan-Do-Study-Act cycles.

PDSA sequence: Plan the change and prediction, Do the test on a small scale and record observations, Study the results against the prediction, Act by adopting, adapting, or abandoning.

Six Sigma uses DMAIC (Define, Measure, Analyze, Improve, Control) for reducing variation and defects in existing processes.

Lean targets waste (defects, overproduction, waiting, non-utilized talent, transportation, inventory, motion, extra processing) and emphasizes flow and value from the customer's perspective.

A run chart plots a measure over time with a median line; a control chart adds statistically derived control limits, and special-cause variation is signaled by points beyond limits or nonrandom patterns.

Common-cause variation is inherent to a stable process and requires process redesign; special-cause variation arises from assignable causes and warrants investigation of that specific event.

Donabedian's model evaluates quality through structure, process, and outcome; structure enables care, process describes care delivered, and outcome reflects results experienced by patients.

Balancing measures detect unintended consequences of a change in other parts of the system; process measures confirm whether the change was delivered as intended.

Program evaluation designs include formative (improving implementation) and summative (judging merit or worth); logic models link inputs, activities, outputs, and short-, intermediate-, and long-term outcomes.

A key driver diagram connects the aim to primary drivers and secondary drivers, then to specific change ideas that can be tested.

Informed consent requires disclosure of material risks and alternatives, decisional capacity, voluntariness, and comprehension; documentation supports but does not replace the consent process.

Research ethics principles from the Belmont Report are respect for persons, beneficence, and justice; their applications are informed consent, assessment of risks and benefits, and fair selection of subjects.

Institutional review board review categories: exempt, expedited, and full board; exemption determinations are made by the institution, not by the investigator.

The Declaration of Helsinki and the Common Rule govern human-subjects research conduct; equipoise is the ethical basis for randomizing participants in a trial.

Conflict-of-interest management includes disclosure, recusal from decisions, divestment, or independent oversight; disclosure alone may not neutralize bias.

Interprofessional collaboration competencies center on values and ethics, roles and responsibilities, interprofessional communication, and teams and teamwork.

TeamSTEPPS communication tools include SBAR (Situation, Background, Assessment, Recommendation), briefs, huddles, debriefs, and CUS (Concerned, Uncomfortable, Safety issue) for speaking up.

Patient-centered care respects patient values and preferences and supports shared decision-making, which integrates evidence, clinician expertise, and patient goals.

Ethical analysis frameworks weigh beneficence, nonmaleficence, autonomy, and justice; distributive justice concerns fair allocation of limited resources.

Health literacy and cultural humility affect how evidence is communicated; plain-language summaries and teach-back support comprehension.