Research Methodology, Evidence-Based Practice and Ethics
This domain tests whether you can appraise a paper and apply its result to a patient, plus the ethical and regulatory rules governing research and publication.
Study design and bias
- Randomized controlled trials answer therapy questions; cohorts answer prognosis and harm; case-control studies suit rare outcomes; cross-sectional studies measure prevalence; diagnostic accuracy studies compare a test with a reference standard.
- Intention-to-treat analysis preserves randomization; per-protocol analysis can exaggerate benefit. Assess allocation concealment, blinding, and completeness of follow-up.
- Bias: selection, information, recall, confounding, publication bias. Confounding is addressed by randomization, restriction, matching, stratification or multivariable regression.
- Reporting standards: CONSORT for trials, STROBE for observational studies, PRISMA for systematic reviews, STARD for diagnostic accuracy.
Interpreting test performance
- Sensitivity equals true positives over all diseased; specificity equals true negatives over all non-diseased. Positive predictive value depends on prevalence and falls as prevalence falls.
- Likelihood ratios: positive LR equals sensitivity divided by (1 minus specificity); a positive LR above 10 is a strong rule-in, negative LR below 0.1 a strong rule-out.
- Number needed to treat is the reciprocal of absolute risk reduction. Confidence intervals convey precision; a P value does not measure effect size.
Evidence grading and guidelines
- GRADE rates certainty as high, moderate, low or very low, considering risk of bias, inconsistency, indirectness, imprecision and publication bias.
- Strong recommendations suit high certainty or high-stakes benefit; weak recommendations require shared decision-making.
Ethics
- Informed consent requires capacity, disclosure, voluntariness and documentation. Research ethics committee approval and Declaration of Helsinki compliance are mandatory.
- Vulnerable groups (children, prisoners, pregnant women, cognitively impaired) need additional safeguards.
- Declare conflicts of interest, avoid plagiarism and duplicate publication, and follow authorship criteria (substantial contribution, drafting or critical revision, final approval, accountability).
Worked example
A FIT test has sensitivity 70% and specificity 95% for colorectal cancer in a population with 1% prevalence. Positive predictive value equals (0.70 x 0.01) divided by ((0.70 x 0.01) + (0.05 x 0.99)), which is 0.007/0.0565, about 12%. Only 12 of 100 positive FIT results represent cancer; in a 5% prevalence population the same test would yield a much higher PPV.
Traps
- Quoting sensitivity when the clinical question is about PPV.
- Ignoring prevalence when comparing two tests.
- Treating a P value above 0.05 as proof of no effect.
- Omitting committee approval or consent documentation in an ethics question.