Evidence-Based Medicine and Research
This domain tests whether you can turn a clinical question into a searchable one, judge study validity, and interpret the numbers you quote to patients. Calculation items are common and are usually simple arithmetic spoiled by one conceptual error.
Core rules
- Frame the question as PICO (population, intervention, comparison, outcome). The question type dictates the design: therapy from randomised trials, diagnosis from cross-sectional accuracy studies, prognosis from cohort studies, harm from cohort or case-control studies.
- Validity checks: randomisation and allocation concealment, blinding, intention-to-treat analysis, completeness of follow-up, and baseline comparability.
- Bias to name: selection, measurement, recall, publication, lead-time and length-time bias, confounding.
- Key statistics: sensitivity and specificity are properties of the test; positive and negative predictive values depend on prevalence; likelihood ratios are prevalence-independent. NNT = 1/absolute risk reduction. A p-value is the probability of data this extreme if the null hypothesis were true - not the probability that the null is true.
- Confidence intervals convey precision; a wide interval crossing no effect means the study is inconclusive rather than negative.
- Grades of evidence run from systematic reviews of trials down to expert opinion; guideline strength separates strong from conditional recommendations.
- Audit measures practice against a standard and does not need research ethics approval; research generates new generalisable knowledge.
Worked example
A test with 90% sensitivity and 95% specificity is applied to 1000 people with 10% prevalence. Of 100 diseased, 90 test positive. Of 900 healthy, 45 test positive (5%). PPV = 90/(90+45) = 67%; the same test in a 1% prevalence population gives a PPV near 15%. Reasoning: predictive values move with prevalence, so the test behaves differently in screening versus diagnostic settings.
Traps that cost marks
Quoting PPV without prevalence; reading a non-significant result as proof of no effect; confusing relative with absolute risk reduction; rounding NNT upward.