Evidence-Based Practice, Research and Critical Appraisal
Evidence-based practice integrates the best available evidence, clinical expertise and patient values. Questions test whether you can frame a question, find and appraise evidence, and apply it to a specific patient. Build the question with PICO (population, intervention, comparison, outcome), then match the design to the question: randomised trials for therapy, cohort studies for prognosis, diagnostic accuracy studies for tests, and systematic reviews with meta-analysis when available.
Concepts to own
- Levels of evidence and the hierarchy of designs, plus the idea that a well-conducted lower-level study may beat a flawed higher-level one.
- Bias domains: selection, performance, detection, attrition and reporting, and how randomisation, allocation concealment, blinding and intention-to-treat analysis reduce them.
- Interpreting results: p-values versus effect size, confidence intervals and precision, minimal clinically important difference versus minimal detectable change, number needed to treat, and the difference between odds ratios and risk ratios.
- Diagnostic statistics: sensitivity, specificity, likelihood ratios, and the fact that positive predictive value depends on prevalence.
- Measurement properties of outcome measures: reliability, validity, responsiveness, floor and ceiling effects.
- Applicability: internal versus external validity, and whether the study population resembles your patient.
Traps
Equating statistical significance with clinical importance is the most common error; a large trial can produce a tiny, unimportant difference with a very small p-value. Other traps: assuming correlation proves causation, treating a confidence interval that crosses the null as proof of no effect, and importing a diagnostic accuracy result without considering prevalence in your setting.
Rule of thumb
Ask three questions in order: is the study valid, what is the magnitude of effect with its confidence interval, and does it apply to my patient? A change must exceed the minimal detectable change to be distinguishable from measurement error, and should exceed the minimal clinically important difference to matter to the patient.