Evidence-Based Practice, Literature Evaluation, and Research
Design and hierarchy
Know the hierarchy: systematic reviews and meta-analyses of randomized trials at the top, then individual RCTs, cohort studies, case-control studies, cross-sectional surveys, case series, and expert opinion. Match design to question: RCTs for efficacy, cohort studies for harms and prognosis, case-control for rare outcomes, and diagnostic accuracy studies for screening tools.
Internal validity
Randomization, allocation concealment, and blinding protect against selection and ascertainment bias. Intention-to-treat analysis preserves randomization but dilutes effect; per-protocol analysis can inflate it. Watch for confounding, attrition, surrogate endpoints, composite outcomes driven by one component, post hoc subgroup claims, and industry funding. A p-value says how incompatible the data are with the null, not how large or important the effect is.
Effect sizes and numbers
Absolute risk reduction, relative risk reduction, and NNT are frequently confused. Relative risk sounds larger; absolute risk reflects real benefit. Odds ratios approximate risk ratios only when the outcome is rare. Confidence intervals convey precision; if an interval crosses the minimally important difference, the result is inconclusive regardless of the p-value. For meta-analyses, examine I-squared for heterogeneity, funnel plots for publication bias, and whether the populations resemble your patient.
Causal and harm assessment
Use the Naranjo scale to structure adverse drug reaction causality and the Bradford Hill considerations for causal inference. GRADE separates certainty of evidence from strength of recommendation.
Worked example
An RCT reports response in 20% with drug versus 15% with placebo. RRR = 5/15 = 33%, which sounds impressive, but ARR = 5%, so NNT = 1/0.05 = 20. Twenty patients treated for one extra response. That framing is what a question is usually testing.