Evidence-Based Practice, Research Methods and Informatics
What this domain tests
Evidence-based practice, research methods and informatics test whether you can appraise evidence and use information systems safely. Expect study designs, bias, randomization, intention-to-treat, confidence intervals, p-values, NNT/NNH, systematic reviews, meta-analysis, diagnostic test characteristics, pharmacoepidemiology, and informatics (CPOE, clinical decision support, privacy, interoperability).
Core rules
- Randomized controlled trials best control confounding; observational studies are useful for rare or long-term outcomes but are vulnerable to selection and confounding.
- Intention-to-treat preserves randomization and gives a pragmatic estimate; per-protocol can exaggerate benefit.
- A p-value is the probability of data at least as extreme if the null hypothesis is true; it is not the probability that the null is true.
- Confidence intervals show precision and magnitude; a wide CI means uncertainty.
- NNT = 1 / absolute risk reduction; NNH = 1 / absolute risk increase. Calculate from absolute, not relative, risks.
- Critical appraisal: validity, results, applicability to your patient.
- Informatics: clinical decision support should be evidence-based, transparent, and interruptive only for high-value alerts; privacy and audit trails are mandatory.
Worked example
An RCT shows a cardiovascular event in 4% of control patients and 2% of treated patients. Relative risk reduction = 50%, but absolute risk reduction = 2% (0.04 − 0.02). NNT = 1 / 0.02 = 50 patients treated for the study period to prevent one event. If the 95% CI for ARR is 0.5% to 3.5%, the NNT CI is 29 to 200 (take reciprocals and reverse the limits). The relative reduction sounds impressive, but the absolute benefit and the CI tell you how much certainty to place in it.
Traps
Do not call a p-value the probability the hypothesis is true; do not calculate NNT from relative risk; do not ignore loss to follow-up; do not accept surrogate outcomes as patient-important; do not assume a meta-analysis is free of publication bias; do not implement an alert without measuring its effect on workflow and safety.