Drug Information, Biostatistics and Evidence-Based Practice
Drug Information
Source hierarchy
Primary literature reports original research (randomised trials, cohort studies, case reports). Secondary literature indexes and abstracts primary sources (PubMed/MEDLINE, Embase, International Pharmaceutical Abstracts, Cochrane CENTRAL). Tertiary literature compiles and summarises (drug monographs such as Lexicomp or Micromedex, textbooks, treatment guidelines). For a quick dosing question, a tertiary source is efficient; for a novel safety signal, go to primary literature.
Answering a request systematically
Use the modified systematic approach: secure the requester's details and the real question, refine into a PICO-style structure, search efficiently, evaluate the evidence, formulate a response, and follow up. Document the request, sources, date and answer.
Critical appraisal
Assess study design, randomisation and allocation concealment, blinding, sample size and power, appropriateness of statistical tests, handling of dropouts (intention-to-treat versus per-protocol), size and precision of effect estimates, and applicability to your patient. Bias types to recognise: selection, performance, detection, attrition, publication bias. A surrogate endpoint (e.g. HbA1c) may not equal a patient-important outcome (e.g. cardiovascular events).
Traps
- Quoting a single trial as definitive when a systematic review contradicts it.
- Confusing statistical significance with clinical importance.
- Using a paediatric or healthy-volunteer study to answer a question about older adults with organ impairment.
- Ignoring industry funding or conflict-of-interest disclosures.
Rule of thumb
Match the question type to the source: background/definition → tertiary; current evidence → secondary database; novel or safety question → primary study. Always report the level of evidence alongside the answer.
Biostatistics and Evidence-Based Practice
Design and inference
Know the hierarchy: systematic reviews and meta-analyses of randomised trials, then randomised controlled trials, cohort, case-control, cross-sectional, case series, expert opinion. Randomisation and blinding control confounding and bias; observational designs can show association but not causation. P-values describe the probability of data at least as extreme under the null hypothesis - they do not measure effect size. Confidence intervals convey precision; a wide interval crossing the null indicates an uncertain result.
Effect measures
Relative risk and odds ratios compare groups; absolute risk reduction (ARR) and number needed to treat (NNT = 1/ARR) translate results into clinical terms. Diagnostic tests use sensitivity (SnNout: sensitive test, negative rules out) and specificity (SpPin), plus positive and negative predictive values - which depend on prevalence. Likelihood ratios are prevalence-independent.
Errors and traps
- Type I error: false positive (alpha, usually 0.05); Type II error: false negative (beta); power = 1 − beta.
- PPV falls when prevalence falls, even with the same sensitivity and specificity.
- Intention-to-treat analyses preserve randomisation but dilute effect size.
- Multiple comparisons inflate false-positive risk without correction.
Worked example
A trial reduces events from 20% to 15%. ARR = 5% (0.05), so NNT = 1 ÷ 0.05 = 20 patients treated to prevent one event. The relative risk reduction is 25%, which sounds more impressive than "5 in 100" - always convert to absolute terms before deciding clinical value.