Key Facts

Epidemiology is the study of disease distribution and determinants in populations. Medical statistics provides the mathematical framework for drawing valid conclusions from clinical research. Both are essential for evidence-based medicine.

Epidemiology

Evidence-based medicine is a cornerstone of modern clinical practice and medical education. Understanding study design and statistical concepts is heavily examined in MRCP Part 1, PLAB, and UKMLA. The ability to critically appraise evidence underpins clinical guideline development, including NICE recommendations.

Aetiology

Study designs (hierarchy of evidence, ascending):

1. Expert opinion/case reports: Lowest level
2. Case-control studies: Compare cases with disease to controls without; retrospective; measure odds ratio
3. Cohort studies: Follow exposed vs unexposed groups; prospective or retrospective; measure relative risk
4. RCTs: Randomise to intervention vs control; minimise confounding; measure treatment effect
5. Systematic reviews/meta-analyses: Pool data from multiple studies; highest level of evidence

Pathophysiology

Bias types:

• Selection bias: Non-random selection of participants (e.g., Berkson's bias, healthy worker effect)
• Information/measurement bias: Systematic errors in data collection (e.g., recall bias in case-control studies)
• Confounding: A third variable associated with both exposure and outcome — addressed by randomisation, matching, or statistical adjustment
• Observer bias: Knowledge of exposure status influences outcome assessment — blinding addresses this
• Publication bias: Positive results more likely published — funnel plot detects this in meta-analyses

Reducing bias:

• Randomisation (allocation concealment)
• Blinding (single, double, triple)
• Intention-to-treat analysis
• Adequate sample size (power calculation)

Application to Clinical Decision-Making

• Screening programmes: Must meet Wilson-Jungner criteria — disease must be important, detectable early, with acceptable treatment
• Diagnostic testing: Pre-test probability × likelihood ratio = post-test probability
• Treatment decisions: Assess NNT, NNH, absolute vs relative risk reduction
• Prognosis: Survival analysis, Kaplan-Meier curves, hazard ratios

UK National Screening Programmes

• Breast cancer: Mammography, women aged 50-70, every 3 years
• Cervical cancer: Cervical screening (HPV primary), age 25-64
• Bowel cancer: FIT (faecal immunochemical test), age 60-74 (expanding to 50-74)
• AAA: USS for men aged 65
• Newborn: Bloodspot screening (PKU, CF, sickle cell, CHT, MCADD + 4 others)

Red Flags in Research Interpretation

• Relative risk reduction quoted without absolute risk reduction — may overstate benefit
• Surrogate endpoints used instead of patient-important outcomes
• Subgroup analyses not pre-specified — hypothesis-generating only
• Industry-funded trials without independent replication

Diagnostic Test Properties

• Sensitivity: Proportion of true positives correctly identified — TP/(TP+FN)
• Specificity: Proportion of true negatives correctly identified — TN/(TN+FP)
• PPV: Proportion of positive tests that are truly positive — TP/(TP+FP)
• NPV: Proportion of negative tests that are truly negative — TN/(TN+FN)
• Likelihood ratio positive: Sensitivity/(1−Specificity)
• Likelihood ratio negative: (1−Sensitivity)/Specificity

Statistical Tests

• Parametric (normally distributed data): t-test (2 groups), ANOVA (>2 groups), Pearson correlation
• Non-parametric (non-normal data): Mann-Whitney U (2 groups), Kruskal-Wallis (>2 groups), Spearman correlation
• Categorical data: Chi-squared test, Fisher's exact test
• Survival analysis: Kaplan-Meier curves, log-rank test, Cox regression (hazard ratio)

Key Statistical Concepts

• p-value: Probability of observing result if null hypothesis is true; <0.05 conventionally "significant"
• Confidence interval: Range within which true value lies with 95% probability; if CI for OR/RR includes 1, result is not significant
• Power: Probability of detecting a true effect (usually target 80%)

Non-pharmacological

• Critical appraisal skills should be applied to all clinical evidence
• NICE guidelines synthesise best available evidence for UK practice
• Clinical audit cycle: Set standard → collect data → compare → implement change → re-audit

Pharmacological

Key concepts in treatment evaluation:

• ARR (absolute risk reduction): Control event rate − treatment event rate
• RRR (relative risk reduction): ARR/control event rate
• NNT: 1/ARR — number needed to treat to prevent one event
• NNH: 1/absolute risk increase — number needed to harm one patient

Example: If a drug reduces MI risk from 8% to 6%:

• ARR = 2%, RRR = 25%, NNT = 50
• This means treating 50 patients for one to benefit

Referral Criteria

• Evidence-based practice should guide all clinical referrals
• Consider NICE guidelines, local protocols, and individual patient factors
• Use validated clinical prediction rules (Wells score, CHA2DS2-VASc, FRAX) for risk stratification

• Evidence-based medicine has transformed clinical practice, improving outcomes across all specialties
• Cochrane reviews provide the gold standard for systematic evidence synthesis
• Implementation of NICE guidelines has standardised UK healthcare quality
• The replication crisis highlights the importance of large, well-designed studies
• Approximately 85% of health research is estimated to be "avoidable waste" (poor design, non-publication, etc.)

2×2 Table for Diagnostic Tests

Levels of Evidence

Common Statistical Errors in Exams