Epidemiology, biostatistics and health needs assessment
Scope
This domain tests whether you can choose and interpret measures of disease frequency, association and test performance, and use them in health needs assessment. Calculations are unavoidable.
Core rules
- Incidence is new cases per population at risk per time; prevalence is all existing cases at a point or period. Under steady conditions prevalence ≈ incidence × average duration.
- Crude rates are distorted by age structure; age-standardisation allows comparison between populations with different age profiles.
- Cohort studies give incidence and relative risk; case-control studies give odds ratios. A rare outcome makes the odds ratio approximate the risk ratio.
- Sensitivity is the proportion of diseased people correctly identified; specificity is the proportion of non-diseased correctly identified. Positive predictive value depends on prevalence, so it cannot be computed from sensitivity and specificity alone.
- Confounding is a third factor associated with exposure and outcome; randomisation, restriction, matching and adjustment address it. Bias arises from selection, information or recall; effect modification is a real difference in effect across strata and is reported, not adjusted away.
- Health needs assessment uses Bradshaw's taxonomy: normative, felt, expressed and comparative need. Screening requires a valid test, an accepted treatment, and evidence that earlier detection improves outcome.
Worked example
A test has 90% sensitivity and 95% specificity; prevalence is 1% in 10,000 people. Diseased: 100, of whom 90 test positive. Non-diseased: 9,900, of whom 5% (495) are false positives. Positive predictive value = 90/(90 + 495) = 15.4%. The reasoning: most positives come from the large healthy group, so a good test still yields many false alarms when prevalence is low. Negative predictive value is 9,405/9,415 ≈ 99.9%.
Traps
Do not calculate PPV without prevalence; do not call prevalence a risk; do not interpret an odds ratio as a risk ratio when the outcome is common; do not adjust for a variable that is on the causal pathway or is an effect modifier.