Public health, epidemiology and research methods in occupational medicine
Measuring disease in working populations
Epidemiology questions reward precision about denominators. Prevalence is existing cases in a population; incidence is new cases per person-time. Cohort studies give incidence and relative risk directly but are inefficient for rare outcomes; case-control studies give odds ratios cheaply but cannot yield incidence; cross-sectional studies give prevalence and associations, not causation; RCTs are rare in occupational settings for ethical and practical reasons.
The signature trap in occupational cohorts is the healthy worker effect: working populations are healthier than the general population, so an SMR below 100 may reflect selection rather than a protective exposure. Other biases to name: healthy worker survivor effect, selection at hire, information and recall bias (cases remember exposures more readily), and confounding by smoking, age, socioeconomic position and co-exposure. Adjust, stratify or match - and remember residual confounding.
Bradford Hill criteria (strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, analogy) are aids to judgement, not a checklist for proof; temporality is the only essential one.
Screening theory: sensitivity and specificity are properties of the test, but positive predictive value depends heavily on prevalence. In a low-prevalence population even a good test generates mostly false positives.
Worked example. A test with 90% sensitivity and 95% specificity in a population with 1% prevalence: of 100,000 people, 900 of the 1,000 cases test positive, and 5% of the 99,000 non-cases (4,950) also test positive. PPV = 900/5,850 ≈ 15% - roughly four in five positives are false.
Rule of thumb: always ask 'what is the denominator, and who was excluded from it?'