Master of Public Health (MPH)

Welcome to Master of Public Health (MPH). Follow Causal Inference in Epidemiology: Chance, Bias, Confounding and the Bradford Hill Considerations, then practise in the learning system. Sign in and use the study header to open your authorised course. Select this published lesson. Start Course Tutor with Auto audio on: it teaches continuously, answers a typed question, then resumes. 1. Strength of association — larger relative risks are harder to explain by confounding. 2. Consistency — repeated in different populations, places, times and designs. Scenario (invented figures). A cohort study reports that alcohol drinkers have twice the risk of a lung disease. Crude table. Drinkers: 1,000 people, 30 cases (3.0%). Non-drinkers: 1,000 people, 15 cases (1.5%). Crude RR = 3.0 divided by 1.5 = 2.0. Suspect confounder: smoking. Drinkers in this population smoke more, and smoking independently causes the disease. Stratify: Stratum: Smokers; Drinkers: cases / total: 28 / 800; Risk: 3.5%; Non-drinkers: cases / total: 7 / 200; Risk: 3.5%; RR: 1.0. Stratum: Non-smokers; Drinkers: cases / total: 2 / 200; Risk: 1.0%; Non-drinkers: cases / total: 8 / 800; Risk: 1.0%; RR: 1.0. Check totals: drinkers 28 + 2 = 30 cases, 800 + 200 = 1,000 people; non-drinkers 7 + 8 = 15 cases, 200 + 800 = 1,000 people. Stratum risks: 28 divided by 800 = 0.035; 7 divided by 200 = 0.035; 2 divided by 200 = 0.010; 8 divided by 800 = 0.010. Interpretation. Within each smoking stratum the RR is 1.0, so the crude RR of 2.0 is entirely due to confounding by smoking. Drinking is associated with smoking (80% of drinkers smoke against 20% of non-drinkers), and smoking raises risk 3.5-fold in each drinking group. The Mantel–Haenszel pooled RR would also be 1.0. Contrast — effect modification. Suppose instead that the RR for an exposure were 5.0 among smokers and 1.2 among non-smokers. The strata differ from each other; smoking would be an effect modifier, and the correct report is the two stratum-specific RRs, not a single pooled figure. Applying Hill. For tobacco and oral cancer, evidence shows strength (large relative risks), consistency (many countries and designs), temporality (cohort studies), dose–response (risk rising with duration and quantity), plausibility (known carcinogens), and experiment (lower risk after cessation). The causal judgement is strong even without randomised trials. Ask Course Tutor to explain this step: Check totals: drinkers 28 + 2 = 30 cases, 800 + 200 = 1,000 people; non-drinkers 7 + 8 = 15 cases, 200 + 800 = 1,000 people. Say what you tried. Pause and try this source task. Crude OR is 3.0; stratum-specific ORs are 1.1 and 1.0. What does this suggest? The source answer is: Confounding by the stratifying variable; the adjusted association is close to null. Compare your method and result, correct an error, then continue. Use Stop to end Course Tutor.