Epidemiology Exam 3 Comprehensive Review Notes

Risk Difference (RD) / Attributable Risk (AR)

• Definition – absolute measure: excess incidence in the exposed versus non-exposed.
• Formula: RD = AR = IE - I{NE} = \frac{a}{a+b} - \frac{c}{c+d}
• Components (Fig 12-1C)
– IE = background risk + risk attributable to exposure. – I{NE} = background risk only.
• Interpretation (requires significant association):
– AR > 0 ⇒ number of cases among exposed attributable to exposure; cases prevented if exposure removed.
– AR = 0 ⇒ no association (RR or OR = 1).
• Clinical & public-health value: quantifies disease burden and potential impact of removing exposure; less useful for etiologic (causal) research.

Example – Benign Breast Disease (BBD) → Invasive Cancer

• 2×2 table (N = 300):
– BBD +: 10 cases / 90 non-cases (100)
– BBD –: 4 cases / 196 non-cases (200)
• Risks: IE = 10/100 = 0.10; I{NE} = 4/200 = 0.02
• RD = 0.10 - 0.02 = 0.08 ⇒ 8 per 100 excess cancers among women with BBD.
• Interpretation: among every 100 women with BBD, 8 of the 10 cancers are due to BBD history.

Attributable Risk Percent (AR %) – Etiologic Fraction (EF)

• Relative measure: proportion of cases among exposed that is due to exposure.
• Cohort: AR\% = \frac{RD}{I_E} \times 100\% = \left(1 - \frac{1}{RR}\right) \times 100\%
• Case-control (cannot measure incidence): AR\% = \left(1 - \frac{1}{OR}\right) \times 100\%
• BBD example: AR\% = 0.08/0.10 = 0.80 \rightarrow 80\% or 1 - 1/5 = 0.80.

Preventive Fraction (PF)

• Used when exposure is protective ( IE < I{NE} ).
• Cohort: PF = \frac{I{NE}-IE}{I_{NE}} \times 100\% = (1-RR) \times 100\%
• Case-control: PF = (1-OR) \times 100\%.

Case-control PF Example (Exercise & MI)

• Cases (MI): 236 exercisers / 764 non-exercisers.
• Controls: 379 exercisers / 621 non-exercisers.
• Design: case-control; compute OR, then PF.

Relative Risk (RR/OR) vs. Attributable Risk (AR)

• RR / OR – strength of association, causal inference.
• AR – absolute impact, number of cases preventable; used after causality assumed.
• Cohort: RR = IE/I{NE}; AR = IE - I{NE}.
• Case-control: OR = ad/bc; cannot compute AR directly.

Study Designs

Observational vs Experimental

• Experimental (Randomized Controlled Trial – RCT)
– Manipulate exposure, randomize subjects, strongest for cause-effect.
– Evaluate efficacy/effectiveness of interventions.
– More rigorous, but ethical/feasibility limits, costly, limited generalizability if eligibility strict.
• Observational
– No randomization; used for etiologic studies when experiments impossible.
– More potential bias/confounding.

Randomization – Purpose & Design

• Primary purpose: eliminate conscious/unconscious selection bias & control unmeasured confounders.
• Simple, stratified (block) randomization improves comparability.
• Trial flow: Defined population → randomize → new vs. current tx → outcomes.

Phases of Clinical Trials

• Phase I: safety, dosage, toxicity (5-60 healthy).
• Phase II: preliminary efficacy, short-term safety (100-300 patients).
• Phase III: full-scale RCT for effectiveness (1,000-3,000); basis for FDA approval.
• Phase IV: post-marketing surveillance – rare/long-term effects.

RCT Methodological Features

• Blinding (masking): single, double; reduces information bias.
• Placebo: identical inert pill; increases compliance, allows blinding.
• Run-in period: pre-randomization compliance test.
• Washout period (in crossovers): remove residual drug effects.
• Attrition → threatens generalizability & validity.
• Ethical issues: equipoise, informed consent, withholding effective therapy.

Strengths & Weaknesses of RCTs

• Strengths: randomization, blinding, controlled exposure, causal inference, rapid results.
• Weaknesses: cost, ethics, feasibility, rare outcomes, compliance, generalizability.

Measuring Treatment Effect

• Efficacy (% ↓ risk): \frac{Risk{control} - Risk{treatment}}{Risk_{control}} \times 100\%
• Drug X vs placebo example: risks 0.162 vs 0.104 → 36\% mortality reduction.

Intention-to-Treat (ITT)

• Analyze participants in original randomized groups regardless of adherence/crossover.
• Preserves benefits of randomization; provides conservative estimate of effectiveness.
• Including non-compliers gives less optimistic (not overly optimistic) efficacy.

RCT Design Variants

• Parallel: each group receives one treatment concurrently.
• Crossover: each subject receives all treatments sequentially; serves as own control; needs washout.
• Factorial: test ≥2 interventions simultaneously; participants receive combinations.

Placebo Considerations

• Serious use when outcome subjective (e.g., pain).
• Less acceptable when outcome is survival/death.

Statistical Inference – p-Value & Confidence Interval (CI)

• p-value: probability observed (or more extreme) result due to chance if null true; p<0.05 ⇒ significant.
• CI: range of plausible true values accounting for sampling variability.
– 95 % CI contains true parameter 95 % of repeated samples.
– Narrow CI = high precision.
– If CI includes 1 (for RR/OR) → not significant.
• Example: OR = 1.6; CI 0.4–2.8 → not significant (p ≥ 0.05).

Bias

Selection Bias (unequal selection probabilities)

• Case-control: Berkson, prevalence-incidence (Neyman).
• Cohort: non-response, loss to follow-up.
• Avoidance: select cases/controls independently of exposure.

Information Bias (misclassification)

• Recall, interviewer, reporting, surveillance, measurement error.

Misclassification

• Non-differential: same error rate → bias toward null (underestimate).
• Differential: different rates → bias toward or away from null; may create or hide association.
• Example (breast-cancer & diet): correcting 30 % under-reported exposure changed OR from 1.0 → 2.3 (bias away from null originally underestimated).

Confounding

• Third variable associated with exposure & disease, not on causal pathway.
• Assessment: stratified ORs ≈ each other but differ from crude by > 15 %.
• Control – design: randomization, restriction, matching; analysis: stratification, multivariable regression.
• Quiz: “correlated with exposure but not with disease” → FALSE (must be risk factor for disease).

Effect Modification (Interaction)

• Association varies by levels of a third factor; natural phenomenon to report.
• Detection: stratified measures differ > 15 % between strata.
• Types: synergism (positive), antagonism (negative).
• Example: crude OR 8.4; non-drinkers 2.2 vs drinkers 14.5 → alcohol is effect modifier.

Interaction Models

• Additive (AR): no interaction ⇒ AR{AB} = ARA + ARB. • Multiplicative (RR): no interaction ⇒ RR{AB} = RRA \times RRB.
• Practice tables provided (risks 3/7/8).
– Additive: combined risk 12.
– Multiplicative: combined risk 18.6.

Cohort Example – Omega-3 & AMD Progression

• Initial RR: 0.1/0.8 = 0.125 (protective).
• After re-classification (20 % progressed): RR = 0.375 (bias toward null; protection appears weaker).

Hill’s Causal Criteria (selected)

• Temporality (required) – cohort best design.
• Strength (magnitude of RR).
• Biological gradient (dose-response).
• Consistency (replication across studies).
• Figure shown depicts biological gradient.

Exam 3 Preparation Tips

• Review lecture & review slides (notes section).
• Practice RQ, HW, exercise problems, especially calculations.
• Consult exam guideline module; email TA/professor with questions.