Epidemiology Study Guide

📖 Core Concepts Epidemiology – Study of how health‑related states are distributed in who, when, where and why they occur. Descriptive vs. Analytic vs. Experimental – Descriptive (time, place, person); Analytic (test exposure‑outcome hypotheses); Experimental (intervention with control). Study Designs – Case‑control – Retrospective; participants selected by disease status. Cohort – Prospective (or retrospective) follow‑up of exposure groups. Randomized Controlled Trial (RCT) – Participants randomly assigned to intervention or control. Measures of Association – Odds Ratio (OR) for case‑control: $\displaystyle \text{OR}= \frac{AD}{BC}$ (A=exposed cases, B=exposed controls, C=unexposed cases, D=unexposed controls). Relative Risk (RR) for cohort: $\displaystyle \text{RR}= \frac{Pe}{Pu}$ where $Pe=\frac{A}{A+B}$, $Pu=\frac{C}{C+D}$. Bias & Error – Random error (sampling variability) vs. systematic error (bias). Key bias types: selection, information (recall), immortal time, confounding. Causal Inference – Correlation ≠ causation; use Bradford Hill criteria and causal‑pie model. Epidemiologic Triad – Interaction of host, agent, environment in outbreak analysis. --- 📌 Must Remember Temporality is the only non‑negotiable Hill criterion – exposure must precede outcome. OR > 1 ⇒ exposure is a risk factor; OR < 1 ⇒ protective. Same interpretation for RR. Recall bias inflates OR when cases remember exposures better than controls. Selection bias matters only when the factor influencing participation is related to both exposure and outcome. Immortal time bias → artificially low event rates in “exposed” group; avoid by proper person‑time assignment. Confounding exists when a third variable is associated with both exposure and outcome and is not on the causal pathway. Confidence interval (CI) width = precision; narrower CI → higher precision (less random error). Age adjustment = standardize rates to a common age distribution for fair comparison. --- 🔄 Key Processes Constructing a 2 × 2 Table (Case‑Control) Populate A, B, C, D → compute OR = AD/BC. Calculating RR (Cohort) Compute incidence in exposed $Pe = A/(A+B)$, in unexposed $Pu = C/(C+D)$ → RR = $Pe/Pu$. Assessing Bias Identify potential sources → evaluate direction (toward/away from null) → apply design or analytic controls (randomization, matching, stratification, multivariable modeling). Applying Bradford Hill Criteria (quick checklist) Strength → Consistency → Specificity → Temporality → Biological gradient → Plausibility → Coherence → Experiment → Analogy. Age Adjustment (Direct Standardization) Multiply age‑specific rates by a standard population age distribution → sum to obtain age‑adjusted rate. --- 🔍 Key Comparisons Case‑control vs. Cohort Selection: disease status vs. exposure status. Direction: retrospective vs. prospective. Measure: OR vs. RR. Cost/Time: faster/cheaper vs. more expensive, longer. Selection Bias vs. Information Bias Source: who enters the study vs. how variables are measured. Effect: distorts exposure–outcome relationship vs. misclassifies exposure/outcome. Random Error vs. Systematic Error (Bias) Nature: chance variability, reduced by larger sample → precision. Nature: consistent deviation from truth, not fixed by sample size. --- ⚠️ Common Misunderstandings “OR = RR” – Only true when the outcome is rare (<10% prevalence). “A non‑significant p‑value proves no association” – May reflect low power or wide CI, not absence of effect. “Recall bias always overestimates risk” – It can under‑estimate if cases forget exposures more than controls. “Confounding is a type of bias” – Conceptually different: confounding is a mixing of causal effects; bias is measurement error. “Age adjustment changes the true disease rate” – It does not; it creates a comparable summary across populations. --- 🧠 Mental Models / Intuition Causal Pie – Think of a disease as a pie that requires several “slices” (component causes); removing any slice (e.g., a modifiable exposure) can prevent the whole pie. Triad Triangle – Visualize an outbreak as a triangle; shifting one corner (host immunity, agent virulence, environmental conditions) can tip the balance. Bias as a Lens – Random error blurs the picture; systematic error (bias) tilts the lens, consistently shifting the view away from truth. --- 🚩 Exceptions & Edge Cases OR ≈ RR when disease incidence < 10 % (rare disease assumption). Immortal time bias most common in pharmaco‑epidemiology where drug exposure is defined after cohort entry. Confounding by indication – occurs in observational treatment studies where disease severity drives treatment choice. Selection bias without outcome association – harmless; only problematic when selection relates to both exposure and outcome. --- 📍 When to Use Which Choose Case‑Control when the disease is rare, latency is long, or rapid results are needed. Choose Cohort when you need temporality, can measure incidence, or when exposure is common. Use RCT for evaluating efficacy of an intervention where ethical randomization is feasible. Apply Age Adjustment when comparing disease rates across populations with differing age structures. Use Mendelian Randomization when you have genetic instruments to infer causality and want to avoid confounding. --- 👀 Patterns to Recognize “Exposure → Dose‑Response → Stronger Association” → supports biological gradient criterion. Consistent direction of effect across diverse study designs → signals consistency (Hill). Large OR/RR with narrow CI → high strength & precision → strong causal hint. Differential loss to follow‑up clustered in one exposure group → red flag for selection bias. --- 🗂️ Exam Traps Mistaking OR for RR in common outcomes – distractor answers often present OR values as if they were RRs. Choosing “recall bias” for a prospective cohort – recall bias is an information bias typical of retrospective designs. Selecting “selection bias” when participation differs but is unrelated to outcome – such a scenario does not bias the estimate. Confusing “immortal time” with “lead‑time bias” – immortal time bias inflates protective effect; lead‑time bias relates to earlier detection, not person‑time classification. Answer choices that list all Hill criteria except “temporality” – temporality is mandatory; its omission signals a wrong answer. ---