Core Curriculum
Module 2: Study Design & Research Methods
Core Topics
From descriptive studies to randomized trials, systematic reviews, and qualitative research — the foundational toolkit for generating and interpreting medical evidence.
📝 2.1 Descriptive Studies
Case reports (single patient) and case series (collection of similar patients) describe novel diseases, rare adverse events, or unusual presentations. They generate hypotheses but cannot establish causation due to lack of comparison groups.
📌 Examples: First AIDS cases (1981), thalidomide birth defects. Lowest level of evidence, but essential for early investigation.
🌍 2.2 Observational Studies: Cross‑Sectional & Ecological
Cross‑Sectional (Community Surveys)
• "Snapshot" assessing exposure and outcome simultaneously
• Measures prevalence and prevalence ratios
• Cannot establish temporality (chicken‑or‑egg)
• Examples: NHANES, BRFSS
Ecological Studies
• Population‑level comparisons (countries, regions)
• Unit of analysis = groups, not individuals
• Ecological fallacy: assuming group‑level association applies to individuals
• Useful for hypothesis generation
🔄 2.3 Case‑Control Studies
Design: Start with outcome (cases vs. controls), look backward at exposure. Efficient for rare diseases, multiple exposures, long latency.
Biases: recall bias, selection bias (control selection critical). Matching controls for confounders (age, sex) is common. When disease rare (<10%), OR approximates Relative Risk.
⏳ 2.4 Cohort Studies
Design
• Start with exposure, follow for outcome
• Prospective (current) or retrospective (historical records)
• Can calculate incidence and relative risk
Relative Risk (RR)
RR = [a/(a+b)] / [c/(c+d)]
RR = 1 → no association; RR > 1 → increased risk; RR < 1 → protective.
Strengths: establishes temporality, minimizes recall bias, can study multiple outcomes. Limitations: expensive, loss to follow‑up, inefficient for rare diseases. Landmark examples: Framingham Heart Study, Nurses’ Health Study.
🎲 2.5 Experimental Studies: Randomized Controlled Trials
Gold standard for causality. Randomization balances known/unknown confounders; blinding (single/double/triple) reduces performance and detection bias. Allocation concealment prevents selection bias.
Trial Designs
• Parallel: one intervention per participant
• Crossover: participants receive both (washout period)
• Cluster: groups randomized
Trial Goals
• Superiority: new better than control
• Non‑inferiority: new not worse by margin
• Equivalence: similar within margin
📊 Analysis: Intention‑to‑Treat (ITT) preserves randomization; Per‑Protocol may overestimate efficacy.
🏘️ 2.6 Community & Cluster Randomized Trials
Randomization of groups (hospitals, schools, communities) when individual randomization is infeasible or contamination risk is high. Intracluster correlation (ICC): individuals within clusters are more similar → reduces effective sample size. Design effect = 1 + (m‑1)×ICC.
📌 Examples: school‑based nutrition programs, hospital hand hygiene campaigns, community water fluoridation. Stepped‑wedge designs (all clusters eventually receive intervention) are ethical and efficient.
📚 2.7 Systematic Reviews & Meta‑Analysis
Systematic review: comprehensive, reproducible synthesis of all available evidence on a focused question (PICO). Meta‑analysis: statistical combination of results, producing a pooled estimate.
Forest Plot
• Each study: square (point estimate) and horizontal line (95% CI)
• Diamond = pooled estimate
• CI crossing null (1.0) → not significant
Heterogeneity (I²)
• I² < 25%: low; 25‑50%: moderate; 50‑75%: substantial; >75%: considerable
• Funnel plot asymmetry suggests publication bias.
🎯 2.8 Study Conduct & Sampling
Sampling methods: probability (simple random, stratified, cluster) vs. non‑probability (convenience, purposive). Inclusion/exclusion criteria define the target population and affect generalizability.
Intention‑to‑Treat (ITT): analyze participants according to randomized group regardless of adherence. Preserves randomization, reflects real‑world effectiveness (conservative).
Per‑Protocol: analyzes only adherent participants; may overestimate efficacy.
📌 Missing data: multiple imputation preferred; complete case analysis and last observation carried forward (LOCF) have limitations.
💬 2.9 Qualitative Research Methods
Explores experiences, meanings, and perspectives using non‑numerical data. Methods include in‑depth interviews, focus groups, ethnography, document analysis. Rigor assessed by credibility, transferability, dependability, confirmability (triangulation).
Applications
• Patient illness experiences
• Barriers to care
• Implementation research
• Understanding health behaviors
Mixed Methods
Combine qualitative and quantitative approaches for comprehensive understanding (sequential or concurrent).