Epidemiological study designs

Thinking about design

• Ask two questions of any study: did the investigator assign the exposure? (yes: experimental; no: observational) and in which direction does it look? (from exposure to outcome, from outcome back to exposure, or both at one moment).
• The answer fixes what the study can measure (prevalence, incidence, odds) and which biases threaten it.

Hierarchy of evidence (highest first)

• Systematic review and meta-analysis of randomised trials.
• Randomised controlled trial (RCT).
• Cohort study.
• Case-control study.
• Cross-sectional survey and ecological study.
• Case series and case reports.
• Expert opinion and mechanistic reasoning.

Descriptive designs

• Case report and case series: describe new diseases or drug effects (first AIDS cases, thalidomide); no comparison group, so no test of association.
• Ecological (correlational) study: compares groups or populations, not individuals (national salt intake against stroke rates). Quick and cheap, but prone to the ecological fallacy: a link between group averages may not hold for individuals.
• Cross-sectional (prevalence) survey: exposure and outcome measured at the same moment in a defined population, like a snapshot. It measures prevalence and is ideal for planning services (national nutrition and demographic surveys).
• Its key weakness for causation is temporality: in a one-day survey of smoking and chronic cough, you cannot tell which came first. Prevalence surveys also over-represent long-lasting cases and miss those who died or recovered quickly.

Exam trap A cross-sectional study cannot establish that exposure preceded disease, so it cannot show cause.

Case-control study

• Start with people with the disease (cases) and without it (controls), then look back for past exposure.
• Measure: odds ratio (OR), because incidence cannot be calculated when the investigator chose how many cases and controls to include.
• Strengths: best for rare diseases and long latency; quick, cheap, needs few subjects; can study many exposures.
• Weaknesses: recall bias (cases search memory harder), selection bias in choosing controls, no incidence or direct relative risk, poor for rare exposures.
• Matching controls for age and sex reduces confounding; a nested case-control study draws cases and controls from an existing cohort.
• Worked example: 40 of 100 lung cancer cases and 20 of 100 controls had asbestos exposure. Odds of exposure in cases = 40/60; in controls = 20/80. OR = (40 × 80) / (60 × 20) = 3,200 / 1,200 = 2.67.

Cohort study

• Start with people free of the outcome, classified by exposure, and follow them forward to see who develops it. This is the design for comparing individuals with and without a risk factor to assess the development of an outcome.
• Prospective (followed into the future) or retrospective/historical (past records, but still exposure-to-outcome direction).
• Measures: incidence, relative risk, attributable risk.
• Strengths: clear temporal sequence; many outcomes of one exposure; suits rare exposures (occupational hazards); less recall bias.
• Weaknesses: costly and slow, loss to follow-up (attrition bias), inefficient for rare diseases, exposure may change. Examples: Framingham Heart Study, British Doctors Study.

Randomised controlled trial

• The investigator allocates the intervention by chance.
• Randomisation removes selection bias in allocation and balances both known and unknown confounders between groups, so any difference in outcome can be attributed to the treatment. This is its main purpose.
• Allocation concealment hides the next assignment from recruiters, protecting randomisation at enrolment. Blinding acts after allocation: single blind (participants unaware), double blind (neither participants nor investigators/assessors know who gets what), triple blind (analysts too). Blinding prevents performance and detection bias; placebo makes it possible.
• Intention-to-treat analyses everyone in the group they were randomised to, even if they stopped or switched; it preserves randomisation and reflects real-world effectiveness. Per-protocol analysis includes only those who complied and can exaggerate benefit.
• Designs:

• Phases: I (safety and dose in healthy volunteers), II (efficacy and side effects in patients), III (large comparative RCTs for licensing), IV (post-marketing surveillance for rare adverse effects).
• Ethics: equipoise, informed consent, ethics committee approval, data safety monitoring, registration, CONSORT reporting.
• Field trials test preventive measures in healthy people (vaccines); community trials test whole communities (water fluoridation).

Vaccine efficacy

• Vaccine efficacy = (attack rate in unvaccinated − attack rate in vaccinated) ÷ attack rate in unvaccinated = 1 − relative risk.
• If 10% of unvaccinated and 2% of vaccinated people catch the disease: (10 − 2) ÷ 10 = 80%. It is the relative risk reduction produced by the vaccine.

Systematic review and meta-analysis

• A systematic review uses an explicit search strategy, inclusion criteria and quality appraisal (PRISMA, Cochrane).
• A meta-analysis pools results statistically, shown as a forest plot: each study is a square (size = weight) with its confidence interval line; the diamond is the pooled estimate; the vertical line is no effect (1 for ratios, 0 for differences).
• Heterogeneity (I²) decides fixed or random effects models; publication bias shows as an asymmetrical funnel plot.

Other designs

• Qualitative research (interviews, focus groups) explores experiences and meanings.
• Diagnostic accuracy study: new test and gold standard applied to the same patients (cross-sectional).
• Audit compares practice with an agreed standard (set standard, measure, change, re-audit); research creates new knowledge.
• Economic evaluation: cost-effectiveness (cost per natural unit), cost-utility (cost per QALY), cost-benefit (both in money), cost-minimisation.

Choosing a design

Remember Case-control goes from disease back to exposure (odds ratio); cohort goes from exposure forward to disease (relative risk).

Exam trap Randomisation balances unknown confounders; matching and restriction handle only the ones you know about.

How FCPS asks about this

• Main purpose of randomisation: removes selection bias and balances known and unknown confounders.
• Double-blind trial: neither participants nor investigators/assessors know the allocation.
• Crossover trial: each participant gets both treatments in turn, with a washout.
• Odds ratio from 40/100 exposed cases and 20/100 exposed controls: (40 × 80)/(60 × 20) = 2.67.
• Vaccine efficacy when attack rates are 10% unvaccinated and 2% vaccinated: 80%.
• Main limitation of a one-day survey for causation: cannot show which came first (no temporality).
• Design comparing people with and without a risk factor for later outcome: cohort study; rare disease points to case-control.