Study designs and the biases that break them
The hierarchy of study designs from case report to meta-analysis, which measure each yields, and the classic biases to name in the exam.
The ladder of evidence
Designs are ranked by how well they defend against bias and confounding. Climb the ladder and control improves; the strongest observational design still cannot match a randomised trial for causation.
| Design | What it does | Direction | Measure | Key strength / weakness |
|---|---|---|---|---|
| Case report / series | Describes one or a few patients | — | None | Hypothesis-generating only; no comparison group |
| Cross-sectional | Exposure and disease measured at one point | Snapshot | Prevalence, prevalence ratio | Fast, cheap; cannot tell which came first |
| Case-control | Start with disease, look back at exposure | Backward | Odds ratio (OR) | Good for rare diseases; prone to recall & selection bias |
| Cohort | Start with exposure, follow to outcome | Forward | Relative risk (RR), incidence | Gives incidence & temporality; slow, costly, loss to follow-up |
| RCT | Randomly allocate exposure/treatment | Forward | RR, risk difference | Randomisation controls known and unknown confounders |
| Meta-analysis / systematic review | Pools multiple studies | — | Pooled OR/RR | Top of the pyramid; only as good as its inputs (garbage in, garbage out) |
OR versus RR — the trap
You cannot calculate relative risk from a case-control study because incidence is unknowable when you have hand-picked the cases. Case-control gives an odds ratio; cohort and RCT give relative risk. When disease is rare, OR approximates RR — the exam's favourite rare-disease assumption.
Biases worth naming
| Bias | Where it bites | One-line tell |
|---|---|---|
| Selection bias | Case-control, cross-sectional | Study group not representative (e.g. Berkson's, healthy-worker effect) |
| Recall bias | Case-control | Cases remember exposures better than controls |
| Lead-time bias | Screening | Earlier diagnosis lengthens apparent survival without changing death date |
| Length-time bias | Screening | Screening preferentially catches slow, indolent disease |
| Confounding | All observational | A third factor linked to both exposure and outcome (age, smoking) |
Randomisation defeats confounding; blinding defeats observer and information bias; a prospective design defeats recall bias.
Anchor: Case-control → odds ratio, looks backward, best for rare disease. Cohort/RCT → relative risk, looks forward. Screening's two con-artists are lead-time and length-time bias.
Sources
- Park's Textbook of Preventive and Social Medicine, 27e
- Gordis Epidemiology, 6e