Epidemiology Study Designs
On this page
Direct answer
Study designs in epidemiology divide into descriptive studies, which portray disease by time, place and person and generate hypotheses, and analytical studies, which test hypotheses by quantifying exposure–outcome relationships. Case–control studies work backwards from outcome to exposure and yield an odds ratio; cohort studies follow exposed and unexposed forwards and yield relative risk; randomised controlled trials assign the exposure themselves and sit at the top of the evidence hierarchy; ecological studies compare populations rather than individuals and invite the ecological fallacy. The examinable core for NEET-PG PSM is the direction of enquiry, the measure of association each design produces, and its signature biases — temporality is strongest in a cohort, cost-efficiency in a case–control, and causal inference in a randomised trial.
What you must remember
- Descriptive studies: case reports, case series, cross-sectional surveys and surveillance data describe disease distribution and generate hypotheses; they usually cannot test them.
- Cross-sectional: exposure and outcome measured simultaneously — a prevalence snapshot that cannot establish which came first (the chicken-and-egg problem).
- Case–control: starts with cases and controls, computes an odds ratio (ad/bc), suits rare diseases, is quick and cheap, but is haunted by recall and selection bias and yields no incidence.
- Cohort: starts with exposure, follows for outcome, gives incidence and relative risk plus attributable risk; temporality is its unique strength, loss to follow-up its classic weakness; may be prospective or retrospective.
- Randomised controlled trial: randomisation, allocation concealment and blinding neutralise confounding and bias; phases run I (safety/pharmacokinetics), II (dose/efficacy signal), III (comparative efficacy), IV (post-marketing surveillance).
- Field versus community trials: field trials enrol healthy individuals (classic: vaccine trials), community trials randomise whole groups (classic: water fluoridation, salt iodisation).
- Ecological studies: unit of analysis is the group; correlation at population level cannot be applied to individuals — the ecological fallacy.
- Evidence hierarchy: meta-analysis and systematic reviews on top, then RCTs, cohorts, case–control, cross-sectional, case series and expert opinion at the base.
How to place a design in an exam vignette
Read the first sentence for the starting point. A study that begins with 200 women already diagnosed with cervical cancer and 200 without, then interviews both about lifetime sexual partners, is a case–control — the outcome selected the subjects, so only an odds ratio is legitimate, and the answers will be distorted by differential recall. A study that enrols 5,000 oral contraceptive users and 5,000 non-users, then tracks venous thromboembolism for ten years, is a prospective cohort — it can compute incidence in each arm, relative risk, and attributable risk, and it establishes that exposure preceded disease. A study that randomises hypertensives to two drugs is an RCT; one that randomises villages to a sanitation intervention is a community trial; one that correlates state-level per capita fat intake with state breast cancer mortality is ecological, and attributing individual risk from it is the trap.
Now layer the follow-up questions the examiner likes: which design suits a rare disease (case–control), a rare exposure (cohort), an outbreak in progress (case–control or cohort seeded from a line list), and which can calculate population attributable risk (cohort with known exposure prevalence). That single framing device answers most design questions asked in recent papers.
How the exam frames it
Two patterns recur. One gives a two-line study description and asks for the design or its measure of association; the reward goes to whoever identifies the sampling direction, not the disease. The second pattern tests vocabulary: nested case–control (cases and controls sampled from an established cohort, saving stored sera), case-cohort, and the difference between a clinical trial and a community trial. A favourite trick labels a retrospective cohort a "case–control" because it looks backwards in calendar time — but if exposure groups were defined first and outcomes ascertained afterwards, it remains a cohort. Time direction of data collection is not what defines the design; the sampling logic is.
Frequently asked questions
Which study design yields an odds ratio as its measure of association?
The case–control design, because subjects are sampled by outcome status; true incidence in the source population cannot be computed, so the cross-product ad/bc is used.
Why is relative risk not calculable in a case–control study?
Since the case and control numbers are chosen by the investigator, the denominators do not reflect the population at risk, making the risk ratio meaningless — though the odds ratio approximates it when the disease is rare.
Which design establishes temporality most convincingly?
The prospective cohort, because exposure is documented before the outcome occurs; the randomised trial shares this strength and additionally removes confounding by allocation.
What is a nested case–control study?
Cases arising within a followed cohort are matched to controls from the same cohort, exploiting stored baseline samples — cohort validity at case–control cost.
What is the ecological fallacy?
Inferring individual-level associations from group-level data, such as concluding that a person in a high-fat-consuming state personally has higher breast cancer risk because state averages correlate.