Study designs: how to tell them apart
- Concept
Three questions name almost any design: who assigned the exposure, is there a comparison group, and where in time does the study start?
What it is
A study design is the plan for whom to study, what to measure and in what order. Observational studies only watch what already happens; in experimental studies the investigator assigns the intervention. Descriptive studies count cases without a comparison group, while analytic studies compare groups to test a hypothesis. Recognising the design is the first step to knowing which measure of association to use and which biases to look for.
Worked example
- Vignette: 800 bus drivers without hypertension are recruited, 400 on night shifts and 400 on day shifts, and followed for 5 years.
- Question 1: did the investigator assign the shift? No, it already existed: the study is observational.
- Question 2: is there a comparison group? Yes, day shift: it is analytic.
- Question 3: where does it start? From the exposure (shift) towards a future outcome: a prospective cohort.
- Data: 60 of 400 night-shift and 40 of 400 day-shift drivers develop hypertension.
- Risks: 60/400 = 0.15 (15%) and 40/400 = 0.10 (10%).
- RR = 0.15 / 0.10 = 1.5.
- Reading: in this cohort, night shifts were associated with 1.5 times the 5-year risk of hypertension.
Designs side by side
| Design | Starting point | Direction in time | Main measure |
|---|---|---|---|
| Case report or series | Patients with the outcome | None (description) | Counts and percentages |
| Ecological | Population groups | Usually one point in time | Correlation between rates |
| Cross-sectional | A sample of a population | One point in time | Prevalence, prevalence ratio |
| Case-control | The outcome (cases and controls) | Backwards, to exposure | Odds ratio (OR) |
| Cohort | The exposure (exposed and unexposed) | Forwards, to the outcome | Incidence, relative risk (RR) |
| Randomised controlled trial | An intervention assigned at random | Forwards | RR, absolute risk reduction, NNT |
Markers
- RR
- OR
- RCT
In practice
- In an exam, first find who decided the exposure, then look for a comparison group, and last check the direction in time.
- A cohort can be retrospective (historical): exposure and outcome have already happened, but the analysis still runs from exposure to outcome.
- A design is named by how participants were selected, not by the date the data were analysed.
Common pitfalls
- Mixing up case-control and retrospective cohort: case-control selects by outcome; a cohort selects by exposure.
- Calling any study with a control group experimental: without assignment by the investigator it is still observational.
- Computing an RR in a case-control study: there the investigator fixes the number of cases, so there is no incidence.
Test yourself
Think of the answer, then tap the question to see it.
Researchers enrol 100 patients with a rare cancer and 200 without it and ask about past exposures. What design is this?
A case-control study: it starts from the outcome and looks back at exposure.
What separates an experimental study from an observational one?
In an experimental study the investigator assigns the exposure or intervention; in an observational study they only record what happens.
Why is a cross-sectional study weak for showing causation?
Because it measures exposure and outcome at the same time and does not show which came first.
Which measure of association goes with a cohort study?
The relative risk (or the rate ratio), because a cohort measures incidence in the exposed and unexposed.
Connections
Educational content on epidemiological and statistical methods; the examples use made-up numbers to teach the calculation. Not a clinical or legal guide: current Colombian rules are in the official sources.
Sources: Principles of Epidemiology | Lesson 1 - Section 7 (Centers for Disease Control and Prevention, 2012); Principles of Epidemiology | Lesson 1 - Section 6 (Centers for Disease Control and Prevention, 2012); Chapter 24: Including non-randomized studies on intervention effects (Cochrane, 2024).
Spotted a mistake?We are always improving for the community. Tell us what is wrong and our team will review it.Report a mistake
Representative model, for educational purposes only.