Research Study Designs in Health Sciences
A study design is the overall plan linking a research question to participant selection, measurement, comparison and analysis. Design choice determines which conclusions are defensible and which biases must be controlled.
Learning objectives
- Classify exploratory, descriptive, analytical and experimental designs.
- Compare cross-sectional, case-control, cohort and trial designs.
- Select a design that matches the question, ethics and resources.
- Identify major strengths, limitations and measures of association.
Classification by purpose
Exploratory
Clarifies a poorly understood issue, generates concepts and develops hypotheses.
Descriptive
Measures frequency, characteristics, distribution or trends without testing a causal comparison.
Analytical
Compares groups to examine associations and explanations.
Experimental
Assigns an intervention to test its effect under controlled conditions.
Core observational designs
| Design | Starting point | Main output | Major limitation |
|---|---|---|---|
| Cross-sectional | Population at one period | Prevalence and associations | Temporal sequence often unclear |
| Case-control | People with and without outcome | Odds ratio | Selection and recall bias |
| Cohort | Exposed and unexposed groups | Incidence, risk/rate ratio | Time, cost and loss to follow-up |
| Ecological | Groups or populations | Group-level association | Ecological fallacy |
Cross-sectional scenario
A survey measures current burnout and working hours among hospital clinicians. It can estimate burnout prevalence and test association with working hours, but cannot prove long hours preceded or caused burnout. It may miss absent workers with severe burnout.
Case-control scenario
Investigators select newborns with sepsis as cases and newborns without sepsis as controls, then compare prior maternal and delivery exposures. This is efficient for an uncommon outcome. Controls must represent the exposure distribution of the population that produced the cases.
Cohort scenario
Pregnant women with and without anaemia are followed to delivery to compare low-birth-weight incidence. Exposure precedes outcome, multiple outcomes can be studied, and absolute risks are estimable. Confounding and loss to follow-up still threaten validity.
Experimental and quasi-experimental designs
| Design | Allocation | Example |
|---|---|---|
| Randomised controlled trial | Random assignment | New reminder system versus usual care |
| Cluster trial | Facilities or communities randomised | Hand-hygiene programme across wards |
| Before-and-after | No concurrent randomised control | Waiting times before and after triage redesign |
| Interrupted time series | Repeated measures before and after intervention | Monthly antibiotic use around a stewardship policy |
Randomisation, allocation concealment and blinding
- Randomisation balances known and unknown prognostic factors on average.
- Allocation concealment prevents foreknowledge of the next assignment during enrolment.
- Blinding reduces performance or assessment bias after allocation.
- Intention-to-treat analysis preserves original assignment in analysis.
Design-selection matrix
Choose cross-sectional for prevalence; case-control for rare outcomes or long latency; cohort for incidence, prognosis and multiple outcomes; trials for ethically assignable interventions; qualitative designs for meaning and experience; mixed methods when numerical patterns require contextual explanation.
Validity threats
- Selection bias from non-comparable recruitment or follow-up.
- Information bias from inaccurate exposure or outcome measurement.
- Confounding from common causes of exposure and outcome.
- Contamination when comparison groups receive the intervention.
- History and secular trends in uncontrolled before-and-after studies.
Worked design exercise
Question: Does household air pollution increase childhood pneumonia? A cross-sectional study is fast but temporality may be unclear. A case-control study efficiently compares prior exposure in pneumonia cases and controls but may suffer recall bias. A cohort establishes exposure before incident disease but requires more time and resources. The best choice depends on available data, disease frequency, measurement quality and feasibility.
Common mistakes
- Calling a cross-sectional association an “effect”.
- Selecting controls who could not have become cases.
- Ignoring clustering in facility- or community-level interventions.
- Using random sampling and claiming the study is a randomised trial.
- Choosing design labels that do not match actual timing and procedures.
Review questions
- Compare cross-sectional, case-control and cohort designs.
- Differentiate random selection from random allocation.
- When is a cluster trial preferable?
- Design a study of risk factors for a rare occupational cancer.
Return to the Clinical Medicine Year 2 curriculum
References
- WHO. Health Research Methodology.
- Rothman KJ. Epidemiology: An Introduction.
- UAHEB. Guidelines on Research Proposal and Report Writing.
Educational content only. Study protocols require appropriate scientific and ethics review.
