By the end of this session, you should be able to:

• Explain target, source, study, and sample populations.
• Distinguish reference population from study population.
• Describe cross-sectional, case-control, and cohort studies.
• Explain the basic logic of experimental studies and trials.
• Select a suitable design for a clear research question.

🎯 Core Message: A good study design starts with a clear population and a clear question.

💡 Key Principle: Design is not chosen because it is popular. It is chosen because it best answers the research question.

• Population tells us who the findings are about.
• It defines who can be included and who cannot.
• It helps us judge whether results can be applied elsewhere (generalisation).
• It prevents confusion between people available and people of interest.

📝 Exam Tip: Always name your populations by person, place, and time. Vague terms like "community members" or "patients" lose marks. Be specific: "Children aged 6 to 59 months in Village X, July 2026."

⚠️ Critical Rule: Exclusion criteria must be applied consistently and before data collection. Changing rules mid-study introduces selection bias and destroys validity.

• Saying "community members" without naming place or time.
• Using facility attendees to represent people who never attend facilities.
• Changing eligibility rules during data collection.
• Ignoring non-response and missing records.
• Reporting sample findings as if they cover everyone.

Research Question: Is mosquito-net use associated with malaria fever among hostel students during July 2026?

📝 Exam Tip: In a scenario question, always identify all four population levels explicitly. This shows you understand the hierarchy from broad interest to actual data.

The reference population answers: "To whom are we trying to generalise?"

• It is the group to which study findings are intended to apply.
• It is often similar to the target population.
• It should be named before the study starts — not invented afterwards to make the study sound more important.
• It helps readers judge whether results are relevant to their setting.

Example: "Mothers attending ANC in urban public facilities in Uganda." A study conducted at Mulago Hospital could reasonably generalise to this group. It could NOT generalise to rural mothers who never attend ANC.

The study population answers: "Who is actually eligible in our study?"

• It is the specific group from whom data are collected.
• It must be described by eligibility, setting, and time.
• It is usually narrower than the reference population.
• It determines internal validity (are the findings true for this group?) and practical feasibility (can we actually do this?).

Example: "ANC mothers attending Mulago outpatient ANC clinic from July to August 2026, who are ≥18 years old, speak English or Luganda, and provide written consent." This is precise, measurable, and reproducible.

These two concepts are the backbone of study quality. Every study must balance them.

🩺 The Situation: During outreach, nurses screen adults who attend a health camp. 18 of 80 have high blood pressure readings. Define the populations and discuss validity:

• Study population: Adults who attended and were screened at the health camp. This is precise and measurable.
• Reference population: May be "all adults in that community" — but this is a stretch.
• The problem: People who attend outreach may differ from those who stay home. Attendees may be:
  • More health-conscious (they came for screening).
  • Older or retired (they had time to attend).
  • Living closer to the venue (better access).
  • Female (more likely to seek preventive care in many cultures).

Interpretation must mention this limitation: "Our findings apply to adults who attended the health camp. They may not represent adults who never attend outreach, especially younger men and those living far from services."

⚠️ Exam Trap: Never claim your sample represents "the community" just because you did outreach. Always question: "Who did NOT come?" Those missing people may be the most important.

Research question: "Among children under five in Village A, is untreated drinking water associated with diarrhoea during July 2026?"

In observational studies, the researcher does not assign the exposure. Participants are observed as they naturally are. The researcher simply measures what already exists. These designs are essential when experiments are unethical, costly, or impractical.

📝 Key Idea: In observational studies, we observe exposure and outcome without controlling who receives the exposure. We cannot randomise people to "smoke" or "drink contaminated water" — that would be unethical. So we watch what happens naturally.

Definition: A cross-sectional study measures exposure and outcome at the same time — like a photograph. It provides a snapshot of a population at one point in time.

• Estimating prevalence — how common a disease or risk factor is right now.
• Describing the distribution of health problems by person, place, and time.
• Planning health services ("How many hypertensive patients do we have?").
• Generating hypotheses for future research.

Example: Survey students today to record net use and malaria fever history. You ask: "Do you sleep under a net?" AND "Have you had fever in the past 2 weeks?" Both questions are answered at the same time.

Past ➔ [MEASURE NOW] ➔ Future
Exposure and outcome are measured together in the same survey or short period.

Definition: A case-control study begins with the outcome. You find people who HAVE the disease (cases) and people who DO NOT have the disease (controls), then look backward in time to compare their past exposures.

• Rare diseases — you do not need to follow thousands of people; you just find the few who already have the disease.
• Outbreak investigations — "What did the sick people eat that the healthy people did not?"
• Diseases with long latency — e.g., cancer (it would take decades to follow people in a cohort study).
• When you need quick answers with limited resources.

Example: Compare pupils with diarrhoea (cases) and pupils without diarrhoea (controls) by asking about their water-tank use in the past week. You start with the outcome (diarrhoea yes/no) and look back for the exposure (tank water yes/no).

[LOOK BACK] Compare previous exposure ➔ [START HERE] Cases (Outcome +) vs. Controls (Outcome −)

📝 Exam Tip — Recall Bias: This is the biggest weakness of case-control studies. A mother whose child died from diarrhoea may remember every detail of what the child ate and drank. A mother whose child is healthy may not remember what her child ate last week. This differential memory creates a false association. Always mention recall bias when critiquing a case-control study.

Definition: A cohort study begins with exposure. You find people who ARE exposed and people who ARE NOT exposed, then follow them forward in time to see who develops the outcome. A "cohort" is simply a group of people who share a common characteristic.

• Prospective cohort: You identify exposed and unexposed people NOW and follow them INTO THE FUTURE to see who gets the disease. This is the "gold standard" of observational studies.
• Retrospective cohort: You look BACKWARD in time using existing records. You find people who were exposed or unexposed in the past and check whether they already developed the outcome. Faster than prospective, but depends on good records.

• Measuring incidence and risk — how many new cases occur over time.
• Establishing temporal sequence — exposure is measured before outcome, so causation is more plausible.
• Studying multiple outcomes from one exposure (e.g., smoking causes lung cancer, heart disease, COPD, and stroke).
• Studying the natural history of disease.

Example: Follow 100 students who sleep under nets (exposed) and 100 students who do not (unexposed) for one semester. Count how many in each group develop malaria. You start with exposure and move forward to outcome.

[START WITH EXPOSURE] Exposed vs. Unexposed ➔ [FOLLOW FORWARD] ➔ Outcome? YES / NO

📝 Exam Tip — Loss to Follow-Up: This is the biggest weakness of cohort studies. If 30% of your exposed group drops out (because they got sick, moved away, or died), your results are biased. The people who remain may be healthier or more compliant than those who left. Always mention loss to follow-up when critiquing a cohort study.

Use this decision table to match your research question to the best design:

In experimental studies, the researcher assigns an intervention or exposure. This is the key difference from observational studies. The researcher actively manipulates the exposure and then compares outcomes between groups.

• The goal is to test whether an intervention causes an effect.
• Experimental designs provide the strongest evidence of causation.
• They are essential for evaluating new drugs, vaccines, health education programs, and clinical protocols.
• Ethics and safety are central. You cannot experiment on people without ethical approval, informed consent, and careful monitoring.

💡 Core Idea: Assign ➔ Compare ➔ Follow up ➔ Interpret.

The gold standard of experimental designs. Participants are randomly allocated to either the intervention group or the control group. Randomisation means chance decides group allocation — not the researcher's preference, not the patient's choice, and not convenience.

Eligible Participants ➔ [RANDOMISATION] ➔ Intervention Group & Control Group ➔ [FOLLOW UP] ➔ Measure Outcome ➔ [COMPARE] Is the difference statistically significant?

• It balances known confounders (age, sex, income) between groups.
• It also balances unknown confounders — factors you did not think to measure.
• It eliminates selection bias — the researcher cannot put sicker patients in the intervention group because they "need it more."
• It allows you to say: "Any difference in outcome is likely due to the intervention, not to pre-existing differences between the groups."

📝 Exam Tip: When asked "Why is randomisation useful?" always mention: (1) it balances known confounders, (2) it balances unknown confounders, and (3) it eliminates selection bias. These three points earn full marks.

Not all interventions can be randomised. Quasi-experimental designs are used when randomisation is impossible, unethical, or impractical.

• An intervention is introduced, but participants are not randomly assigned.
• Common in service delivery and quality improvement settings.
• Examples: before-and-after studies (measure outcomes before and after an intervention in the same group) and controlled before-and-after studies (compare one group that got the intervention with another group that did not, measured at two time points).
• Interpretation challenge: Other changes may have occurred at the same time (e.g., a national health campaign, seasonal change). You cannot be sure the intervention alone caused the difference.

Example: Measure hand hygiene compliance before and after a ward training intervention. If compliance rises from 40% to 75%, was it the training? Or was it a new infection control policy announced at the same time? Or increased supervision? Quasi-experimental designs cannot fully separate these effects.

Experimental studies involve active manipulation of people's health. Ethics are not optional — they are mandatory.

• Informed consent: Participants must understand what the study involves, the risks, the benefits, and their right to withdraw. Consent must be voluntary — no coercion.
• Reasonable chance of benefit: The intervention should be based on prior evidence that it might work. You cannot test something that is known to be harmful or useless.
• Minimise risks: Risks must be as low as possible. Participants must be monitored for adverse effects.
• No denial of essential care: The control group cannot be denied care they would normally receive. If there is an effective treatment, the control group gets it (the new intervention is tested against standard care, not against nothing).
• Confidentiality and respectful care: Participant data must be protected. Identities must not be revealed. Participants must be treated with dignity.
• Ethical review: All experimental studies must be approved by an Institutional Review Board (IRB) or Ethics Committee before starting.

⚠️ Critical: You cannot randomise people to a harmful exposure (e.g., "smoking group" vs. "non-smoking group"). That would be unethical. For harmful exposures, use observational studies.

🩺 Research Question: "Does a handwashing education package reduce diarrhoea among pupils over one school term?"

• Cluster RCT: Randomise entire schools (not individual pupils) to intervention or control. This prevents contamination — pupils in the intervention group might teach handwashing to control-group friends if they are in the same school.
• Quasi-experimental: If you cannot randomise schools, use a before-and-after design in the same schools, comparing diarrhoea rates in Term 1 (before) vs. Term 2 (after intervention).

💡 Key Principle: You cannot do an RCT for everything. You cannot randomise people to "smoking." For harmful exposures, use observational studies. For testing new treatments, use RCTs.

Use PICO to break down any research question:

💡 Mnemonic: "Please Identify Clear Outcomes" = PICO.

• Design: Cross-sectional.
• Reason: Measures prevalence at a single point in time.
• Limitation: Attendees may differ from non-attendees (selection bias).

• Design: Case-control.
• Reason: Starts with outcome (diarrhoea), looks back for exposure (tank water).
• Limitation: Recall bias — parents of sick children may remember water exposure differently.

• Design: Cohort.
• Reason: Starts with exposure (net use), follows forward to outcome (malaria).
• Limitation: Loss to follow-up; confounding (net users may be more health-conscious).

• Design: Cluster RCT.
• Reason: Tests an intervention; randomise by school to prevent contamination.
• Limitation: Expensive; Hawthorne effect (being observed changes behaviour).

• P: Mothers attending ANC.
• O: Knowledge of danger signs.
• Design: Cross-sectional survey.

• P: Mothers who delivered in the facility.
• E: Hand hygiene, instrument sterilisation, duration of labour.
• O: Wound infection.
• Design: Case-control (fast) or Cohort (measures incidence).

• P: Children in the affected community.
• E: Vaccination status.
• O: Measles infection.
• Design: Cohort (follows vax vs. unvax) or Case-control (compare vax status of cases vs. controls).

• P: Outpatients at the clinic.
• I: New triage system.
• C: Standard triage.
• O: Waiting time (minutes).
• Design: Quasi-experimental (before-and-after) or Cluster RCT if two similar clinics exist.

Target = broad group findings should apply to. Sample = actual participants studied. The sample is a subset of the study population, which is a subset of the source population, which is a subset of the target population.

Case-control starts with outcome and looks backward for exposure. Cohort starts with exposure and follows forward for outcome. Case-control is efficient for rare diseases; cohort measures incidence.

Cross-sectional. It takes a snapshot at one point in time.

The researcher assigns the exposure/intervention. This is the defining feature.

It distributes known and unknown confounders evenly between groups, reducing selection bias and strengthening causal inference.

Internal = Are findings correct for the people studied? External = Can findings apply to other settings?

Exposure and outcome are measured simultaneously — we cannot establish which came first.

Cases remember past exposures differently (often more accurately) than controls. Most affects case-control studies.

Participants drop out before outcome measurement. If dropouts differ systematically between groups, the comparison is biased. Most affects cohort studies and RCTs.

When randomisation is not feasible or ethical — e.g., ward-wide quality improvement, school-wide programs, or when you cannot deny an intervention to a control group.

A third factor associated with both exposure and outcome, creating a false association. Example: Coffee appears linked to lung cancer, but the confounder is smoking. Smokers drink more coffee AND have higher lung cancer risk.

• Gordis, L. (2013). Epidemiology (5th ed.). Saunders Elsevier.
• Webb, P., & Bain, C. (2010). Essential Epidemiology: An Introduction for Students and Health Professionals (2nd ed.). Cambridge University Press.
• Rothman, K. J., Greenland, S., & Lash, T. L. (2008). Modern Epidemiology (3rd ed.). Lippincott Williams & Wilkins.
• Hulley, S. B., Cummings, S. R., Browner, W. S., Grady, D. G., & Newman, T. B. (2013). Designing Clinical Research (4th ed.). Lippincott Williams & Wilkins.