An epidemiologic study design is the plan for collecting and comparing data to answer a health question. The key questions for classifying any study are:

1. Did the investigator assign the exposure? Yes → experimental. No → observational.
2. Is there a comparison group? No → descriptive. Yes → analytic.
3. What is the starting point? Exposure (cohort), disease (case-control), or both at one time (cross-sectional).

Overview

Descriptive epidemiology answers who, where, and when (person, place, time). A nurse who compiles official data on cases by age, sex, barangay, and month is doing descriptive epidemiology.

Cross-sectional study. Exposure and disease are measured at the same time in a defined population (a "snapshot"). It yields prevalence, is quick and inexpensive, and suits community health needs assessment. Limitation: it usually cannot show which came first (temporality).

Ecological study. Uses group data (e.g., per-capita cigarette sales and lung cancer mortality by country). Risk: the ecological fallacy — assuming a group-level association holds for individuals.

Case-control study. Start with people who have the disease (cases) and people who do not (controls), then look back at past exposures. Efficient for rare diseases, diseases with long latency, and outbreak investigations in which the exposed population cannot be fully listed (e.g., a community-wide outbreak). For an outbreak in a small group with a complete list, such as a wedding or fiesta, a retrospective cohort of everyone who attended is preferred. The measure of association is the odds ratio (OR); incidence and relative risk cannot be calculated directly because the investigator chose how many cases and controls to include. Weakness: recall bias and difficulty choosing comparable controls.

Cohort study. Start with people free of the disease, classify them by exposure, and follow them to see who develops the disease. It gives incidence in each group and the relative risk (RR), and shows temporality most clearly among observational designs. Prospective cohorts follow people forward from now; retrospective (historical) cohorts use existing records to reconstruct exposure and follow-up. Weaknesses: costly, long, and loss to follow-up; inefficient for rare diseases.

Randomized controlled trial (RCT). The investigator randomly assigns participants to intervention or control (placebo or usual care), often with blinding. Randomization balances known and unknown confounders, so the RCT gives the strongest evidence for cause and effect of an intervention.

Measures of association

Error and bias

• Selection bias — groups differ in how they were chosen (e.g., hospital controls)
• Information bias — measurement error, including recall bias (cases remember exposures more than controls) and interviewer bias
• Confounding — a third factor linked to both exposure and disease distorts the association (e.g., age confounding the link between grey hair and heart disease)
• Controlling confounding: randomization, restriction, matching (design stage); stratification and multivariable analysis (analysis stage). Increasing sample size improves precision (reduces random error) but does not remove confounding.

Hierarchy of evidence (for interventions, strongest first): systematic reviews/meta-analyses of RCTs → RCTs → cohort → case-control → cross-sectional and ecological → case series/reports → expert opinion.

Worked example 1 — Cohort study (RR). 1,000 smokers and 2,000 non-smokers, all free of lung disease, are followed for 10 years. 30 smokers and 10 non-smokers develop the disease.

• Incidence in exposed (Ie) = 30 ÷ 1,000 = 0.030 = 30 per 1,000
• Incidence in unexposed (Iu) = 10 ÷ 2,000 = 0.005 = 5 per 1,000
• RR = 0.030 ÷ 0.005 = 6 — smokers had 6 times the risk.

Check: 30 ÷ 5 = 6 ✓

• Attributable risk = 30 − 5 = 25 per 1,000 smokers over 10 years
• Attributable risk percent = 25 ÷ 30 × 100 = 83.3%

Check: 25 ÷ 30 = 0.8333 ✓

Worked example 2 — Case-control study (OR). 100 cases of a disease and 200 controls are asked about a past exposure.

• OR = (a × d) ÷ (b × c) = (60 × 150) ÷ (50 × 40) = 9,000 ÷ 2,000 = 4.5

Check: odds of exposure in cases = 60 ÷ 40 = 1.5; in controls = 50 ÷ 150 = 0.333; 1.5 ÷ 0.333 = 4.5 ✓

• Interpretation: the odds of exposure among cases were 4.5 times those among controls, suggesting the exposure is a risk factor. OR > 1 = positive association; OR = 1 = none; OR < 1 = protective.

Worked example 3 — Protective effect. In a trial, the incidence of the disease was 2 per 1,000 among vaccinated and 10 per 1,000 among unvaccinated children. RR = 2 ÷ 10 = 0.2. Vaccine efficacy = (10 − 2) ÷ 10 × 100 = 80%.

Check: 8 ÷ 10 = 0.8 ✓

Matching designs to nursing questions

• Recognizes the design of a published study and judges its strength before applying it to practice
• Collects data accurately (interviews, questionnaires, records) using standardized tools to limit information bias
• Participates in outbreak case-control studies by interviewing cases and controls with the same questionnaire
• Protects participants' rights and welfare as a data collector or researcher
• Translates findings into programs (evidence-based community health nursing)

• Informed consent: participants must understand purpose, risks, benefits, and the right to withdraw without losing services.
• Ethics review: research involving human participants should be reviewed by a research ethics committee before data collection. Routine outbreak investigation under public health authority is part of surveillance and response, not optional research.
• Privacy: research and surveillance data are sensitive personal information under the Data Privacy Act of 2012 (RA 10173); RA 11332 limits surveillance data to public health purposes.
• Equipoise and fairness in trials: a control group should not be denied an intervention already known to be effective.
• Honest reporting: fabricating or selectively reporting data is scientific misconduct.

Case 1. After a barangay wedding with a complete guest list, 40 guests become ill. The nurse helps find the food responsible. Which design fits, and which measure is used?

• Answer: retrospective cohort study of all guests; food-specific attack rates and RR (as in the fiesta example in Topic 9).
• Why: the whole exposed group can be listed, so everyone is classified by what they ate and whether they became ill.
• Use a case-control study (OR) instead when the exposed population cannot be fully listed, for example a community-wide outbreak linked to a store-bought product; then ill persons are compared with well controls about past exposures.

Case 2. A school nurse records the BMI and daily soft-drink intake of all Grade 6 pupils in one week and finds an association.

• Answer: cross-sectional study.
• Caution: it cannot show whether soft drinks came before overweight.

Case 3. Researchers randomly assign 20 barangays to receive a new household water-treatment program and 20 to continue usual care, then compare diarrhea incidence.

• Answer: community (cluster) trial — experimental.
• Why: the investigator assigned the intervention by random allocation of whole communities.

• Calling a study "prospective" just because data are collected now; the defining feature of a cohort is starting from exposure.
• Computing RR from a case-control study — use OR.
• Saying case-control studies are good for rare exposures — they suit rare diseases; cohorts suit rare exposures.
• Believing a larger sample eliminates confounding.
• Treating a cross-sectional association as proof of cause.
• Applying group-level (ecological) findings to individuals.
• Reading RR < 1 as "no effect" — it means a protective association.

• Experimental = investigator assigns exposure; observational = does not.
• Cross-sectional = snapshot → prevalence; weak on temporality.
• Case-control = start with disease, look back → OR; best for rare diseases and outbreaks without a complete list of the exposed; recall bias. Complete guest list → retrospective cohort (attack rates, RR).
• Cohort = start with exposure, follow forward → incidence and RR; best observational evidence of temporality.
• RCT = randomization controls confounding → strongest single-study evidence.
• RR = Ie ÷ Iu; OR = ad ÷ bc; AR = Ie − Iu; AR% = (Ie − Iu) ÷ Ie × 100.
• RR or OR of 1 = no association; > 1 = risk factor; < 1 = protective.
• Confounding control: randomization, restriction, matching, stratification, multivariable analysis.
• Ecological studies use group data; beware the ecological fallacy.