Chapter 5. Study design — deciding what to look at through the lens of epidemiology

Key points

• Study designs divide broadly into descriptive and analytic
• Knowing cross-sectional from longitudinal (cohort, case-control) shows you how to collect
• Structuring the question as PICO/PECO makes every conversation with AI sharper

What you look at decides how you collect

Once a hypothesis exists, the next question is how to confirm it — the study design. Start collecting before choosing a design and you meet the tragedy of realising, later, that this data cannot answer this question. Learn the shapes first.

In epidemiology and social research, studies divide broadly in two by purpose:

• Descriptive — grasping a situation, observing cases (describing what is happening, and how much)
• Analytic — analysing the relationship between cause and effect (exploring why it happens)

The same perspective applies directly to market research and user analysis in business.

The divide is in how time is handled

Analytic studies change character according to how they handle time.

Cross-sectional study. Records the state of a population at a single point in time and looks for associations. Example: the relationship between memory scores and exercise habits, right now. Straightforward, but weak on which came first.

Longitudinal study. Follows change and influence over time. It divides again:

• Cohort study — follows a population forward (e.g. compare cognitive function after three years between groups with and without an exercise habit)
• Case-control study — works backward from the outcome to look for factors (e.g. compare past lifestyle between people whose cognitive function declined and people whose did not)

Case reports and case studies. Observe and describe an individual or a small number of cases in detail. Useful for generating hypotheses. A therapist's case report is exactly this. You already carry out perfectly respectable descriptive research.

The magic shape for structuring a question — PICO/PECO

Researchers worldwide use a shape for questions: PICO/PECO. It breaks a question about the effect of an intervention or a factor into four elements.

• P (Population) — the people or group in question (e.g. community-dwelling older adults)
• I (Intervention) / E (Exposure) — the intervention, or the factor (e.g. a three-times-weekly exercise programme / the presence of an exercise habit)
• C (Comparison) — what it is compared against (e.g. a non-exercising group, the existing programme)
• O (Outcome) — what is measured as the result (e.g. memory test scores)

"I" is used where you intervene yourself; "E" where you observe a factor that exists naturally.

So the vague "is exercise good for the brain?" becomes:

P: community-dwelling people in their seventies E: exercising three or more times a week, or not C: the non-exercising group O: memory scores after one year

That alone makes it suddenly clear what to measure, from whom, and how.

AI as a design consultant

Designing the study is exactly where AI helps. Tell it your vague interest, have it organised into PICO/PECO, and ask for suitable designs. Once a draft exists, you trim it to fit the constraints of your setting — time, numbers, budget.

If you keep the various design templates as notes in Obsidian, combining them with a hypothesis you have just had becomes very smooth. Feed those notes to the AI and prompts like "using this cross-sectional design template, write a study protocol to test the hypothesis that ◯◯" become easy to write.

One caution. AI proposes plausible designs with complete confidence. Whether they are valid, feasible and ethically sound must be confirmed by you, and preferably by a supervisor or co-investigator.

A prompt to try

You are a researcher who knows epidemiology and study design well. I want to investigate the following.

[What I want to investigate] (e.g. whether our facility's exercise programme raises users' life satisfaction)

1. Structure this question as PICO (or PECO).
2. Propose possible study designs (descriptive, cross-sectional, cohort, case-control), with the advantages and disadvantages of each.
3. Narrow to a single design that is realistically feasible in a setting with limited numbers and budget, and recommend it.
4. List the ethical and practical points that need particular care in that design.

Chapter 5 summary

• Studies divide into descriptive and analytic, and by time into cross-sectional and longitudinal
• A therapist's case report is respectable descriptive research
• PICO/PECO makes both the design and the conversation with AI sharper