How to Study Epidemiology: A Practical Guide

Start with the question, not the definition

Epidemiology studies the distribution and determinants of health events in populations, and uses that knowledge to prevent and control health problems. To study it well, practise moving between three things: the question being asked, the design used to answer it, and the measure that follows from that design. Memorising terms without making those connections makes it harder to interpret a paper or choose the right calculation.

Begin with the topics in your course outline or learning objectives. Sort them into a small number of connected areas: frequency of disease, study designs, measures of association, bias and confounding, screening and diagnostic tests, and causal reasoning. Keep a question list beside the outline. For every topic, ask: What does this measure? What data do I need? What can this design establish? What could distort the result?

Build your revision material from your own lecture slides, readings and notes. For each concept, capture a short definition, its formula or decision rule where relevant, and a worked example. Keep exceptions and assumptions close to the rule they qualify. A board on this topic ends up looking like this:

Learn the measures by deriving them

Do not treat epidemiological formulae as isolated items to memorise. First identify the population and time frame, then classify each quantity as a proportion, rate or odds. Write the denominator in words before substituting numbers. This catches a common error: using the entire population when the correct denominator is only people at risk, or confusing person-time with a fixed follow-up population.

For example, if 20 of 200 people at risk develop a condition during follow-up, the cumulative incidence is 20/200, or 10% over that stated period. If participants contribute different lengths of follow-up, calculate the incidence rate using total person-time instead. A rate has units, such as cases per 1,000 person-years; a risk is a proportion over a specified period. They are not interchangeable.

Practise reading a 2 × 2 table until you can reconstruct the measures from the labels. For a cohort table, calculate risk among exposed and unexposed, then divide to obtain the risk ratio. Calculate the risk difference by subtraction. In a case-control study, participants are sampled according to outcome status, so the usual association measure is the odds ratio; the sampled proportions do not directly provide population risks. When an outcome is rare, the odds ratio may approximate the risk ratio, but that approximation depends on rarity and should not be assumed automatically.

Use short retrieval practice after learning a formula: close your notes, state what the numerator and denominator represent, and solve a fresh example. The MySummaries core checklist can be used to check whether the essential rules are still retrievable:

Match the study design to the question

For each design, learn what is sampled, when exposure and outcome are measured, and what the design can estimate. Then practise choosing a design for a research question and naming its main limitation.

A cohort study starts with exposure status and follows people for outcomes. It can establish that measured exposure preceded the outcome and can estimate incidence when follow-up and ascertainment support it. A case-control study starts with outcome status and looks back for exposure; it is often useful for rare outcomes or outcomes with long latency, but selection of appropriate controls and accurate exposure measurement are important. A cross-sectional study measures exposure and outcome around the same time, making it useful for estimating prevalence but often unable to establish which came first.

A randomised trial assigns an intervention, usually to compare outcomes between groups. Randomisation can balance confounders on average, but does not guarantee that every trial is unbiased: consider allocation, adherence, loss to follow-up, outcome measurement and analysis. Ecological studies compare groups rather than individuals. Be alert to the ecological fallacy: a group-level association does not necessarily hold for individuals within those groups.

Make yourself explain the design choice aloud in one or two sentences. “This is a cohort because participants are classified by exposure before outcome follow-up; the measure of interest is the risk ratio” is more useful than merely naming “cohort”. In your notes, pair every design with a plausible question and a caveat.

Practise bias, confounding and causal reasoning

These concepts are easier to distinguish using a specific study. Bias is a systematic error in design, conduct or measurement. Confounding occurs when the exposure-outcome association is mixed with the effect of another factor. Random error, by contrast, is variation that affects precision. A larger sample can reduce random error, but does not automatically fix selection bias, measurement error or confounding.

For any reported association, ask how participants were selected, how exposure and outcome were measured, what happened to people lost to follow-up, and which potential confounders were considered. Confounding may be addressed through design, such as restriction or randomisation, and through analysis, such as stratification or regression adjustment. Adjustment is not a cure-all: the variables must be measured adequately and selected with a defensible causal rationale. Overadjusting for a mediator or collider can introduce its own problems.

A causal diagram can help you make assumptions visible. Draw arrows between the exposure, outcome and plausible third variables before deciding what to adjust for. Then explain what evidence would strengthen or weaken a causal interpretation: temporality, consistency, a plausible mechanism, and alternative explanations. No single criterion turns an observational association into proof of causation.

Interpret uncertainty and diagnostic results

When you read a result, state the estimate, the confidence interval and the practical meaning. A narrow interval usually indicates greater precision than a wide one, given the model and data. If a confidence interval for a ratio includes 1, the data are compatible with no relative association at the corresponding confidence level; the interval also shows a range of effect sizes compatible with the data under the analysis. Avoid reducing interpretation to whether a p-value crosses a threshold. Statistical significance does not tell you whether an effect is important, unbiased or causal.

For diagnostic tests, learn the definitions with the direction of the conditional probability clear. Sensitivity is the probability of a positive test among people who have the condition; specificity is the probability of a negative test among people who do not. Positive predictive value is the probability of having the condition given a positive result. Predictive values depend on prevalence in the tested population, so they can differ when the same test is used in populations with different pre-test probabilities.

Screening is not simply testing many people. Consider the condition’s natural history, test performance, the balance of benefits and harms, and what happens after a positive result. Check your course materials for the screening frameworks and local recommendations it expects; recommendations can vary by population and jurisdiction.

Turn revision into a repeatable session

Use a session sequence that makes you retrieve, apply and correct knowledge rather than reread notes from beginning to end:

1. Retrieve: spend five minutes writing definitions or drawing a design from memory.
2. Apply: solve a small set of calculations or interpret a study abstract. Label the denominator and the design before calculating.
3. Explain: give a short verbal account of what the result means and what it cannot establish.
4. Correct: compare with your source notes. Record the exact error, such as “used total population instead of population at risk”.
5. Review: turn that error into a concise question and revisit it later.

Keep flashcards to one point each. Ask for a definition, a denominator, a design feature or a distinction; avoid cards that demand an entire lecture. A short review deck from this board might look like this:

That is a MySummaries deck, filled with epidemiology material. Yours is written from your own notes. Start free

Do not use cards as a substitute for working through tables or interpreting results. After answering a card, rate whether you recalled it accurately and promptly. If you missed a denominator or qualification, make that distinction the focus of the next practice problem.

Use audio for consolidation, not first learning

An audio explanation can be useful when reviewing a connected process, such as choosing a measure from a study design. It is less suitable for learning a calculation that you have never attempted: pause and solve the calculation yourself, then compare your steps with the explanation. Listen for the reasoning between the definitions, not just the final formula.

The MySummaries lecture screen can read a focused section of your board back as a short explanation:

How MySummaries helps

MySummaries lets you build a revision board from your own epidemiology slides, PDFs and notes, then use that board to make retrieval cards and short audio lectures. That keeps definitions, examples and qualifications tied to the material you are studying. You can start at portal.mysummaries.app.