Start with the question, not the formula

Biostatistics becomes easier to study when you treat it as a sequence of decisions. First identify the research question and study design. Then identify the outcome and the data type. Next choose a suitable summary or analysis, calculate or interpret the result, and state what it means in context.

That sequence is more useful than memorising a catalogue of tests. A test name alone does not tell you whether it fits the data, whether its assumptions are met, or what conclusion its result supports.

The exact emphasis varies across biomedical science courses. Use your course outline and past or sample assessments to find the topics and question styles you need. Do not assume that every course tests the same calculations or software. A reliable baseline is to be able to explain common measures, select an appropriate method, interpret uncertainty and recognise limitations.

Build a map of the subject

Divide your notes into connected sections rather than keeping one long formula sheet:

  1. Study design and variables: trial, cohort, case-control and cross-sectional studies; exposure, outcome and confounding; categorical and numerical variables.
  2. Description and uncertainty: proportions, means, medians, standard deviation (SD), standard error (SE) and confidence intervals (CIs).
  3. Comparisons and association: common statistical tests, risk measures and regression.
  4. Diagnostic tests: sensitivity, specificity and predictive values.
  5. Interpretation: p-values, effect size, precision, bias and limits on causal claims.

For each section, write a short explanation in your own words, add a worked example, and list the mistake you are most likely to make. Put facts that need repeated recall—such as formula meanings or interpretation rules—into flashcards. Keep longer reasoning, assumptions and worked solutions on the board.

A MySummaries board on this topic might organise those notes like this:

Biostatistics Choosing and interpreting analysesStudy
Biomedical science — core methodsChoosing and interpreting analyses5 sections · 3 columns
Study design and variables3 due
  • Cohort study — participants are grouped by exposure and followed for outcomes
  • Case-control study — participants are selected by outcome; compare prior exposure
  • Outcome type guides the summary and analysis
Uncertainty and tests

A 95% CI describes uncertainty using a method with 95% long-run coverage under its assumptions. A p-value is the probability, assuming the null model, of results at least as incompatible with it as those observed.

Centre and spread
MeasureUse or meaning
Mean, SDNumerical data; SD describes spread of observations
Median, IQRUseful for skewed numerical data; IQR is Q3 − Q1
SEPrecision of an estimate; for a sample mean, SD/√n
Diagnostic accuracy
  • Sensitivity = TP/(TP + FN)
  • Specificity = TN/(TN + FP)
  • Predictive values depend on prevalence in the population tested
Effect measures
  • Risk = events ÷ people at risk
  • RR = risk in exposed ÷ risk in comparison group
  • OR = odds in one group ÷ odds in the other; common in case-control studies

Learn the core ideas before adding detail

Know what each number represents, not just how to produce it. SD describes variation among observations. SE describes how much an estimate would vary across repeated samples under the model; for a sample mean, SE = SD/√n. A larger sample generally reduces the SE, but does not necessarily reduce bias or make a poorly designed study valid.

For a CI, describe the estimate, its interval and the uncertainty it conveys. Avoid saying that there is a 95% probability the fixed true value lies in this particular calculated interval. In a frequentist interpretation, 95% refers to the long-run performance of the interval method under its assumptions. If a CI for a difference includes zero, the data are compatible with no difference at the corresponding two-sided level; that does not establish that the groups are equivalent.

A p-value is not the probability that the null hypothesis is true, nor a measure of effect size. Interpret it alongside the estimated effect, CI, study design and potential sources of bias. Statistical significance alone does not establish clinical importance or causation.

For risk measures, keep the denominator in view. If 12 of 200 people in one group and 24 of 200 in another experience an event, the risks are 0.06 and 0.12. The risk ratio is 0.06/0.12 = 0.5; the absolute risk reduction is 0.12 − 0.06 = 0.06, or 6 percentage points. The number needed to treat is 1/0.06 = 16.7, conventionally rounded up to 17, over the stated follow-up period. This interpretation assumes the comparison and outcome are suitable for that calculation; do not omit the follow-up period.

A compact checklist helps you retrieve these distinctions accurately:

Key pointsChoosing and interpreting analyses
Mean describes the arithmetic centre; SD describes spread of observations; for a sample mean, SE = SD/√n.
Risk = events ÷ people at risk; RR = risk in group 1 ÷ risk in group 2; ARR = comparison risk − intervention risk.
Sensitivity = TP/(TP + FN); specificity = TN/(TN + FP). Predictive values change with prevalence.
A p-value is calculated assuming the null model; it is not the probability that the null hypothesis is true.
Report an effect estimate with its CI and context; statistical significance alone does not show clinical importance or causation.

Choose a method from the design and outcome

Practise the choice as a short explanation. Name the outcome type, state whether you are describing groups or estimating an association, then give the method and any key assumption you would check. For example, a comparison of a numerical outcome between two independent groups may call for an independent-samples t-test if its assumptions are reasonable; a paired before-and-after measurement calls for an approach that accounts for pairing. If the data are skewed or assumptions are not reasonable, consider an appropriate alternative rather than applying the test by habit.

For categorical outcomes, a chi-square test is commonly used to assess association between categorical variables when its expected-count assumptions are suitable. With sparse tables, another method may be needed. For time-to-event outcomes, methods such as Kaplan–Meier estimation and the log-rank test account for event timing and censoring in ways that a simple comparison of proportions does not. Your course may teach additional methods; use its specified assumptions and terminology.

Regression extends these comparisons by estimating associations while accounting for other included variables. Linear regression is used for a numerical outcome under its model assumptions; logistic regression models the log odds of a binary outcome. Adjustment does not automatically remove confounding: the variables measured, the model and the study design all matter. In a case-control study, an odds ratio is generally available, while risks and risk ratios cannot usually be calculated directly from the sampled case and control counts.

Turn each mistake into a recall prompt

After solving a problem, record the exact step that caused trouble. Was the denominator wrong? Did you confuse SD with SE? Did you call an association causal? Make the prompt narrow enough to answer in one or two sentences. A small set of precise cards is more useful than a card that asks for a whole chapter.

A short MySummaries card drill could test facts from this board. Answer each prompt, turn the card, then grade your recall: choose Again if you could not retrieve it, Hard if the answer took effort, Good if it was correct, or Easy if it was immediate and secure.

Cards — Choosing and interpreting analyses6 due

What does SD describe?

The spread of individual observations around their mean.

All 6 cards
What does SD describe?The spread of individual observations around their mean.
For a sample mean, how is the standard error calculated?SE = SD/√n.
Write the formula for a risk ratio.Risk in group 1 divided by risk in group 2.
What is sensitivity?TP/(TP + FN): the proportion of people with the condition who test positive.
What does a p-value describe under the null model?The probability of results at least as incompatible with the null model as those observed.
Why can predictive values change between populations?They depend on prevalence as well as the test's sensitivity and specificity.

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

Practise interpretation as well as calculation

For every numerical exercise, write a conclusion in plain language after the arithmetic. Include the groups or population, the outcome, the effect estimate and its uncertainty where available. Check that your denominator matches the question and that percentages and percentage-point differences are not confused.

For a diagnostic test, keep the conditional direction straight: sensitivity starts with people who have the condition; specificity starts with people who do not. Positive and negative predictive values start with test results and ask about condition status. A highly sensitive test may be useful for ruling out a condition when negative in an appropriate setting, but the interpretation still depends on the test and clinical context. Do not infer predictive value from sensitivity and specificity alone without considering prevalence.

Use a consistent practice cycle. First attempt a question without notes. Mark the first incorrect step, not only the final answer. Rework it from the beginning, then explain why the method fits. A week later, answer a similar question with changed numbers. This distinguishes understanding from memorising one worked example.

Listen to explanations while revisiting a weak section

Audio is useful after you have tried the material yourself: listen for the reasoning that links the design to the measure, rather than treating the script as background noise. Pause before a conclusion and say what the result supports, what it does not support and which assumptions matter. Then return to the board and correct any imprecise wording in your notes.

A MySummaries lecture from the same board might revisit the interpretation of uncertainty and effect measures:

Lecture — Choosing and interpreting analyses8 min
From estimate to conclusionConnects effect size, uncertainty and cautious interpretation in a biomedical study.
03:1608:02
Speed1×1.25×1.5×2×

Transcript · tap any word to jump there

Start with the estimate, not the p-value. In a comparison of two groups, the estimate tells you the direction and size of the observed difference. Ask whether the measure is a difference in means, a difference in risks, a ratio, or something else; the label determines how you explain its value.

Next, read the confidence interval. A wide interval signals substantial uncertainty about the size of the effect. If an interval for a difference includes zero, the data are compatible with no difference under the method used, but that is not proof that the groups are equivalent. A narrow interval can still describe a biased estimate if the study design or measurement is flawed.

Finally, state a conclusion that matches the evidence. A small p-value does not tell you the probability that the null hypothesis is true, and it does not establish clinical importance. Describe the observed association, its uncertainty and the limits of the design. If the study is observational, be cautious about claiming that an exposure caused the outcome.

A repeatable study session

Use a 40–60 minute session to keep practice active: spend about 10 minutes retrieving definitions and formulas without notes, 20–30 minutes solving or interpreting a few problems, and the remaining time correcting errors and making one or two focused recall prompts. Adjust the time to your workload. If you have access to course questions, use them to match the language and depth expected in your assessments; do not infer an assessment format from a general guide.

At the end of each session, check whether you can explain the method choice and the result without relying on a memorised phrase. Revisit topics where you made repeated errors before adding more formulas. This method builds calculation skill and statistical judgement together, which is what makes biostatistics usable in biomedical research.

How MySummaries helps

Build a revision board from your own biostatistics notes, slides, PDFs or photographed handwritten pages. From that board, you can practise spaced-repetition cards, generate an audio lecture from a difficult section and work through written mock questions marked against your material. Start at portal.mysummaries.app.