Why Biostatistics Matters in Medicine

Overview

Biostatistics is the application of statistical reasoning to questions about health, disease and healthcare. It helps clinicians and researchers move from individual observations to conclusions that can be examined, challenged and used responsibly.

The subject is not merely a collection of tests or formulas. It is a framework for asking precise questions, collecting suitable data, describing what was observed, estimating effects, quantifying uncertainty and deciding how much confidence to place in a conclusion. These tasks are central to clinical practice, medical research, public health and evidence-based medicine.

This lecture introduces the role of biostatistics through one simulated clinical scenario. The aim is not to teach a particular statistical test. Instead, it is to show the reasoning that should occur before, during and after any analysis.

Why this matters

Every medical decision is made under uncertainty. A clinician may be uncertain whether a treatment will benefit a particular patient. A researcher may be uncertain whether an apparent difference between groups reflects a real effect or random variation. A hospital may be uncertain whether a new care pathway is worth implementing. A public-health team may be uncertain whether a programme has improved outcomes across a population.

Evidence-based medicine combines the best available research evidence with clinical expertise and the values and circumstances of patients [1]. Biostatistics contributes to the research-evidence component. It does not replace clinical judgement, biological understanding, ethical reasoning or patient preference. It helps make the evidence more transparent and interpretable.

Poor statistical reasoning can produce harmful consequences. A treatment may be adopted because a relative effect sounds impressive even though the absolute benefit is small. A useful intervention may be dismissed because a study was too small to estimate its effect precisely. A difference caused by bias or confounding may be mistaken for a treatment effect. A statistically detectable result may be treated as clinically important when the actual difference is trivial.

Key concepts

Biostatistics connects a question to a decision

The clinical problem comes first. Statistical methods are chosen only after the question, population, outcome and intended comparison have been defined. A technically correct analysis cannot compensate for a vague question or an unsuitable study design. This principle has long been central to medical statistics: design, analysis and interpretation are connected rather than separate activities [2].

A useful sequence is:

1. Define the question. What decision needs to be informed?
2. Define the target population. To whom should the conclusion apply?
3. Define the outcome and comparison. What will be measured, and between which groups or conditions?
4. Collect suitable data. Are the measurements valid, complete and obtained without avoidable bias?
5. Summarise and estimate. What happened, and how large was the observed difference?
6. Quantify uncertainty. How precise is the estimate?
7. Assess credibility. Could bias, confounding, missing data or analytical choices explain the result?
8. Integrate the evidence. How does the result fit with other evidence, clinical expertise, patient values, feasibility, harms and costs?

Data are not automatically evidence

Data are recorded observations, such as age, blood pressure, laboratory results, treatment received and survival status. A dataset may contain thousands of observations but still provide weak evidence if the measurements are inaccurate, the study population is unrepresentative or the comparison is biased.

Statistical information is produced when data are organised and summarised. Examples include a mortality risk of 14%, a mean systolic blood pressure of 132 mmHg or a median hospital stay of 5 days.

Research evidence requires more than a summary. The study design, data quality, effect estimate, uncertainty, assumptions and possible sources of bias must all be considered.

A decision goes further still. Decisions may depend on the balance of benefits and harms, certainty of evidence, resources, equity, feasibility and patient preferences. Structured evidence-grading approaches similarly separate the estimated effect from the certainty of evidence and the strength of a recommendation [3].

Biostatistics has four broad roles

1. Describing health and disease

Descriptive statistics organise complex observations into understandable summaries. In clinical practice, they describe a patient’s measurements over time. In research, they describe study participants and outcomes. In public health, they quantify disease frequency, mortality, service use and inequalities.

Good description requires the correct denominator. Reporting 140 deaths is incomplete without knowing how many people were followed. In the scenario, 140 deaths among 1,000 participants represents a 30-day mortality risk of 14.0%.

2. Estimating effects and associations

Biostatistics helps quantify differences, relationships and changes. The result should be expressed in a measure that answers the question, such as a risk difference, risk ratio, mean difference, prevalence ratio or hazard ratio.

Different measures can describe the same data from different perspectives. The absolute risk difference shows the change in percentage points and is often useful for planning services. The risk ratio expresses the outcome in one group relative to another and is often useful for comparing proportional effects.

3. Quantifying uncertainty

A study observes a sample rather than every possible patient who could receive the intervention. Another sample would not produce exactly the same result. This natural sample-to-sample variation is called sampling variation.

A confidence interval describes the precision of an estimate under the statistical model and assumptions used. A wide interval indicates greater uncertainty; a narrow interval indicates greater precision. A 95% confidence interval should not be interpreted as a 95% probability that the true value lies inside this particular calculated interval. It is better understood as a range of parameter values reasonably compatible with the data and analysis assumptions [4,5].

4. Assessing the credibility of a conclusion

Statistical analysis must be interpreted in the context of how the data were generated. Important threats include:

• selection bias, when the included participants differ systematically from the intended population;
• measurement bias, when exposure, intervention or outcome measurements are inaccurate or differ between groups;
• confounding, when another factor is related to both the exposure and outcome and distorts an observed association;
• missing data, when unavailable observations are related to prognosis, treatment or outcome;
• lack of independence, when repeated or clustered observations are analysed as though they came from unrelated individuals; and
• selective analysis or reporting, when only favourable outcomes, subgroups or models are presented.

Randomisation can reduce confounding at baseline in a well-conducted trial, but it does not prevent every problem. Loss to follow-up, poor adherence, outcome-measurement differences and selective reporting can still reduce credibility.

Biostatistics across medicine and health

Clinical practice

Clinicians use statistical evidence when interpreting diagnostic tests, estimating prognosis, discussing treatment benefits and harms, and applying clinical prediction tools. Statistical literacy helps a clinician ask whether the evidence is relevant to the patient in front of them, whether the estimated effect is large enough to matter and how uncertain that estimate is.

Medical research

In research, biostatistics contributes from protocol development to publication. It helps define outcomes, select a design, calculate a defensible sample size, plan data collection, choose analyses, handle missing data, present results and avoid interpretations that exceed the study design.

Current reporting guidance for randomised trials and observational studies requires transparent reporting of study methods and statistical analyses so that readers can evaluate what was done [6,7]. Reporting guidance does not guarantee a good study, but incomplete reporting makes critical appraisal difficult.

Public health

Public-health decisions often concern populations rather than individual patients. Biostatistics is used to estimate disease burden, compare regions or time periods, evaluate screening and vaccination programmes, detect outbreaks, monitor inequalities and allocate resources.

Absolute effects are particularly important in public health. A modest reduction in individual risk may prevent many events when applied to a large population. Conversely, a large relative effect may have little population impact when the event is rare.

Evidence-based decision-making

Evidence-based decisions require more than asking whether a result is statistically significant. Decision-makers should consider:

• the size and direction of the effect;
• the precision of the estimate;
• the validity and applicability of the study;
• consistency with other evidence;
• benefits, harms and opportunity costs;
• feasibility and resource requirements; and
• patient and community values.

Worked example

Simulated 30-day mortality data

simulated data

Table 1. Thirty-day mortality by allocated care group

Allocated group	Deaths within 30 days	Total participants	Mortality risk
New sepsis pathway	140	1,000	14.0%
Usual care	180	1,000	18.0%

Note: All values are simulated for teaching. Follow-up is complete for all 2,000 participants.

Why absolute and relative effects should be reported together

The relative reduction of 22% may sound more impressive than the absolute reduction of 4 percentage points, although both describe the same comparison. Reporting both prevents selective framing and helps different readers answer different questions.

For a clinician, the absolute effect helps communicate expected benefit to patients. For a hospital, it helps estimate the number of events potentially prevented. For researchers, the relative effect may help compare results across populations with different baseline risks. Neither measure should be interpreted without its confidence interval.

Why the P-value is not the decision

The P-value of 0.015 does not show that there is a 98.5% probability that the pathway is effective. It also does not show that the pathway is clinically worthwhile. The P-value is influenced by both the observed effect and the sample size. With a very large sample, a clinically trivial difference can produce a small P-value. With a small sample, a clinically important effect can remain imprecisely estimated and produce a larger P-value.

The American Statistical Association advises that scientific conclusions should not be based only on whether a P-value crosses a particular threshold [4]. The effect estimate, uncertainty, design, data quality and clinical context should remain central.

From study result to target estimand MMed extension

A deeper analysis asks exactly what treatment effect the study intends to estimate. This target quantity is called the estimand. It includes the population, treatment conditions, outcome, time point and summary measure of interest.

In the scenario, one possible estimand is the difference in 30-day all-cause mortality risk between allocation to the new pathway and allocation to usual care among eligible adults with suspected sepsis. The interpretation may change if the question instead concerns treatment actually received, centres with full implementation, patients who completed every pathway step or mortality at 90 days.

Defining the estimand before analysis prevents a statistical result from drifting away from the clinical question.

What biostatistics can and cannot do

Biostatistics can make uncertainty visible, compare outcomes on meaningful scales and reveal whether a conclusion depends on analytical assumptions. It can also identify patterns that deserve further investigation.

Biostatistics cannot rescue poor measurement, remove all bias from an unsuitable design or prove that an association is causal. It cannot determine whether an effect is ethically acceptable, affordable or aligned with patient preferences. These require broader scientific and clinical judgement.

A sophisticated model applied to biased data may produce a precise but misleading estimate. Conversely, a simple, well-planned analysis of high-quality data may provide more trustworthy evidence. The aim is not maximum mathematical complexity. The aim is an analysis that answers the question with appropriate transparency.

Common mistakes

Knowledge check

In the simulated study, why is '140 deaths' insufficient as a result?

Answer and reasoning

A count needs a denominator and a defined follow-up period. Here, 140 deaths occurred among 1,000 participants within 30 days, giving a mortality risk of 14.0%. Without the denominator and time frame, the frequency cannot be compared or interpreted properly.

What is the difference between the absolute risk difference of −4.0 percentage points and the risk ratio of 0.78?

Answer and reasoning

The risk difference describes the absolute change: the new sepsis pathway group had 4 fewer deaths per 100 patients, or 40 fewer per 1,000, than usual care. The risk ratio describes the relative comparison: mortality in the new sepsis pathway group was 0.78 times that in the usual-care group. Both are valid, but they answer different practical questions.

Does p = 0.015 mean that there is a 1.5% probability that the pathway has no effect?

Answer and reasoning

No. A P-value is calculated under a specified statistical model, including a null hypothesis. It does not provide the probability that the null hypothesis is true. The effect estimate, confidence interval, study design and possible biases must also be considered.

Why can a statistically detectable difference still be unimportant clinically?

Answer and reasoning

Statistical detectability depends partly on sample size. A very large study can estimate a very small difference precisely and produce a small P-value. Clinical importance depends on the magnitude of benefit or harm, the outcome, patient priorities, feasibility and costs, not only on statistical significance.

What should the hospital committee consider besides the estimated mortality reduction?

Answer and reasoning

The committee should consider the precision of the estimate, study quality, possible harms, implementation fidelity, staffing and resource requirements, applicability to its patients, consistency with other studies, equity and patient or community values. Biostatistics informs the decision but does not make the decision by itself.

Key take-home messages

References

1. Sackett DL, Rosenberg WMC, Gray JAM, Haynes RB, Richardson WS. Evidence based medicine: what it is and what it isn’t. BMJ. 1996;312(7023):71–72. https://doi.org/10.1136/bmj.312.7023.71
2. Altman DG. Practical statistics for medical research. London: Chapman & Hall; 1991.
3. Guyatt GH, Oxman AD, Vist GE, Kunz R, Falck-Ytter Y, Alonso-Coello P, et al. GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ. 2008;336(7650):924–926. https://doi.org/10.1136/bmj.39489.470347.AD
4. Wasserstein RL, Lazar NA. The ASA’s statement on p-values: context, process, and purpose. Am Stat. 2016;70(2):129–133. https://doi.org/10.1080/00031305.2016.1154108
5. Greenland S, Senn SJ, Rothman KJ, Carlin JB, Poole C, Goodman SN, et al. Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. Eur J Epidemiol. 2016;31(4):337–350. https://doi.org/10.1007/s10654-016-0149-3
6. Hopewell S, Chan AW, Collins GS, Hróbjartsson A, Moher D, Schulz KF, et al. CONSORT 2025 statement: updated guideline for reporting randomised trials. BMJ. 2025;389:e081123. https://doi.org/10.1136/bmj-2024-081123
7. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Ann Intern Med. 2007;147(8):573–577. https://doi.org/10.7326/0003-4819-147-8-200710160-00010