- Atlas
- Medicine & Health
- Biostatistics
Biostatistics
Biostatistics is the branch of statistics that deals with health: measuring how much disease a population carries, deciding whether a treatment really works, and separating a real effect from the play of chance.
11 frameworks, from 1662 to today · Public Health · Medicine & Health
- Political Arithmetic and Vital Statistics 1662-1900
- Bayesian Inference 1763-Present
- Numerical Method in Clinical Medicine 1820-1880
- Sanitary Statistics 1839-1900
- Frequentist Inference 1900-Present
- Randomized Controlled Trial 1948-Present
- Survival Analysis 1950-Present
- Observational Study Design and Causal Criteria 1951-Present
- Regularization and Statistical Learning 1970-Present
- Rubin causal model 1974-Present
- Meta-analysis 1976-Present
Overview
Biostatistics is the branch of statistics that deals with health: measuring how much disease a population carries, deciding whether a treatment really works, and separating a real effect from the play of chance1. Its practitioners design studies, count and model data, and draw conclusions that guide medicine and public health1. The stakes are concrete. A vaccine judged effective on faulty arithmetic can save lives or waste them; a death rate computed the wrong way can hide an epidemic or invent one. The history of biostatistics is the story of how people learned to count the sick and the dead honestly, and then how they learned to reason about uncertainty when the counts were in.
Counting the people: political arithmetic and vital statistics
The earliest strand is the systematic tallying of births and deaths. In 1662 the English merchant John Graunt published Natural and Political Observations upon the Bills of Mortality, analyzing London's weekly death records. Graunt estimated the city's population from the number of christenings, noted that more boys than girls are born, and constructed one of the first life tables—a table showing, for each age, how many people survive to it. His numbers were rough and his methods ad hoc, but he showed that routine administrative records could yield regularities worth studying.
A generation later, the astronomer Edmond Halley—better known for the comet—published in 1693 a life table built on data from Breslau, using it to price annuities. The problem he solved was practical: insurers and governments needed to know how long people lived on average to set fair prices. The life table became the foundational tool of demography and actuarial science, and it remains the backbone of survival analysis.
The eighteenth century added the idea of quantifying risk itself. The mathematician Abraham de Moivre worked on annuities and the mathematics of chance. In medicine, James Lind's 1747 trial of scurvy remedies aboard a naval ship compared several treatments in small groups of sailors—an early controlled comparison, though Lind did not analyze it statistically. The physician Pierre-Charles-Alexandre Louis in Paris in the 1820s and 1830s went further, tabulating outcomes of bloodletting for pneumonia and concluding, controversially, that the treatment did not help. Louis's "numerical method" was attacked by clinicians who trusted individual experience over aggregated counts; the dispute established the basic argument that group comparison can correct for the misleading impressions of single cases.
Sanitary arithmetic and the public health movement
The nineteenth century made vital statistics a tool of reform. In Britain, William Farr, appointed to the General Register Office in 1839, built a national system for classifying causes of death and computing mortality rates, including standardized comparisons between districts. Farr's work let reformers show that death rates varied enormously with living conditions. Florence Nightingale, returning from the Crimean War, used statistical tables and her famous polar-area diagrams to argue that most soldiers died of preventable disease rather than battle wounds, and she campaigned for sanitary reform with numbers as her weapon.
The most celebrated episode is John Snow's investigation of the 1854 cholera outbreak in London's Soho. Snow mapped deaths and traced them to a single water pump on Broad Street, arguing that cholera spread through contaminated water rather than "miasma," foul air. Snow's reasoning was epidemiological, but his method—comparing death rates among customers of different water companies—was statistical, and it is often cited as an early natural experiment. The sanitary movement that followed, with its sewers and clean water, was driven by exactly this kind of counting: rates of death by district, compared over time.
Inference: from Bayes to the significance test
The mathematics of drawing conclusions from data developed alongside. Thomas Bayes's essay, published posthumously in 1763, set out what became Bayes' theorem: a rule for updating a probability in light of new evidence23. In modern form, Bayesian inference treats an unknown quantity—a treatment effect, a disease rate—as having a probability distribution, which is revised as data arrive4. The approach is natural for medical judgment, since it lets prior knowledge be combined with new observations, but for two centuries it was computationally and philosophically awkward, and it receded as the dominant method.
The alternative was frequentist inference, built on the idea that probability means the long-run frequency of an event in repeated trials5. Karl Pearson, working at University College London from the 1890s, developed the chi-squared test for comparing observed counts with expected ones, and founded the journal Biometrika in 1901. Ronald Fisher, at Rothamsted Experimental Station, transformed the field in the 1920s6. He introduced the analysis of variance, the design of experiments, and the concept of the null hypothesis tested against data. Fisher's significance test asked: if there were no real effect, how often would data as extreme as these arise? A small probability—the p-value—was grounds for doubting the null hypothesis. Fisher also insisted on randomization, assigning treatments by chance so that no systematic bias could creep in.
Fisher's approach competed with that of Jerzy Neyman and Egon Pearson (Karl Pearson's son), who in the 1930s formalized hypothesis testing with explicit alternative hypotheses, and distinguished Type I errors (false positives) from Type II errors (false negatives). The hybrid that most textbooks teach today—a null hypothesis, a p-value, a threshold such as 0.05—is a blend of Fisher's and Neyman–Pearson's ideas, and its logic has been debated ever since. The p-value's meaning is routinely misread, and the "reproducibility crisis" of recent decades has renewed criticism of significance testing as a basis for scientific claims.
The randomized controlled trial
The idea of comparing treatments by chance assignment matured in the twentieth century. Austin Bradford Hill, working in Britain, designed the 1948 randomized trial of streptomycin for pulmonary tuberculosis, run by the Medical Research Council—widely regarded as the first properly randomized clinical trial7. Patients were allocated at random to streptomycin or bed rest, and outcomes were assessed without knowledge of assignment. The trial demonstrated a clear benefit and established the template: randomization to remove selection bias, a control group, and blinded assessment.
The randomized controlled trial (RCT) became the gold standard for causal questions about treatment. Its logic is simple: if assignment is random, the groups differ only by chance, so any difference in outcome can be attributed to the treatment. The method spread through medicine in the following decades, and its statistical machinery—sample size calculation, interim analysis, intention-to-treat analysis—grew with it. The 1962 amendments to the U.S. Food, Drug, and Cosmetic Act, requiring evidence of efficacy, made the controlled trial a regulatory necessity.
Observational studies and causal criteria
Not everything can be randomized. Smoking cannot be assigned, nor can diet, occupation, or pollution. For these questions, biostatisticians developed observational study design: cohort studies following exposed and unexposed groups forward in time, and case-control studies comparing the past exposures of the sick and the well89. The British doctors study, begun by Richard Doll and Austin Bradford Hill in 1951 and following tens of thousands of physicians, provided the decisive evidence linking smoking to lung cancer. Its design—a prospective cohort with careful follow-up—showed how observational data could establish causation when experiment was impossible.
Bradford Hill's 1965 paper set out "viewpoints" for judging whether an association is causal: strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, and analogy. These were never a formal checklist, and Hill himself warned against treating them as rigid criteria, but they became the standard vocabulary for causal reasoning in epidemiology. The challenge they address is confounding—a third factor that influences both exposure and outcome and can create a spurious association10. Observational design and analysis are largely a campaign against confounding, through matching, stratification, and regression adjustment.
Survival analysis and the proportional hazards model
Many health outcomes are times: time to death, to relapse, to recovery. Survival analysis handles such data, and its central difficulty is censoring—some subjects are still alive or lost to follow-up when the study ends, so their event times are unknown but bounded. The Kaplan–Meier estimator, developed in the 1950s, estimates the survival curve from censored data11. The log-rank test compares survival between groups.
The decisive advance came in 1972, when David Cox published the proportional hazards model. It relates the hazard—the instantaneous risk of an event at a given time, given survival so far—to covariates through an exponential function, while leaving the baseline hazard unspecified. The model's great virtue is that it estimates the effect of each covariate (a hazard ratio) without needing to assume a particular shape for the underlying risk over time. It became, and remains, the workhorse of clinical and epidemiological research on time-to-event outcomes.
Meta-analysis and evidence synthesis
As trials multiplied, the problem became combining them. Meta-analysis is the statistical synthesis of results from multiple studies, producing a pooled estimate and a measure of heterogeneity—how much the studies disagree beyond chance. Early examples appear in the 1970s in education and psychology; in medicine, the approach was championed in the 1980s, notably by Richard Peto and colleagues, who used it to pool trials of beta-blockers after heart attack and of aspirin in vascular disease. The Cochrane Collaboration, founded in 1993, institutionalized systematic review and meta-analysis as the basis of evidence-based medicine. The method's power is to increase precision and detect small effects; its vulnerability is publication bias—positive studies are more likely to be published—and the temptation to pool studies that are not comparable.
Potential outcomes and modern causal inference
A deeper framework for causation emerged from the work of Donald Rubin and others from the 1970s onward: the potential outcomes model. Each person is imagined to have two potential outcomes, one under treatment and one under control; the causal effect is the difference between them, but only one is ever observed. This "fundamental problem of causal inference" makes causal effects estimable only under assumptions, and the framework clarifies exactly what those assumptions are. It connects naturally to randomization (which makes the groups exchangeable) and to observational methods such as propensity scores—the probability of treatment given covariates—which can balance treated and untreated groups on measured confounders. The framework, sometimes called the Rubin causal model, has become the common language linking trials, observational studies, and modern "target trial" designs that emulate a randomized trial using observational data.
Computation, high dimensions, and the present
The late twentieth and early twenty-first centuries brought data at a scale earlier statisticians could not have imagined: genomic sequences, electronic health records, imaging, and wearable sensors. High-dimensional data—where the number of variables exceeds the number of subjects—broke classical regression, which cannot estimate more parameters than observations. The response was regularization: ridge regression, the lasso (introduced by Robert Tibshirani in 1996), and their descendants, which shrink or select variables to produce stable estimates. Machine learning methods—classification trees, random forests, neural networks—entered biostatistics as tools for prediction, and the field now debates how to validate them and how to move from prediction to causal understanding.
Bayesian methods, long dormant, revived with computing. Markov chain Monte Carlo, developed from the 1950s and popularized in statistics from the 1990s, made it possible to fit complex Bayesian models that had been analytically intractable. Bayesian approaches are now common in clinical trial design (adaptive trials that update as data arrive), in disease mapping, and in evidence synthesis.
The questions still open are recognizably the old ones in new dress: how to draw causal conclusions from observational data, how to interpret p-values and communicate uncertainty, how to handle the multiplicity of comparisons that large datasets invite, and how to make analyses reproducible12. The field's history is a record of learning, again and again, that counting carefully is harder than it looks—and that the discipline of stating assumptions and checking them against data is what turns numbers into knowledge.
References
- Biostatistics, Wikipedia
- Bayesian statistics, Wikipedia
- Bayes' theorem (History), Wikipedia
- Bayesian inference, Wikipedia
- Frequentist inference, Wikipedia
- Ronald Fisher, Wikipedia
- Randomized controlled trial (History), Wikipedia
- Case–control study, Wikipedia
- Cohort study, Wikipedia
- Confounding, Wikipedia
- Kaplan–Meier estimator, Wikipedia
- P-value, Wikipedia
The frameworks of Biostatistics
Each framework is a way of working the field has used: when it held, and the question it answers.
- 1662-1900
Political Arithmetic and Vital Statistics
How many people are born, live and die?
- 1763-Present
- 1820-1880
Numerical Method in Clinical Medicine
Do treatments work better than what they are compared with?
- 1839-1900
- 1900-Present
- 1948-Present
Randomized Controlled Trial
Does this treatment really cause better outcomes?
- 1950-Present
Survival Analysis
How long until an event happens, when some are lost to follow-up?
- 1951-Present
Observational Study Design and Causal Criteria
When does an association between exposure and disease reflect cause?
- 1970-Present
Regularization and Statistical Learning
How can many predictors be used without overfitting?
- 1974-Present
- 1976-Present
How the ideas connect
Each arrow runs from an earlier idea to the later one it shaped. The list below says what passed between them.
Sanitary Statistics grew out of Political Arithmetic and Vital Statistics
William Farr built the General Register Office's national system of cause-of-death classification and district mortality rates directly on the political-arithmetic tradition of Graunt and the life table, making vital statistics the instrument of sanitary reform.
Carried by William Farr; John Graunt; Natural and Political Observations upon the Bills of Mortality (1662).
Randomized Controlled Trial influenced by Frequentist Inference
Fisher's insistence on randomisation and significance testing supplied the design and analysis logic of the 1948 streptomycin trial, and the Neyman-Pearson error framework gave the trial its criteria for judging a treatment effect.
Carried by Ronald Fisher; Jerzy Neyman; Austin Bradford Hill.
Survival Analysis grew out of Political Arithmetic and Vital Statistics
The life table of Graunt and Halley is the ancestor of survival analysis, and the Kaplan-Meier estimator generalises the actuarial life-table calculation to individual censored event times.
Carried by John Graunt; Edmond Halley; Edward L. Kaplan.
Randomized Controlled Trial influenced by Numerical Method in Clinical Medicine
Louis's numerical method established the argument that aggregated group comparison corrects the misleading impressions of single cases, the premise on which the 1948 streptomycin trial's controlled comparison rests.
Carried by Pierre-Charles-Alexandre Louis; Austin Bradford Hill.
Observational Study Design and Causal Criteria influenced by Sanitary Statistics
Snow's comparison of cholera death rates by water company, an early natural experiment within sanitary statistics, supplied the model of comparing exposed and unexposed groups that observational cohort design later formalised.
Carried by John Snow; Richard Doll; Austin Bradford Hill.
Observational Study Design and Causal Criteria influenced by Frequentist Inference
Bradford Hill's causal viewpoints presuppose the frequentist significance test: strength and consistency of association are judged by whether observed rate differences exceed chance variation as computed in the Neyman-Pearson framework.
Carried by Austin Bradford Hill; Ronald Fisher.
Meta-analysis grew out of Frequentist Inference
Meta-analysis was built as a frequentist procedure for pooling effect estimates and testing their homogeneity, applying the significance-testing and estimation machinery of Fisher and Neyman-Pearson to collections of studies.
Carried by Gene V. Glass.
Meta-analysis influenced by Survival Analysis
Meta-analysis of clinical trials needed methods for combining time-to-event outcomes, and it borrowed the effect measures and censoring-aware estimates developed in survival analysis.
Beyond the field
Where Biostatistics's ideas came from in other fields, and what they led to there: the strongest 12 of 98 links.
Political Arithmetic and Vital Statistics led to Vital Statistics and Population Rates in Clinical Epidemiology
The English vital statistics system of the General Register Office grew directly out of political arithmetic, the seventeenth- and eighteenth-century tradition of counting births, deaths and populations. William Farr explicitly built on Graunt and the political arithmeticians when he constructed national mortality rates.
Carried by John Graunt; William Farr.
Political Arithmetic and Vital Statistics led to Life table in Epidemiologic Modeling
Graunt's life table in the Natural and Political Observations (1662) was built directly from the bills of mortality that political arithmetic collected, making the life table a branch of that enterprise.
Carried by John Graunt; Natural and Political Observations (1662).
Political Arithmetic and Vital Statistics led to Demographic statistics in Demography
Demographic statistics descends directly from the political-arithmetic and vital-statistics tradition of counting births and deaths and computing rates. The later discipline is the systematised continuation of that counting programme.
Carried by John Graunt; William Farr.
Political Arithmetic and Vital Statistics influenced Civil Registration and Vital Statistics in Global Health Metrics And Evaluation
The civil registration system introduced in England and Wales in 1837 drew directly on the political arithmetic tradition of Graunt and the sanitary statisticians, who had shown that routine counts of births, deaths and causes could be analysed for public health purposes. William Farr, the first compiler of abstracts, applied that quantitative tradition to the new national register.
Carried by John Graunt; William Farr.
Political Arithmetic and Vital Statistics influenced Actuarial and Frequency Tradition in Probability Theory
Political arithmetic, the statistical analysis of births and deaths by Graunt and Petty, supplied the empirical frequency data and the habit of reasoning from observed ratios that the actuarial tradition built upon.
Carried by John Graunt; William Petty.
Political Arithmetic and Vital Statistics led to Program Statistics and Vital Statistics Reporting in Public Health Program Evaluation
Political arithmetic, the numerical study of births and deaths begun by Graunt and Petty, supplied the rate calculations and population reasoning that civil registration and program statistics later institutionalised.
Carried by John Graunt; William Petty.
Political Arithmetic and Vital Statistics led to Vital statistics in Public Health Informatics
Political arithmetic, the counting of births and deaths for state and insurance purposes from Graunt onward, supplied the concepts and tables on which vital statistics was built. The nineteenth-century registration system generalised that tradition into routine national statistics of population health.
Carried by John Graunt; Natural and Political Observations (1662).
Political Arithmetic and Vital Statistics influenced Medical Statistics (Numerical Method) in History Of Medicine
Political arithmetic and vital statistics gave medical statistics its core instruments, the life table, rates and the comparison of grouped observations. The numerical method applied these demographic techniques to clinical and therapeutic questions.
Bayesian Inference led to Bayesian Survival Analysis in Survival Analysis
Bayesian survival analysis applies Bayesian inference to survival models: priors on hazard or regression parameters combined with censored-data likelihoods. The parametric and semiparametric Bayesian survival models of the 1980s grew directly out of Bayesian inference as a branch application.
Bayesian Inference influenced Bayesian Diagnostic Reasoning in Clinical Decision Science
Bayesian diagnostic reasoning applies Bayes's theorem to clinical evidence; Ledley and Lusted introduced Bayesian inference into medical diagnosis in 1959, treating symptoms as evidence updating disease probabilities.
Carried by Robert S. Ledley; Lee B. Lusted.
Bayesian Inference influenced Bayesian pharmacometrics in Pharmacometrics
Bayesian inference supplies the prior-likelihood-posterior machinery, MCMC sampling and predictive checks that Bayesian pharmacometrics applies to nonlinear mixed-effects PK/PD models.
Bayesian Inference led to Empirical Bayes in Statistical Inference
Empirical Bayes takes a Bayesian posterior analysis and estimates the prior from the pooled data rather than fixing it in advance. Robbins's 1955 article introduced the empirical Bayes approach as a branch of Bayesian inference.
Carried by Herbert Robbins.

Biostatistics is a planet in Nousfarer
Its city has a tower for each of these 11 frameworks. Land, walk in, and climb one concept at a time.
Land on BiostatisticsRelated fields
Fields its ideas are linked to
- Statistical Inference 16 links
- Survival Analysis 10 links
- Causal Inference 6 links
- Clinical Epidemiology 4 links
- History Of Medicine 3 links
- Epidemiology 3 links
- Bayesian Statistics 3 links
- Experimental Design 3 links
- Probability Theory 2 links
- Clinical Decision Science 2 links
- Diagnostic Test Interpretation 2 links
- Epidemiologic Modeling 2 links