Epidemiology: Understanding Disease Spread and Impact

In this article27 sections · 18 min read

Key Definitions The Founding Moment: John Snow and the Broad Street Pump Why Snow's Method Was Revolutionary The Limits of the Founding Myth Epidemiological Study Designs Compared Study Designs: The Epidemiologist's Toolkit Randomized Controlled Trials Cohort Studies Case-Control Studies Cross-Sectional Studies and Ecological Studies Causation: The Bradford Hill Criteria The Nine Criteria Infectious Disease Epidemiology: Epidemic Thresholds and R0 Understanding the Basic Reproduction Number Herd Immunity and Vaccination Thresholds The Framingham Heart Study: Epidemiology's Greatest Achievement COVID-19 and Modern Epidemiology What Worked What the Pandemic Revealed About Weaknesses Epidemiology's Ongoing Challenges The Problem of Nutritional Epidemiology Big Data and Machine Learning Global Health Equity Epidemiology vs. Public Health: Clarifying the Distinction Further Reading and Cross-References Sources & Further Reading Further Reading

Epidemiology: In the summer of 1854, a physician named John Snow walked the streets of Soho, London, knocking on doors and asking questions. More than five hundred people had died of cholera in the surrounding blocks within ten days. Snow had a theory about why, and he needed data to test it.

He was not looking for a bacterium - the germ theory of disease did not yet exist, and Snow had no microscope that could have helped him anyway. He was looking for a pattern in who was dying, where they lived, and where they drank their water. What he found would eventually be recognized as the founding act of modern epidemiology.

Snow mapped every death on a street plan of the area. The deaths clustered around a single water pump on Broad Street. He interviewed residents, tracking down the exceptions - a brewery nearby whose workers were healthy, a widow far away who had nonetheless died.

Each exception strengthened his case: the brewery workers drank beer; the widow's family had the Broad Street water delivered to her because she liked its taste. Snow persuaded local authorities to remove the pump handle.

The outbreak declined. He had identified the cause and implemented the intervention without ever knowing what cholera actually was at a biological level.

The story is, in many ways, too clean - historians have noted that the outbreak was already subsiding when the pump handle came off, and that Snow's later career was more complicated than the founding myth suggests.

But the method Snow demonstrated - systematic observation of disease patterns in populations, careful comparison of exposed and unexposed groups, and the use of evidence to guide public action - remains the core of what epidemiologists do today, whether they are tracking a new respiratory virus, investigating a cluster of cancer cases near an industrial facility, or studying the long-term cardiovascular effects of dietary patterns.

Epidemiology is the science of the distribution and determinants of disease in human populations, and the application of that science to the control of health problems. It is the discipline that asks: who gets sick, when, where, and why?

It is both a method and a practice, both a form of scientific inquiry and a set of tools for improving public health. To understand it is to understand how we know what we know about the causes and patterns of human disease.

"Epidemiology is the basic science of public health, and its practice requires both rigorous analytical thinking and the willingness to act on imperfect evidence under conditions of uncertainty." - Kenneth Rothman, Modern Epidemiology^[4]

Key Definitions

Epidemiology: The study of the distribution (who, when, where) and determinants (why, how) of health-related states and events in specified populations, and the application of this study to the control of health problems.^[5]

Incidence: The rate at which new cases of a disease occur in a population over a specified period.

Prevalence: The proportion of a population that has a disease or condition at a specific point in time or over a specified period.

Relative risk: The ratio of the incidence of disease in an exposed group to the incidence in an unexposed group. A relative risk of 2 means exposed individuals are twice as likely to develop the disease.

Confounding: A distortion of the apparent association between an exposure and an outcome caused by a third variable associated with both.

Bias: Any systematic error in the design, conduct, analysis, or interpretation of a study that causes a deviation from the true value.

Epidemic: The occurrence of a disease in a community or region in excess of what would normally be expected.

Pandemic: An epidemic occurring worldwide, crossing international boundaries and affecting large numbers of people.

R0 (basic reproduction number): The average number of secondary infections generated by one infectious person in a completely susceptible population with no interventions in place.

The Founding Moment: John Snow and the Broad Street Pump

John Snow's investigation of the 1854 cholera outbreak stands as epidemiology's origin story precisely because it demonstrates the field's core intellectual commitment: that systematic observation of disease patterns in populations can reveal causes and guide interventions, even in the absence of biological understanding.

Why Snow's Method Was Revolutionary

In 1854, the dominant explanation for cholera was the miasma theory: the belief that disease spread through 'bad air' emanating from filth, decay, and overcrowding. Snow's alternative hypothesis - that cholera spread through contaminated water - had no established biological mechanism to support it.

He could not point to a specific pathogen. What he had was a spatial pattern that was inconsistent with miasma theory and consistent with waterborne transmission.

The genius of Snow's approach was its combination of what we would now call spatial epidemiology (the geographic distribution of cases), ecological analysis (comparing mortality rates across areas served by different water supplies), and individual-level case investigation (the interviews that explained the exceptions).

His 1849 treatise had already argued for waterborne transmission based on the geographic distribution of cholera cases across London relative to different water company service areas. The Broad Street investigation gave him a concentrated natural experiment.

Snow's analysis of the competing water supplies of South London - some households served by the Southwark and Vauxhall Company, which drew water from the Thames below London's sewage outfalls, and others served by the Lambeth Company, which had moved its intake upstream - produced one of the first natural experiments in epidemiology.

Mortality rates were dramatically higher in households served by Southwark and Vauxhall, even after controlling for poverty and housing conditions. This large-scale analysis, published in 1855, was in some ways even more rigorous than the Broad Street investigation.^[1]

The Limits of the Founding Myth

The cholera bacterium, Vibrio cholerae, was not identified until 1883, when Robert Koch isolated it during an Egyptian outbreak. The biological confirmation of Snow's hypothesis came nearly thirty years after his death in 1858.

This temporal gap is itself instructive: epidemiology routinely identifies associations and informs interventions before the underlying biological mechanism is understood.

The field's power lies in its ability to work from patterns in population data rather than requiring complete mechanistic knowledge.

Epidemiological Study Designs Compared

Study type	Design	Strengths	Weaknesses	Best used for
Randomized controlled trial (RCT)	Participants randomly assigned to intervention or control	Controls known and unknown confounding; strongest causal evidence	Unethical for harmful exposures; expensive; may lack generalizability	Evaluating treatments, vaccines, preventive interventions
Cohort study	Follow exposed and unexposed groups forward in time; measure outcomes	Can study multiple outcomes; establishes temporal order; can calculate incidence	Slow and expensive for rare or slowly developing diseases; loss to follow-up	Chronic disease risk factors; occupational exposures; Framingham model
Case-control study	Compare prior exposures in people with disease vs. comparable controls	Efficient for rare diseases; relatively fast and cheap	Recall bias; cannot directly calculate incidence; susceptible to selection bias	Rare cancers; outbreak investigations
Cross-sectional study	Measure exposure and outcome simultaneously	Fast; cheap; good for prevalence estimation	Cannot establish temporal order; susceptible to prevalence-incidence bias	Prevalence surveys; hypothesis generation
Ecological study	Compare average exposures and outcomes across populations or time periods	Can study population-level exposures; uses existing data	Ecological fallacy: group-level associations may not hold at individual level	Pollution studies; policy evaluations; hypothesis generation
Natural experiment	Exploit policy changes or chance events that mimic random assignment	Approaches causal evidence without deliberate intervention	Cannot always identify clean natural experiments; may have limited generalizability	Policy evaluation; studying ethically impossible exposures

Study Designs: The Epidemiologist's Toolkit

Different research questions require different study designs, and the choice of design shapes both the strength of the evidence that can be produced and the practical constraints of the research.

Randomized Controlled Trials

The randomized controlled trial (RCT) is the gold standard for establishing causal effects. Participants are randomly assigned to receive either an intervention or a control condition.

If the randomization is successful, both known and unknown confounding variables are distributed equally between groups, and any difference in outcomes can be attributed to the intervention.

The RCT's limitation in epidemiology is that it is often unethical or impractical to randomize people to harmful exposures. You cannot randomly assign people to smoke cigarettes, eat diets high in trans fats, or live in communities with high levels of air pollution.

RCTs are therefore most useful for evaluating treatments, preventive interventions, and vaccines - areas where a potentially beneficial intervention is being tested rather than a harmful exposure being studied.

Cohort Studies

A cohort study follows a group of people over time, comparing those exposed to a factor of interest with those unexposed, and measuring how many in each group develop the outcome of interest.

The Framingham Heart Study, begun in 1948 with 5,209 residents of Framingham, Massachusetts, is the most famous cohort study in the history of epidemiology.

Participants were enrolled before they developed cardiovascular disease, their risk factors (blood pressure, cholesterol, smoking, exercise habits, diet) were measured at regular intervals, and they were followed for decades to see who developed disease and died.

The Framingham Study identified the major risk factors for cardiovascular disease - the term 'risk factor' itself was coined in a Framingham paper - and established the evidence base for much of modern preventive cardiology.

The study has been extended to include children and grandchildren of the original participants, providing multi-generational data of extraordinary richness.

Case-Control Studies

Case-control studies begin with the outcome. Researchers identify people who have already developed the disease (cases) and comparable people who have not (controls), then compare how frequently each group was previously exposed to the factor under investigation.

Case-control designs are efficient for studying rare diseases: you begin with a fixed number of people who have already experienced the outcome, rather than following a large cohort and waiting for disease to develop.

The main vulnerability of case-control studies is recall bias: cases may remember and report past exposures differently from controls, particularly when the exposure is something they have been told might be related to their disease.

Detailed, validated questionnaires and the use of objective records (employment files, prescriptions, biological samples) where available can reduce but not eliminate this problem.

Cross-Sectional Studies and Ecological Studies

Cross-sectional studies measure exposure and outcome simultaneously in a population at a single point in time. They are efficient for estimating prevalence and can identify associations, but cannot establish temporal order (whether exposure preceded disease) and are therefore weak evidence for causation.

Ecological studies examine associations at the group level rather than the individual level - comparing average disease rates and average exposure levels across countries, regions, or time periods. The correlation between a country's average dietary fat intake and its heart disease mortality rate is an ecological association.

Ecological studies are useful for generating hypotheses and can analyze exposures that vary at the group level, but are subject to the ecological fallacy: associations observed at the group level may not hold at the individual level.

Causation: The Bradford Hill Criteria

The most fundamental challenge in observational epidemiology is inferring causation from association. A statistical association between an exposure and an outcome can arise from three sources: confounding, bias, or a genuine causal relationship.^[8]

The Bradford Hill criteria, proposed by Sir Austin Bradford Hill in 1965, provide a framework for assessing the totality of evidence and judging how likely an observed association is to be causal.

The Nine Criteria

Strength: Stronger associations are more likely to be causal. A relative risk of 10 is harder to explain away by confounding than a relative risk of 1.2.

Consistency: The association should be observed in multiple studies, across different populations, different investigators, and different study designs.

Specificity: The exposure leads to a specific disease, not a wide range of unrelated outcomes. Hill recognized this criterion was weak - many exposures cause multiple effects - but regarded it as supporting evidence when present.

Temporality: The cause must precede the effect. This is the only criterion Hill regarded as strictly necessary: an association where the supposed effect precedes the supposed cause cannot be causal.

Biological gradient: A dose-response relationship, where increasing exposure is associated with increasing risk, provides stronger evidence for causation.

Plausibility: The association makes biological sense given current knowledge. Hill cautioned that plausibility was limited by current knowledge - an association can be causal even if no mechanism is yet known.

Coherence: The causal interpretation should not fundamentally conflict with the known natural history of the disease.

Experiment: If removal of the exposure (through natural experiments, policy changes, or interventions) reduces disease incidence, this provides strong supporting evidence.

Analogy: Similar exposures have similar effects in related domains, providing prior plausibility for the association under investigation.

Hill was explicit that these were considerations for judgment, not a checklist.^[2] The totality of the evidence, weighed against these criteria, should inform conclusions about causation - always understood as probabilistic rather than certain.

Infectious Disease Epidemiology: Epidemic Thresholds and R0

The epidemiology of infectious disease adds a layer of complexity absent from the study of chronic non-communicable diseases: the transmission dynamic.

Whether an outbreak grows or dies out depends not only on the biology of the pathogen and the characteristics of the host population but on the mathematical relationship between transmission and removal.

Understanding the Basic Reproduction Number

R0, the basic reproduction number, encapsulates this relationship in a single parameter.^[6] An R0 greater than 1 means each infected person infects more than one other person on average, and the outbreak will grow exponentially. An R0 less than 1 means the outbreak will die out.

The speed of exponential growth is determined by both R0 and the serial interval (the time between successive generations of infection).

Measles, with an R0 of 12 to 18, is one of the most contagious pathogens known. The original SARS-CoV-2 strain had an estimated R0 of around 2.5. The Omicron variant had an estimated R0 of 8 to 15, which is why it spread so much more rapidly than earlier variants despite similar or lower rates of severe disease.

R0 is not a fixed property of a pathogen. It is a composite measure that depends on biological factors (how much virus an infected person sheds, how long they remain infectious) and behavioral and social factors (how many people they contact, how close those contacts are).

Interventions that reduce contact rates - social distancing, school closures, mask wearing - reduce the effective reproduction number (Rt), the reproduction number in a population that is neither fully susceptible nor fully immune.

Herd Immunity and Vaccination Thresholds

The concept of herd immunity - the indirect protection that unvaccinated or susceptible individuals receive when a sufficient proportion of the population is immune - follows mathematically from R0. The herd immunity threshold is 1 - (1/R0).

For measles with R0 of 15, approximately 93 percent of the population must be immune to prevent epidemic spread. For a disease with R0 of 2.5, only 60 percent immunity is needed.

These calculations assume uniform mixing of the population - everyone has an equal probability of contacting anyone else.

In reality, populations are structured by geography, social networks, and behavior, which means that herd immunity thresholds can vary significantly across sub-populations and that local outbreaks can occur even when overall population immunity exceeds the theoretical threshold.

The Framingham Heart Study: Epidemiology's Greatest Achievement

The Framingham Heart Study deserves extended attention because it transformed not only understanding of cardiovascular disease but the practice of preventive medicine and the entire concept of risk factor medicine.

In 1948, cardiovascular disease was the leading cause of death in the United States, but its causes were poorly understood. The study's original design enrolled 5,209 men and women aged 30 to 62 from the town of Framingham, Massachusetts.^[3]

They underwent detailed physical examination at enrollment and returned every two years for repeat examination. The study was designed to track participants until they either developed cardiovascular disease or died.

Over the following decades, the Framingham cohort generated a succession of landmark findings. The 1961 paper coining the term 'risk factor' identified elevated cholesterol, hypertension, and smoking as independent predictors of cardiovascular disease.

The concept of attributable risk - quantifying how much disease in the population could be attributed to specific risk factors - emerged from Framingham analyses.

Later studies identified the importance of HDL cholesterol, established that hypertension was a treatable risk factor (not simply an inevitable consequence of aging), and documented the cardiac consequences of obesity, diabetes, and physical inactivity.

The Framingham Offspring Study, begun in 1971 with the children of original participants, allowed multi-generational analysis. The Third Generation Study began in 2002.

By the early 21st century, Framingham investigators were publishing genome-wide association studies linking genetic variants to cardiovascular risk, using the deep phenotyping and decades of follow-up data that no other cohort could match.

COVID-19 and Modern Epidemiology

The COVID-19 pandemic that began in late 2019 was the most consequential public health event since the 1918 influenza pandemic, and it tested epidemiology - as a science, as a professional practice, and as a form of public communication - in ways that will be studied for decades.

What Worked

Seroprevalence studies - population surveys measuring the proportion of people with antibodies to SARS-CoV-2 - rapidly revealed that confirmed case counts dramatically underestimated true infection rates, allowing better-grounded estimates of infection fatality rates.^[7]

Cohort studies quickly identified major risk factors for severe disease. Genomic sequencing, linked to epidemiological investigation, allowed rapid identification and tracking of viral variants. Vaccine trials conducted during the pandemic demonstrated unprecedented speed without sacrificing scientific rigor.

What the Pandemic Revealed About Weaknesses

The pandemic exposed the vulnerability of public health surveillance infrastructure in many countries. Contact tracing - one of the oldest and most effective tools for controlling infectious disease outbreaks - depends on trained staff, established protocols, and community trust built before an emergency.

Countries that had invested in this infrastructure controlled early outbreaks far more successfully than those that had not.

The preprint culture that had developed in science over the preceding decade accelerated during the pandemic. Researchers posted preliminary results before peer review, enabling rapid dissemination but also high-profile retractions and public confusion when early findings were not replicated.

The communication of uncertainty - the honest statement that models make projections under specific assumptions, that evidence is preliminary, that recommendations may change as knowledge develops - proved extraordinarily difficult in a media environment hungry for certainty and in a political environment where uncertainty could be weaponized.

Epidemiology's Ongoing Challenges

Modern epidemiology faces methodological and practical challenges that Snow could not have imagined.

The Problem of Nutritional Epidemiology

Nutritional epidemiology - the study of diet and health - has been criticized for producing an unusually high rate of findings that are not subsequently replicated, for issuing public health guidance that reverses (eggs are bad, eggs are good; dietary fat causes heart disease, dietary fat is not the main culprit), and for relying on self-reported dietary recall data that have known inaccuracies.

The field is attempting to address these problems through better measurement technologies (metabolomics, which measures the metabolic products of dietary intake directly from blood samples) and through better study designs (Mendelian randomization, which uses genetic variants as natural instruments for random assignment to different dietary exposures).

Big Data and Machine Learning

The availability of large electronic health record datasets, genomic data, and consumer behavior data has created new opportunities for epidemiological research at scales previously impossible. Machine learning methods can identify complex patterns in high-dimensional data that traditional regression approaches cannot capture.

But these methods also create new risks: overfitting to data artifacts, finding associations that are statistically robust but biologically meaningless, and - given the proprietary nature of much big data - reducing the reproducibility and transparency that scientific inference requires.

Global Health Equity

Epidemiology has historically been conducted disproportionately in high-income countries, studying diseases prevalent in those populations, with funding from those countries' institutions.

Global health epidemiology requires not only extending data collection to low- and middle-income settings but rethinking research priorities - studying the diseases that kill the most people worldwide rather than the diseases most prevalent in wealthy countries - and building research capacity in lower-income settings so that epidemiology is done locally by local researchers rather than imported from outside.

Epidemiology vs. Public Health: Clarifying the Distinction

The relationship between epidemiology and public health is one of method to practice. Epidemiology is the diagnostic science of public health - the tools of investigation used to identify who is at risk, why, and how that risk can be modified.

Public health is the broader enterprise of using that knowledge to protect and improve population health through policy, education, regulation, and direct service delivery.

An epidemiologist might demonstrate that a specific air pollutant causes measurable cardiovascular harm at concentrations currently permitted by law.

The public health decision - whether to tighten regulations, how quickly, what the economic tradeoffs are, how to communicate the risk to the public - is made by public health officials, policymakers, and ultimately the political process. Epidemiologists inform that process with evidence.

They do not make the policy.

This distinction matters because it clarifies the appropriate scope of epidemiologists' authority. During the COVID-19 pandemic, epidemiologists (and virologists, and other scientists) were sometimes placed in a position of appearing to make policy - to decree lockdowns, mandate vaccines, determine school closures.

In fact, those were political decisions informed by scientific evidence. Conflating the two - treating epidemiological findings as if they directly determined policy - generated backlash against scientists and created the false impression that policy disagreements were scientific disagreements.

Sources & Further Reading

Snow, J. (1855). On the Mode of Communication of Cholera (2nd ed.). John Churchill.
Hill, A.B. (1965). The environment and disease: Association or causation? Proceedings of the Royal Society of Medicine, 58, 295-300.
Dawber, T.R., Meadors, G.F., & Moore, F.E. (1951). Epidemiological approaches to heart disease: The Framingham Study. American Journal of Public Health, 41(3), 279-286.
Rothman, K.J., Greenland, S., & Lash, T.L. (2008). Modern Epidemiology (3rd ed.). Lippincott Williams & Wilkins.
Gordis, L. (2014). Epidemiology (5th ed.). Elsevier Saunders.
Anderson, R.M., & May, R.M. (1991). Infectious Diseases of Humans: Dynamics and Control. Oxford University Press.
Chadeau-Hyam, M., et al. (2020). Latent class modelling of the SARS-CoV-2 serological response. Scientific Reports, 10, 21484.
Szklo, M., & Nieto, F.J. (2019). Epidemiology: Beyond the Basics (4th ed.). Jones & Bartlett Learning.

Frequently asked questions

Epidemiology and public health are related but distinct disciplines that are frequently conflated, even by people working in health sciences. Epidemiology is fundamentally a scientific method, a set of tools and study designs used to investigate the distribution and determinants of disease in populations. Epidemiologists ask questions like: who is getting sick, when, where, and why? They design studies, collect data, analyze patterns, and draw inferences about causation.

Public health, by contrast, is a broader field of practice concerned with protecting and improving the health of entire populations. It encompasses epidemiology as one of its core tools but also includes health policy, health education, environmental health, health administration, and clinical preventive services.

An analogy helps clarify the distinction: if public health is the field of medicine treating society as the patient, then epidemiology is the diagnostic science, the equivalent of laboratory testing and clinical examination, that informs the diagnosis. A public health agency might decide to mandate vaccination against a disease, enforce clean water standards, or launch an anti-smoking campaign. The decision to act and the design of the intervention belongs to public health.

The evidence base for whether those interventions work, who is most at risk, and how the disease spreads comes from epidemiology. The distinction matters practically. Epidemiologists produce knowledge; public health officials apply it.

A classic example: epidemiologists produced decades of evidence linking cigarette smoking to lung cancer before public health authorities used that evidence to introduce restrictions on tobacco advertising, fund cessation programs, and implement smoke-free workplace laws. The epidemiological work and the public health response were separate enterprises that depended on each other.

In academic settings, epidemiology is usually a department within schools of public health, but the research methods of epidemiology are also used in clinical medicine, veterinary science, environmental science, and the social sciences, wherever investigators want to understand the distribution of outcomes in a population and the factors that determine them.

John Snow’s investigation of the 1854 Broad Street cholera outbreak in London is one of the founding narratives of epidemiology, and with good reason: it demonstrated that careful, systematic observation of disease patterns could identify a cause and enable an intervention even in the complete absence of germ theory. In 1854, the dominant medical theory was miasma, the belief that disease spread through bad air produced by filth and decay. Snow was skeptical.

He had already published a treatise in 1849 arguing that cholera was transmitted through contaminated water rather than air. The Broad Street outbreak, which killed over 500 people in ten days in the Soho district of London, gave him the opportunity to test this hypothesis in detail. Snow’s method combined what we would today call a geographic cluster analysis with a careful case-control investigation.

He mapped every death from cholera in the area, marking each one on a street map and noting the location of water pumps. The cluster of deaths was centered unmistakably on the Broad Street pump. But Snow went further: he interviewed households, identified exceptions, and explained them. A brewery nearby had almost no cases, the workers drank beer, not pump water.

A widow who lived far from Broad Street had died of cholera, investigation revealed she preferred the taste of Broad Street water and had it delivered. These details eliminated alternative explanations and strengthened the causal inference. Snow persuaded local authorities to remove the handle from the Broad Street pump, and the outbreak declined. The evidence later showed that the pump’s water supply had been contaminated by a nearby cesspit.

Beyond its immediate impact, Snow’s work established core principles that still define epidemiological practice: map the cases, define the population at risk, compare rates between exposed and unexposed groups, generate a hypothesis, test it against exceptions, and use the evidence to guide intervention.

It also showed that epidemiology could operate as a form of detective work, reasoning from patterns in population data to specific causal claims, before any understanding of the underlying biological mechanism existed.

The Bradford Hill criteria are a set of nine considerations proposed by the British epidemiologist Austin Bradford Hill in a landmark 1965 lecture to the Royal Society of Medicine. Hill’s purpose was to address what is arguably the central methodological challenge in observational epidemiology: how do we move from the observation that two things are statistically associated to the conclusion that one causes the other?

His criteria, often called viewpoints or guidelines rather than rules, provide a structured framework for evaluating that inferential leap.

The nine criteria are: strength of association (stronger associations are more likely to be causal), consistency (the association is observed across multiple studies and populations), specificity (the cause leads to one specific effect), temporality (the cause must precede the effect, the only criterion Hill regarded as strictly necessary), biological gradient (a dose-response relationship, where more exposure leads to more disease), plausibility (the association makes

biological sense given current knowledge), coherence (the causal interpretation does not conflict with known facts about the disease), experiment (evidence from natural experiments or interventions supports the relationship), and analogy (similar causes have similar effects in related domains).

Hill developed these criteria partly as a response to debates over smoking and lung cancer. By the early 1960s, multiple large cohort studies had established a robust statistical association between cigarette smoking and lung cancer, but the tobacco industry and some scientists argued this was not sufficient evidence of causation.

Hill’s criteria provided a framework for arguing that the totality of the evidence, its strength, consistency across countries and populations, biological gradient, and plausibility, justified a causal conclusion even without the ability to conduct a randomized controlled trial in humans. In modern practice, the criteria are used to evaluate evidence from observational studies where randomization is impossible. They are not a checklist that, if ticked, proves causation.

Rather, they are considerations that, taken together, help epidemiologists and policymakers judge how confident they should be that an association is causal, and how much confidence is sufficient to justify action. They remain the standard reference point in environmental health, occupational medicine, and nutritional epidemiology, where randomized trials are often impractical or unethical.

These three study designs sit at the core of epidemiological methodology and represent progressively stronger levels of evidence for causal inference, though each has distinct strengths and appropriate uses depending on the research question. A cohort study follows a group of people (the cohort) over time, comparing those who are exposed to a factor of interest with those who are not, and measuring how many in each group develop the outcome.

The Framingham Heart Study, begun in 1948 with 5,209 residents of Framingham, Massachusetts, is the archetypal prospective cohort study: participants were enrolled before they developed heart disease, their risk factors were measured, and they were followed for decades to see who developed disease. Cohort studies are powerful for measuring incidence (the rate of new disease), calculating relative risks, and studying the natural history of disease over time.

Their main disadvantages are cost, time, and the possibility of attrition, participants dropping out over years or decades. A case-control study works backward from outcome to exposure. Researchers identify people who already have the disease (cases) and comparable people who do not (controls), then compare how often each group was previously exposed to the factor of interest.

Case-control studies are efficient for studying rare diseases, because you begin with a fixed number of cases rather than waiting for disease to develop in a large cohort. They are also faster and less expensive. The trade-off is that they rely on retrospective recall of past exposures, which introduces the risk of recall bias, cases may remember and report past exposures differently than controls.

A randomized controlled trial (RCT) randomly assigns participants to receive either an intervention (such as a drug, vaccine, or behavioral program) or a control condition. Randomization, if done correctly, means that both known and unknown confounding factors are equally distributed between the groups, which makes the RCT the strongest design for establishing causation.

The RCT’s limitation is that randomization is often impossible, you cannot randomly assign people to smoke, eat poorly, or live in polluted areas, and sometimes unethical. RCTs are therefore most common in evaluating treatments, vaccines, and health interventions rather than studying the causes of naturally occurring diseases.

Confounding and selection bias are the two most fundamental threats to the validity of epidemiological studies, and understanding them explains much of the design complexity of modern epidemiology. Confounding occurs when the apparent association between an exposure and an outcome is actually produced, at least in part, by a third variable, the confounder, that is associated with both. A classic example: studies find that coffee drinkers have higher rates of lung cancer.

Does coffee cause lung cancer? Almost certainly not. The confounding variable is smoking: people who drink coffee are also more likely to smoke, and smoking causes lung cancer. The coffee-lung cancer association is confounded by smoking, and once you control for smoking in the analysis, the association largely disappears.

Confounders are controlled for through study design (restriction, matching), statistical analysis (multivariable regression), or through randomization, which distributes confounders equally between groups. Selection bias occurs when the people included in a study are systematically different from the population the study is meant to represent, in ways that distort the results.

A common form is healthy worker bias: studies of occupational exposures often compare workers to the general population, but workers are, by definition, healthy enough to hold jobs, making them a systematically healthier comparison group. Another form is survival bias: if you study patients already in hospital, you have automatically excluded people who died before reaching the hospital, potentially making diseases look less lethal than they are.

Both problems are pervasive and can entirely reverse the apparent direction of an association. They explain why single studies, however large, rarely settle epidemiological questions definitively. The scientific community uses systematic reviews and meta-analyses, pooling results across multiple studies with different designs and different potential biases, to arrive at more robust conclusions.

They also explain why epidemiologists invest heavily in study design: a well-designed study with a modest sample size is often more informative than a large study with serious bias built into its design.

R0, pronounced ‘R naught’ and called the basic reproduction number, is one of the most important summary statistics in infectious disease epidemiology. It represents the average number of secondary infections that one infectious person will generate in a completely susceptible population, that is, before any immunity exists and before any control measures are in place.

R0 is not a fixed property of a pathogen; it depends on the biology of the organism (how long someone is infectious, how much virus they shed), the route of transmission, and crucially, the social and behavioral context, how many people an infected person typically contacts and how close those contacts are. The R0 threshold that matters most is 1.

If R0 is less than 1, each infected person infects fewer than one other person on average, and the outbreak will die out without intervention. If R0 is greater than 1, the number of cases will grow exponentially. The higher R0 is above 1, the faster the epidemic grows. Seasonal influenza has an R0 of roughly 1.2 to 1.4.

Measles, one of the most contagious pathogens known, has an R0 of 12 to 18, meaning one infectious person can infect up to 18 others in a fully susceptible population.

Early estimates of the original SARS-CoV-2 strain placed its R0 around 2.5, while the Omicron variant had an estimated R0 of 8 to 15.R0 also determines the herd immunity threshold, the proportion of a population that needs to be immune (through vaccination or prior infection) to prevent an epidemic from growing. The formula is 1 - (1/R0). For measles with R0 of 15, approximately 93 percent of the population needs to be immune to achieve herd immunity.

For a disease with R0 of 2.5, only 60 percent immunity is required. The concept of R0 is therefore central to vaccine policy, informing how high vaccination coverage needs to be to protect a population. During the COVID-19 pandemic, the rapid evolution of variants with successively higher R0 values meant that herd immunity thresholds kept shifting, complicating vaccination strategy.

The COVID-19 pandemic was both a vindication and a stress test for epidemiology as a science. It demonstrated the value of epidemiological tools while also exposing gaps in data infrastructure, institutional coordination, and the communication of scientific uncertainty to the public. On the positive side, epidemiological methods worked.

Contact tracing, the core epidemiological tool for controlling infectious disease outbreaks, successfully contained COVID-19 in several East Asian countries in the early months.

Seroprevalence studies, which test stored blood samples or population samples for antibodies to estimate how many people had actually been infected (as opposed to diagnosed), revealed that reported case counts dramatically undercounted true infections, allowing better estimation of infection fatality rates. Cohort studies rapidly identified risk factors for severe disease, older age, obesity, diabetes, cardiovascular disease, within weeks of the pandemic’s start.

Vaccine trials conducted during the pandemic were among the fastest and most closely watched RCTs in history, generating robust efficacy and safety data within months. The limitations were equally instructive. Early in the pandemic, data collection was inconsistent and incompatible across countries, making international comparison unreliable. Case definitions changed over time.

Testing availability varied enormously, making case counts an unreliable measure of epidemic trajectory. Preprint culture, researchers posting results before peer review, accelerated the spread of preliminary findings but also caused high-profile retractions and public confusion.

The pandemic also revealed the costs of weakened public health surveillance infrastructure in many countries: when an outbreak demands rapid, high-quality data, the systems to collect it need to exist in advance. The pandemic also raised fundamental questions about how epidemiological findings should be communicated.

Probabilistic, conditional statements, ‘under these assumptions, this model projects this range of outcomes’, proved difficult to convey to a public and media that wanted certainty. The rapid evolution of evidence was interpreted as scientists ‘changing their minds’ rather than as normal scientific updating. These communication challenges have since spurred considerable work on science communication in public health contexts.

Want to write for us?

Pitch a guide to When Notes Fly

Evergreen subjects, primary sources, and a four-stage review. Contributors are chosen for knowledge of the specific topic.

Read the content guide →

Share this article