Risk factors and causes are related but fundamentally different things, and conflating them is one of the most common errors in how health information gets communicated and understood. A risk factor is anything statistically associated with a higher chance of developing a disease or condition. A cause is something that actually produces the disease through a biological or physical mechanism. Every cause is a risk factor, but most risk factors are not causes. The gap between those two categories has led to billions of dollars in failed drug trials, misguided public health advice, and confused patients making decisions based on correlations mistaken for mechanisms.
What a Risk Factor Actually Is
When researchers identify a risk factor, they are saying that people who have or are exposed to that factor develop a given outcome more often than people who are not. That is a statistical observation. It does not tell you whether the factor is directly producing the outcome, whether something else is producing both the factor and the outcome at the same time, or whether the outcome is actually producing the factor in reverse. Risk factors can be sorted into modifiable ones you can change, like smoking or physical inactivity, and non-modifiable ones you cannot, like age, sex, or genetic background.1Frontiers in Public Health. Management and Prevention Strategies for Non-communicable Diseases (NCDs) and Their Risk Factors That classification matters enormously for practical purposes, but it still does not answer whether a given risk factor is a cause.
Consider an analogy. Fire trucks are strongly associated with house fires. The more fire trucks you see, the bigger the fire tends to be. Fire trucks are a reliable “risk factor” for fire damage. But no one would say fire trucks cause the damage. A hidden third variable, the fire itself, explains both the trucks and the destruction. This sounds obvious with fire trucks. It is far less obvious when the relationship involves cholesterol levels, body weight, or a biomarker your doctor measured in a blood test.
Why the Distinction Matters for Your Health
When a risk factor gets mistaken for a cause, the logical next step is to intervene on it. If high levels of a certain blood marker are associated with heart disease, the reasoning goes, then lowering that marker should reduce heart disease. Sometimes that logic works beautifully: high LDL cholesterol is both a risk factor and a causal contributor to cardiovascular disease, and lowering it with medication genuinely reduces heart attacks. But other times, the logic fails spectacularly.
HDL cholesterol, the so-called “good cholesterol,” has been one of the most instructive failures. For decades, observational studies consistently showed that people with higher HDL levels had fewer heart attacks. HDL looked like a protective cause. Drug companies invested heavily in medications designed to raise HDL, reasoning that boosting it would protect the heart. Three major trials of drugs called CETP inhibitors, which successfully raised HDL levels, failed to reduce cardiovascular events.2PubMed. New Therapeutic Approaches to the Treatment of Dyslipidemia Genetic studies using a technique called Mendelian randomization confirmed that HDL cholesterol is probably not a causal factor for heart disease at all. It appears to be a marker that travels alongside genuinely protective lifestyle factors and socioeconomic conditions, not a driver of protection itself.3PubMed Central. High-Density Lipoprotein Cholesterol in Atherosclerotic Cardiovascular Disease Risk Assessment: Exploring and Explaining the “U”-Shaped Curve
The hormone replacement therapy story is another cautionary tale. Observational studies in the 1980s and 1990s found that postmenopausal women who took hormone therapy had lower rates of heart disease. Doctors prescribed it widely for heart protection. Then randomized trials showed the opposite: hormone therapy did not reduce heart disease and in some cases increased it. The original association was likely driven by the fact that women who chose hormone therapy tended to be healthier and wealthier to begin with, a type of bias called “healthy user” confounding.4PubMed. The discrepancy between observational studies and randomized trials of menopausal hormone therapy: did expectations shape experience? The risk factor (not using hormones) was confused with a cause of heart disease, and the result was years of misguided prescribing.
The Three Traps That Create False Causes
Three major pitfalls cause risk factors to be mistaken for causes. Understanding them helps you read health news with sharper eyes.
The first is confounding. This is the fire-truck problem: a hidden variable causes both the exposure and the outcome, creating a statistical link between two things that have no direct relationship. Confounding can distort measures of association and make a harmless exposure look dangerous or a dangerous one look harmless.5PubMed Central. Methodological issues of confounding in analytical epidemiologic studies Aluminium in drinking water has been positively associated with Alzheimer’s disease in multiple studies, but the relative risks are small enough that confounding factors could plausibly explain the entire association.6Brain Research Bulletin. Aluminium as a risk factor in Alzheimer’s disease, with emphasis on drinking water People with higher aluminium exposure may also differ in diet, occupation, or other factors that independently affect brain health.
The second trap is reverse causation. Sometimes the disease causes the risk factor, not the other way around. A striking example involves body weight and death. Being underweight is associated with higher mortality, which might tempt you to conclude that being thin is dangerous. But a 45-year follow-up study found that this excess risk among underweight adults shrank steadily as researchers excluded deaths in the early years of follow-up. The association was driven by people who were underweight because they were already getting sick, not people who got sick because they were underweight.7PubMed Central. Underweight as a risk factor for respiratory death in the Whitehall cohort study: exploring reverse causality using a 45-year follow-up In dialysis patients, lower cholesterol is associated with worse outcomes, but this is not because low cholesterol is harmful. In that population, low cholesterol is a marker of poor nutritional status and underlying illness.8PubMed. Reverse epidemiology: a confusing, confounding, and inaccurate term
The third trap is mediation, where a factor sits on the pathway between a cause and a disease but is not the root cause itself. High blood pressure is a risk factor for stroke, and it genuinely contributes to causing stroke. But high blood pressure is itself caused by things like diet, stress, and kidney function. If you only target the blood pressure number without addressing what drove it up, you may get a different result than expected. The causal chain matters, not just the individual link you happen to measure.
How Scientists Try to Tell the Difference
Epidemiologists have been wrestling with this problem for at least sixty years. In 1965, Austin Bradford Hill published nine considerations, often called the Bradford Hill criteria, for judging whether an observed association is likely to be causal. These include the strength of the association, its consistency across different studies and populations, whether the exposure precedes the disease, whether a dose-response relationship exists, and whether the relationship is biologically plausible.9PubMed Central. Applying the Bradford Hill criteria in the 21st century: how data integration has changed causal inference in molecular epidemiology Hill himself was careful to call these “viewpoints,” not rules. No single criterion is sufficient or necessary to prove causation. They are a checklist for structured thinking, not a formula that spits out a yes or no answer.
Since Hill’s time, causal thinking has advanced considerably. Three major frameworks have emerged. Directed acyclic graphs, or DAGs, are visual maps of how variables relate to each other. They help researchers figure out which variables to adjust for in a study and which ones would actually introduce bias if adjusted for.10PubMed Central. Tutorial on directed acyclic graphs Sufficient-component cause models, sometimes called “causal pies,” describe how diseases result from combinations of factors. No single factor alone is sufficient to cause the disease; instead, different combinations of component causes can each independently produce the outcome.11PubMed Central. The causal pie model: an epidemiological method applied to evolutionary biology and ecology And the counterfactual framework asks what would have happened to a person if they had not been exposed to the factor in question.12PubMed Central. Causal inference based on counterfactuals Each of these tools addresses a different aspect of the gap between association and causation, and modern epidemiologists often use them together with the older Bradford Hill viewpoints.13PubMed Central. Assessing causality in epidemiology: revisiting Bradford Hill to incorporate developments in causal thinking
One tool that has transformed the field in the past two decades is Mendelian randomization. The idea is elegant: because your genetic variants are randomly assigned at conception, they are not affected by the confounding factors that plague observational studies. If a genetic variant that raises your LDL cholesterol also raises your heart attack risk, that is strong evidence that LDL itself is doing the damage, not some lifestyle factor correlated with high LDL. The method works like a natural experiment, mimicking the design of a randomized trial without actually randomizing anyone.14PubMed Central. Mendelian Randomization: Concepts and Scope Mendelian randomization was instrumental in debunking the idea that HDL cholesterol is causally protective, and it has been used to test whether dozens of other risk factors genuinely cause disease or merely tag along with something that does.15PubMed. Mendelian Randomization in Cardiovascular Research: Establishing Causality When There Are Unmeasured Confounders
Most Diseases Do Not Have a Single Cause
Part of why the risk-factor-versus-cause question is so tricky is that most chronic diseases are not caused by one thing. Heart disease, cancer, diabetes, and dementia all arise from webs of interacting factors: genes, behaviors, environmental exposures, and random biological events. The causal pie model captures this well. A given disease outcome requires a complete set of component causes, but multiple different sets can each independently produce it.16PubMed Central. Attributing diseases to multiple pathways: a causal-pie modeling approach Smoking is a component cause in several different causal pies for lung cancer, which is why it is such a powerful risk factor. But some people who never smoke develop lung cancer through an entirely different causal pie involving radon exposure, genetic susceptibility, or other factors.
This means a factor can be genuinely causal without being the sole cause, and it can be a strong risk factor without being causal at all. The relationship between cause and risk is not a clean binary. It is a spectrum, and where a given factor sits on that spectrum determines how useful it is as a target for prevention or treatment.
When Risk Factors Are Good Enough
All of this might make it sound like non-causal risk factors are useless. They are not. For prediction and screening, a risk factor does not need to be a cause to be valuable. Biomarkers that merely correlate with disease progression can still be clinically useful for diagnosing a condition or estimating a patient’s prognosis, even if the biomarker is not on the causal pathway.17PubMed Central. Biomarkers and Surrogate Endpoints In Clinical Trials Your doctor measures C-reactive protein to gauge inflammation and disease risk, not because CRP itself is causing the damage. Age is a powerful risk factor for almost every chronic disease, and while you obviously cannot intervene on aging itself, knowing that age raises risk helps target screening and prevention to the people most likely to benefit.18PubMed Central. Understanding Modifiable and Unmodifiable Older Adult Fall Risk Factors to Create Effective Prevention Strategies
The problem arises specifically when non-causal risk factors are treated as intervention targets. If something merely correlates with disease, fixing the correlation does not fix the disease. A Mendelian randomization study of healthcare costs found that waist circumference, body mass index, and systolic blood pressure had genuine causal effects on costs, with a standard-deviation increase in waist circumference corresponding to roughly a 23% increase in healthcare spending. But certain clinically measured biomarkers like albumin, CRP, and vitamin D showed no causal effect on costs despite being associated with them in observational data.19Nature Communications. Quantifying the causal impact of biological risk factors on healthcare costs Spending money to move a non-causal marker is spending money on nothing.
How the Message Gets Garbled
If the scientific community has all these tools for distinguishing risk factors from causes, why does the confusion persist so stubbornly in the public sphere? A major part of the answer lies in how research findings get communicated. A replication study of health news reporting found that about half of news articles used stronger causal language than the journal articles they were covering. And the distortion was traceable: when university press releases exaggerated the causal language, roughly four out of five associated news stories also exaggerated. When press releases stuck to the original findings, only about one in six news stories exaggerated.20PubMed Central. The association between exaggeration in health-related science news and academic press releases: a replication study The exaggeration pipeline often starts before a journalist even touches the story.
Cognitive shortcuts also play a role. Clinicians and patients alike rely on mental rules of thumb to make fast decisions. These shortcuts can systematically distort how risk factors, disease labels, and treatment choices are recorded in medical systems, creating feedback loops where confounded associations get reinforced through clinical practice itself.21ScienceDirect. Cognitive Bias and the Creation and Translation of Evidence Into Clinical Practice A doctor who believes a risk factor is a cause may document and treat accordingly, and those records then become the observational data that future researchers analyze.
The Problem With Prediction Algorithms
Machine learning and artificial intelligence have added a new layer of confusion. Modern algorithms are extraordinarily good at identifying statistical patterns in large datasets. They can predict who will develop a disease with impressive accuracy. But prediction and causation are fundamentally different tasks. A prediction model can exploit any statistical regularity in the data, causal or not, to improve its accuracy. The parameters of a data-driven prediction model do not necessarily have a causal interpretation, which means using such a model to decide on interventions is risky.22Nature Machine Intelligence. Causal inference and counterfactual prediction in machine learning for actionable healthcare If an algorithm learns that low albumin predicts higher healthcare costs, and a health system tries to reduce costs by supplementing albumin, the intervention will fail if low albumin is just a marker rather than a cause. The push to integrate machine learning into clinical decision-making has made the distinction between risk factors and causes more urgent, not less.
Risk Factors in the Courtroom
The gap between risk factors and causes also shows up in legal proceedings, where the stakes are intensely personal. In toxic tort cases and product liability lawsuits, plaintiffs must prove that an exposure caused their illness, not merely that it raised their statistical risk. Courts have grappled with where to draw this line, and many have used a threshold of a relative risk greater than 2.0 as evidence of causation, on the logic that this implies the exposure more likely than not caused the plaintiff’s disease.23Epidemiology. The Sixteenth Conference of the International Society for Environmental Epidemiology (ISEE): Abstracts Epidemiologists have criticized this bright-line rule because it conflates population-level risk with individual causation. A relative risk of 1.5 means the exposure raises a person’s risk, but you cannot say with certainty that it caused any particular person’s disease. A relative risk of 3.0 means the exposure raises risk a lot, but the same uncertainty about individual cases remains. The legal system needs clean answers; biology rarely provides them.
The Multifactorial Reality of Peptic Ulcers
One of the most famous stories in modern medicine involves peptic ulcers and the bacterium Helicobacter pylori. Barry Marshall and Robin Warren shared a Nobel Prize for demonstrating that this bacterium plays a role in ulcer formation, overturning the long-held belief that ulcers were caused purely by stress and diet. That discovery transformed gastroenterology. But recent work has raised questions about whether the pendulum swung too far. Some researchers have argued that H. pylori’s contribution to ulceration may depend on pre-existing damage to the stomach lining from other processes, including long-term stress pathways, suggesting the bacterium’s role is secondary rather than primary in at least some cases.24Frontiers in Psychiatry. Reconceptualizing peptic ulcers as a psychosomatic disorder: a new etiological theory resolving a long-standing controversy surrounding Helicobacter pylori Whether you find that argument persuasive or not, it illustrates a broader truth: even when a factor is genuinely causal, its role in a multifactorial disease may be more conditional and context-dependent than a simple headline can convey.
Similarly, researchers have explored the epidemiological link between H. pylori infection and Alzheimer’s disease. Laboratory work has shown that bacterial peptides can activate genes associated with Alzheimer’s hallmarks, providing a plausible biological mechanism.25Scientific Reports. The hypothesis that Helicobacter pylori predisposes to Alzheimer’s disease is biologically plausible Biological plausibility is one of the Bradford Hill viewpoints, and it can strengthen an argument for causation, but plausibility alone is not proof. Many biologically plausible hypotheses have turned out to be wrong once tested in randomized trials. The history of medicine is littered with mechanisms that made perfect sense on a whiteboard but did not survive contact with human complexity.
How to Read Risk Factor News
When you see a headline claiming that some food, behavior, or exposure “increases risk” of a disease, a few questions can help you gauge whether you are looking at a real cause or just a statistical companion. Was the study observational or a randomized trial? Observational studies can identify risk factors but are always vulnerable to confounding and reverse causation. Has the finding been confirmed with Mendelian randomization or other designs that control for confounding more rigorously? Does the article describe a mechanism, or just an association? And did the news report use stronger language than the original paper? Given that about half of health news articles upgrade correlational findings to causal language, skepticism is warranted even when the underlying science is solid.20PubMed Central. The association between exaggeration in health-related science news and academic press releases: a replication study
The traditional risk factor approach has generated enormous insight into what makes people sick. But researchers have argued that the framework is fundamentally limited when the goal is to design interventions that reliably improve population health, and that a shift toward explicitly modeling causal relationships within complex systems, rather than simply cataloging statistical associations, is necessary for real progress.26Europe PMC / Wolters Kluwer. Commentary: The Limits of Risk Factors Revisited: Is It Time for a Causal Architecture Approach? Similarly, public health researchers have argued that framing causes as interventions, asking “what would happen if we changed this factor?” rather than “is this factor associated with the disease?”, produces more actionable answers.27PubMed Central. Causal inference in public health The question is not just whether a risk factor is associated with disease. The question is whether changing it would make a difference.