What Is Relative Risk in Epidemiology?

Relative risk is a measure that compares how likely an event is in one group versus another, typically an exposed group versus an unexposed group. If people who eat a certain food get sick twice as often as people who don’t, the relative risk is 2. The number itself is simple, but how it gets interpreted, misinterpreted, and applied in real-world decisions is where things get interesting, because a relative risk of 2 can mean very different things depending on how common the outcome was in the first place.

How the Number Works

Relative risk starts with two groups. One group has been exposed to something of interest, whether that’s a drug, a diet, a chemical, or a behavior. The other group has not. You measure how often each group experiences the outcome you care about and divide one rate by the other. The result tells you how much more (or less) likely the outcome is in the exposed group compared to the unexposed group.

A relative risk of 1 means both groups have the same likelihood of the outcome. Above 1 means the exposed group faces higher risk. Below 1 means the exposed group faces lower risk, which is what you want to see when testing a treatment or vaccine. A relative risk of 0.5 for a vaccine, for instance, means the vaccinated group got sick half as often as the unvaccinated group.

This measure comes most naturally from cohort studies, where researchers follow people forward in time from a point when none of them has the outcome yet. Because everyone starts outcome-free, the actual incidence rate in each group can be directly observed and compared.¹ In case-control studies, where you start with people who already have the outcome and look backward, you can’t calculate a true relative risk because you don’t know the actual incidence in the population. That’s where the odds ratio steps in as a substitute, a distinction explored further below.

Why the Baseline Matters More Than the Multiplier

The most common misunderstanding about relative risk is treating the number as if it tells you everything you need to know about danger. It doesn’t, because relative risk says nothing about how common the outcome is to begin with. A drug that doubles your risk of a side effect sounds alarming, but if the baseline risk is 1 in 100,000, doubling it gives you 2 in 100,000. That’s very different from doubling a risk that starts at 1 in 10.

This is the distinction between relative and absolute risk. Absolute risk is the raw probability of something happening. Relative risk is the ratio between two absolute risks. Both numbers are needed for a complete picture.² When you see a headline announcing that some exposure “triples the risk” of a disease, the follow-up question should always be: triples it from what? Tripling a risk of 0.01% gives you a final risk of 0.03%. Tripling a risk of 10% gives you 30%. The relative risk is identical in both scenarios. The practical significance is worlds apart.

Researchers and clinicians are trained to keep both numbers in view, but public communication frequently strips out the absolute numbers and leaves only the multiplier. That framing makes small risks sound terrifying and large risk reductions sound more impressive than they are.

The Ceiling Relative Risk Hits When Outcomes Are Common

There’s a mathematical constraint on relative risk that rarely gets discussed outside statistics classrooms but has real consequences for how findings are interpreted. Because a probability can never exceed 1 (or 100%), the higher the baseline risk of an outcome, the less room relative risk has to grow. When the baseline incidence is 10%, the maximum possible relative risk is 10, because even if every single exposed person gets the outcome, the incidence can only reach 100%. When the baseline is 50%, the ceiling drops to 2. When the baseline is 80%, the relative risk can’t exceed about 1.25.³

This matters because a relative risk of 1.2 for a common outcome might represent a genuinely strong effect that is being compressed by the math, while a relative risk of 5 for a rare outcome might represent a weaker absolute impact. Comparing relative risks across studies without accounting for the baseline rates in each study can lead to distorted conclusions about which exposures matter most. For common outcomes, rate ratios or other measures often provide a clearer picture of the actual effect size.

Relative Risk Versus the Odds Ratio

Odds ratios show up everywhere in medical research, and people routinely treat them as if they were the same thing as relative risk. They are not. An odds ratio compares the odds of an outcome in two groups, while relative risk compares the probabilities. Odds and probability are related but behave differently, especially when the outcome is common.

When the outcome is rare in the unexposed group, the odds ratio gives a decent approximation of the relative risk. Epidemiologists call this the “rare disease assumption.” But as the base rate of the outcome rises, the odds ratio inflates relative to the true relative risk. An odds ratio of 3 for an outcome that occurs in 40% of the unexposed group translates to a relative risk well below 3.⁴ The gap between the two measures widens as the outcome becomes more common, meaning that for everyday health events like pain, insomnia, or nausea, odds ratios can substantially overstate the size of an effect when interpreted as though they were relative risks.

Logistic regression, the workhorse statistical model in clinical research, produces odds ratios rather than relative risks. This is part of why odds ratios are so prevalent in the literature. Alternative approaches like modified Poisson regression can estimate relative risk directly while adjusting for other factors, and some methodologists have argued these alternatives should be used more often for common outcomes.⁵ When outcomes are common and logistic regression is used, methods exist to convert the resulting odds ratios back into relative risks, though the confidence intervals around those converted estimates require care.⁶

Relative Risk Reduction and Number Needed to Treat

When researchers evaluate a treatment, they often express its benefit as a relative risk reduction. If the control group has an event rate of 20% and the treatment group has an event rate of 15%, the absolute risk reduction is 5 percentage points. The relative risk reduction, on the other hand, is 25%, because 5 is one-quarter of 20. In formula terms, relative risk reduction equals 1 minus the relative risk.⁷

Relative risk reduction sounds more impressive than absolute risk reduction, and that asymmetry is routinely exploited. A drug that reduces heart attack risk from 2% to 1% has a relative risk reduction of 50%, but the absolute risk reduction is just 1 percentage point. The number needed to treat, which is the inverse of the absolute risk reduction, captures this more honestly: you would need to treat 100 people with that drug for one person to avoid a heart attack. That gives clinicians a tangible sense of how many patients they have to expose to a treatment’s side effects and costs for each patient who benefits.

A useful rule of thumb: whenever someone quotes a relative risk reduction, ask for the absolute numbers. A 30% relative risk reduction could mean treating 15 people to prevent one event, or treating 1,500 people to prevent one event, depending on the baseline rate.

What the Confidence Interval Tells You

A single relative risk number is a point estimate, a best guess from one study. The confidence interval around it tells you the range of values that are reasonably compatible with the data. If a study reports a relative risk of 1.8 with a 95% confidence interval from 1.2 to 2.7, you can be fairly confident the true relative risk is somewhere in that range. The key threshold is 1: if the interval includes 1, the data are consistent with there being no difference between the groups at all, and the finding is not considered statistically significant.

One subtlety worth knowing is that confidence intervals for relative risk live on a scale that is not symmetric. The distance from 1 to 2 (a doubling) is the same, in relative terms, as the distance from 1 to 0.5 (a halving). This means that on a standard number line, the interval looks lopsided, wider above the point estimate than below it. Researchers handle this by working with the logarithm of the relative risk, which restores symmetry.⁸ For the reader, the practical point is simpler: pay attention to the width of the interval. A relative risk of 3.0 with a confidence interval from 0.8 to 11 is much weaker evidence than a relative risk of 1.5 with a confidence interval from 1.3 to 1.7. The first is compatible with almost anything; the second is a tight, precise estimate.

From Individual Comparison to Population Impact

Relative risk tells you how much more likely an outcome is for an exposed individual compared to an unexposed one. But from a public health standpoint, the question isn’t just “how dangerous is this exposure?” It’s “how much disease in the population is actually caused by this exposure?” That’s where population attributable risk comes in.

Population attributable risk combines two pieces of information: the strength of the association (the relative risk) and how common the exposure is in the population. An exposure with a relative risk of 5 that only affects 0.1% of the population may contribute very little to the overall burden of disease, while an exposure with a relative risk of 1.3 that affects 60% of the population could be responsible for a large share of cases.⁹ This is why public health campaigns sometimes focus on modest risk factors like sedentary behavior or mildly elevated blood pressure. The relative risk is unspectacular, but because so many people are affected, eliminating or reducing the exposure could prevent more illness than targeting a stronger risk factor that only a few people have.

How the Smoking Research Made Relative Risk a Household Concept

Relative risk as a formal measure existed before the mid-twentieth century, but the concept became central to epidemiology through the landmark smoking studies. In 1950, Richard Doll and Austin Bradford Hill in England, along with Ernst Wynder and Evarts Graham in the United States, published case-control studies showing that tobacco use was a major risk factor for lung cancer.¹⁰ Critics pushed back, arguing that the association could be explained by confounders. The debate intensified in 1954 when cohort studies by both Doll and Hill and by Hammond and Horn showed that the risk of lung cancer was dramatically higher among smokers than among comparable non-smokers.

The British Doctors Study, which followed tens of thousands of physicians over decades, became one of the most influential cohort studies ever conducted and provided the first strong statistical proof of the association between smoking and a wide range of diseases, not just lung cancer.¹¹ The relative risks were large enough, and the data robust enough, that this body of work essentially established the template for how epidemiologists use relative risk to build a causal argument: consistent findings across multiple study designs, a dose-response relationship, biological plausibility, and a large magnitude of effect.

Relative Risk in the Courtroom

Outside of medicine and public health, relative risk plays a surprising role in law. In toxic tort litigation, where plaintiffs claim that exposure to a substance caused their illness, courts have sometimes used a relative risk threshold of 2 as a legal benchmark. The reasoning goes: if the relative risk associated with an exposure is greater than 2, then more than half of the cases among exposed people are attributable to the exposure, meaning it is “more probable than not” that the plaintiff’s disease was caused by it. That framing aligns with the civil legal standard of preponderance of evidence.

Whether that threshold makes scientific sense has been debated for years. Some legal scholars and epidemiologists have argued that a relative risk above 2 is not necessary for proving that a specific individual’s disease was caused by an exposure, because the threshold rests on biological assumptions that are rarely verified for any important disease. At the same time, in certain circumstances, a relative risk above 2 can be sufficient to establish causation at law.¹² The tension here reflects a deeper issue: relative risk is a population-level measure, and applying it to determine what happened to one specific person requires assumptions that population data alone cannot guarantee.

Pooling Relative Risks Across Studies

Meta-analyses routinely combine relative risks from multiple studies to arrive at a single pooled estimate. For most outcomes, this works well, but rare events create a technical headache. When the outcome is uncommon enough that some studies observe zero events in one or both groups, the relative risk becomes mathematically undefined because you can’t divide by zero. Standard methods handle this by either excluding those studies or adding a small correction factor, both of which introduce bias.¹³

This is a real problem for drug safety research, where the adverse events you most want to detect are exactly the ones that happen rarely. Dropping studies with zero events discards information. Adding correction factors introduces assumptions. Specialized meta-analytic models exist to handle these situations, but there’s no single approach that performs well in every scenario. For the reader encountering a meta-analysis of rare adverse events, the practical takeaway is that the pooled relative risk may be less precise than it looks, especially when the total number of events across all studies is small.

When Two Risk Factors Combine

Relative risk gets more complicated when two exposures act together. If smoking has a relative risk of 3 for some disease and asbestos exposure has a relative risk of 5, what is the relative risk for someone exposed to both? If the effects multiply, you’d expect a combined relative risk around 15. If they simply add, you’d expect something closer to 7. The actual answer depends on how the biological mechanisms interact, and researchers test for this by comparing the observed combined risk against what different mathematical models predict.

At low baseline prevalence, synergistic risk factors tend to behave multiplicatively, producing a combined relative risk larger than the sum of the individual effects would predict. When the factors operate through independent pathways, the combined effect often looks additive instead.¹⁴ The distinction between additive and multiplicative interaction isn’t just academic. Public health decisions about where to focus prevention efforts depend on which model applies, because the two models lead to different conclusions about which subgroups benefit most from intervention. If two factors interact super-additively, targeting the combination prevents far more disease than you’d estimate from the individual relative risks alone.

The Hazard Ratio and How It Differs

You’ll often see hazard ratios reported alongside or instead of relative risks, especially in studies that follow people over long periods. A hazard ratio looks similar to a relative risk and is interpreted in roughly the same way: above 1 means higher risk, below 1 means lower risk. But the two measures answer slightly different questions. Relative risk compares the total proportion of people who experience an outcome by a fixed time point. The hazard ratio compares the instantaneous rate of the outcome at any given moment during follow-up, accounting for the fact that people drop out or are followed for different lengths of time.¹⁵

In practice, when the outcome is relatively rare and follow-up time is complete for most participants, the hazard ratio and the relative risk will be close to each other. They diverge when outcomes are common or when there is substantial loss to follow-up. If a clinical trial reports a hazard ratio of 0.7, it means that at any given point during the study, treated patients were experiencing the event at 70% the rate of untreated patients. That’s not quite the same as saying 30% fewer treated patients experienced the event overall, though it’s frequently interpreted that way. The difference is subtle enough that for most practical purposes the two numbers convey similar messages, but it’s worth knowing they aren’t identical, especially when comparing findings across studies that use different measures.