Relative Risk vs Absolute Risk: How To Read Medical Statistics And Make Better Health Decisions (2026 Guide)

relative risk vs absolute risk
Situation Figures Planning use
Postoperative nausea example 10% vs 20%; RR 0.5, described as a 50% reduction in risk A relative figure on its own. Ask for the underlying rates before deciding whether the change matters to you.
Chronic postsurgical pain after a nerve block Falls from 15% to 5%; absolute risk reduction of 10 percentage points; NNT 10 Shows the real-world gap and how many people must be treated for one extra benefit.
Nerve injury after the same nerve block Rises from 1% to 5%; absolute risk increase of 4%; number needed to harm 25 Weigh harms in the same absolute terms as benefits.
Smokers vs nonsmokers, perioperative respiratory complications 8% vs 2%; attributable risk of 6 percentage points Shows how much of the absolute risk is tied to an exposure.
95% confidence interval around a relative risk A narrower interval means a more precise estimate Check the interval before trusting the headline number.

These examples and definitions come from OpenAnesthesia’s explainer on relative risk, odds ratios, absolute risk reduction and attributable risk, while the point about which differences matter most to people comes from Testing Treatments.

Relative risk vs absolute risk is the difference between a ratio of two groups’ event rates and the actual gap between them. For personal decisions, absolute figures usually matter more, provided you also know the baseline risk.

Reading the two measures correctly is one part of knowing how to tell if a health study is reliable. The sections below start with what the choice of measure changes for a decision, then show how each figure is built.

What the choice of measure changes

The same result can sound dramatic or modest depending on which number is quoted. In the nausea example, rates of 10% and 20% give an RR of 0.5, which is a 50% reduction in risk, according to OpenAnesthesia. The ratio tells you the proportional change but not how common the event was to begin with.

  • Benefit: a relative figure shows proportion only. The absolute figures show how many more people actually avoid the outcome.
  • Rare events: in populations with low event rates, relative risk may overstate the clinical significance of findings, so a large-looking ratio on a rare outcome deserves a second look.
  • Personal weight: Testing Treatments states that absolute differences between the effects of two treatments matter more to most people than relative differences, and it lists relative measures of effects as something that can be misleading.
  • Harms: apply the same test to side effects. In the nerve block example, nerve injury rising from 1% to 5% is an absolute increase of 4% and a number needed to harm of 25, meaning one extra injury for every 25 patients.

A treatment that clearly helps on one outcome can still be a poor trade if the harm figure is similar in size, so put benefit and harm side by side in absolute terms.

Supplement and food headlines

Headlines about a food, a vitamin or a supplement typically report a relative change, such as risk being cut by a given proportion. That is the same kind of figure as the 50% in the nausea example, and it carries the same limits: it says nothing about how many people had the outcome in the first place. Testing Treatments advises that researchers, clinicians and journalists who make claims should report absolute as well as relative differences.

A magnified headline word carries emotional weight but no rate, which is why the numbers behind a health headline need to be found separately. Stock illustration only.

When a diet or supplement story gives only a relative figure, use a short checklist before changing what you eat or buy:

  • Find the baseline. Ask what share of people in the comparison group had the outcome. Without it, a relative change cannot be turned into a real difference.
  • Look for the absolute gap. The difference between the two groups’ rates is the number that describes the practical effect.
  • Convert to a count. An NNT or number needed to harm tells you how many people would need to take a product for one person to be affected.
  • Check the interval. A claim that reports no confidence interval gives you no way to judge precision, and Testing Treatments lists reporting confidence intervals as a key concept.
  • Match the study to you. Results from a population with a different baseline may not translate to your own circumstances.

If a supplement or food story offers none of these, treat the headline as a prompt to look up the study, not as a reason to act. Pooled evidence is a stronger starting point than a single result, and Systematic Review Vs Meta-Analysis: How They Differ And When To Use Each (2026 Guide) – Testpledge explains how that evidence is combined.

How the figures are built

Relative risk and baseline risk

Relative risk compares the probability of an event in the treatment group with the probability in the control group, so a result below 1 means fewer events in the treated group. Baseline risk is the starting level that the ratio is applied to. OpenAnesthesia notes that absolute risk reduction assumes the baseline risk does not change between groups, which is why a result from one population should not be carried over automatically to people with a very different starting risk.

A pie chart and line graph display proportions and trends, but like a relative risk they show shares rather than the size of the group behind them. Stock illustration of a general statistics display.

Reading the nerve block example

Chronic postsurgical pain falling from 15% to 5% is a clear absolute improvement, and the same source gives an NNT of 10, meaning ten patients must receive the intervention to prevent one additional case. The same nerve block shows the harm side: nerve injury rises from 1% to 5%. Reading both together gives a more honest picture than either alone.

For the attributable risk example, an 8% complication rate in smokers against 2% in nonsmokers gives an attributable risk of 6 percentage points. That subtraction describes the exposure’s absolute contribution to the outcome in that example.

Precision and sample size

A 95% confidence interval around a relative risk reflects the precision of the estimate, and a narrow interval means a more precise one. Absolute figures are also sensitive to study size: smaller studies may produce wide variability in absolute risk reduction estimates. Treat a striking number from a small study as provisional. How the trial was run, including blinding, affects how far to trust it, as covered in Single-Blind Vs Double-Blind Studies: How To Tell Which Design Gives More Reliable Results In 2026 – Testpledge.

Absolute risk for an individual

Absolute risk can also be projected for a person from their characteristics, not just compared between trial groups. The BCRA package for R, for example, includes a function described as estimating absolute risks of developing breast cancer, based on the Gail model, as listed on RDocumentation. Treat any such estimate as a starting point for a conversation with a clinician rather than a verdict, because it depends on the inputs and model behind it.

Frequently asked questions

Why do headlines use relative risk?

Relative figures are easy to state in a short line, but they can mislead. Testing Treatments advises that anyone reporting claims should give absolute as well as relative differences, and that confidence intervals should be reported.

How do I calculate absolute risk reduction?

Subtract the risk in the treatment group from the risk in the control group. That gives the absolute difference in event rates between the groups, usually expressed in percentage points.

What is NNT?

The number needed to treat indicates how many patients must receive an intervention to prevent one additional adverse event. It equals 1 divided by the absolute risk reduction, so a larger reduction means a smaller NNT.

What does a confidence interval that crosses 1 mean for a relative risk?

It indicates no statistically significant difference between the groups. The estimate may still look impressive as a single number, but the interval shows that the data are compatible with no effect.