Relative risk reduction is the proportional drop in the chance of an outcome in a treatment group compared with a control group, expressed as a fraction or percentage of the control risk: if a control group has a 4 percent event rate and treatment cuts that to 2 percent, the relative risk reduction is 50 percent because half of the baseline risk has been removed. It is the headline number that makes interventions sound dramatic, but it deliberately hides how much risk there was to begin with, which is why the same 50 percent reduction can describe a life-changing benefit or a vanishingly small one. The honest companion figure is the absolute risk reduction, the raw difference in event rates, and understanding how the two diverge is the difference between a result that informs a decision and a result that merely impresses.
How each figure is calculated from the event rates
Both measures start from the same two numbers: the control event rate, the proportion of the control group that has the outcome, and the treatment event rate, the proportion of the treatment group that has it. Call the control rate the baseline risk and the treatment rate the residual risk. From these two probabilities the two reductions are built in two different ways, and that difference in construction is exactly why one number can look enormous while the other looks trivial.
- The absolute risk reduction is the control event rate minus the treatment event rate. It is a plain subtraction, measured in the same units as the risks themselves, such as events per hundred people.
- The relative risk reduction is the absolute risk reduction divided by the control event rate. Equivalently, it is one minus the risk ratio, because the risk ratio is the treatment rate divided by the control rate.
- The risk ratio itself, also called relative risk, is the treatment event rate divided by the control event rate, and it underpins the relative figure.
The structural contrast is simple. The absolute figure keeps the baseline in plain sight, because it is an unscaled difference. The relative figure throws the baseline away by dividing by it, so it reports only the proportion removed and says nothing about how large the starting risk was. A proportion of a tiny number is still tiny, yet the proportion alone reads as impressive. That is the whole problem in one sentence, and the worked example below makes it concrete.
A worked example where a big relative drop is a tiny absolute one
Suppose a preventive drug is tested against placebo. In the placebo arm the event happens to 4 people in every 1000, so the control event rate is 0.004. In the treatment arm it happens to 2 people in every 1000, so the treatment event rate is 0.002. The absolute risk reduction is 0.004 minus 0.002, which equals 0.002, meaning two fewer events per thousand people treated. The relative risk reduction is 0.002 divided by 0.004, which equals 0.50, or 50 percent.
Both numbers are correct, and they describe the identical data, yet they tell very different stories. A press release can truthfully announce that the drug "halves the risk," which sounds like a major breakthrough. The clinical reality is that you must treat 500 people to prevent a single event, because the baseline risk was small to start with. Now imagine a second setting where the control event rate is 0.40 and treatment brings it to 0.20. The relative risk reduction is again 50 percent, an identical headline, but the absolute risk reduction is now 0.20, twenty fewer events per hundred people, a genuinely large effect. The relative figure cannot distinguish these two worlds; only the absolute figure can. If you want to see how these counts behave once they are pooled across several studies, you can enter event data directly into our tool for combining study event counts and read both measures side by side.
Why the absolute figure is the honest one for decisions
A decision, whether by a clinician, a guideline panel, or a patient, is always about a real person with a real baseline risk. What matters is how much that individual risk actually falls, not what fraction of an unknown starting point is removed. The absolute risk reduction answers the decision question because it is anchored to the population's real event rate, while the relative figure floats free of it.
This is why reporting only the relative figure is treated as a form of spin in the methodological literature, and why critical-appraisal checklists ask reviewers to confirm that absolute effects are reported. The same 50 percent reduction can be a public-health triumph or a clinically meaningless change depending entirely on the baseline, and a reader given only the relative number has no way to tell which. Reporting the absolute difference alongside the relative one restores that missing context and lets the reader judge whether the benefit is worth the cost, the burden, and the harms of the intervention.
The link to number needed to treat
The absolute risk reduction has a direct and clinically powerful translation: its reciprocal is the number needed to treat, the count of people who must receive the intervention to prevent one additional event. In the small-baseline example above, the absolute risk reduction was 0.002, so the number needed to treat is one divided by 0.002, which equals 500. In the large-baseline example the absolute risk reduction was 0.20, so the number needed to treat is one divided by 0.20, which equals 5.
The contrast is stark. Treating 5 people to prevent one event is a strong, tangible benefit; treating 500 people to prevent one event is a far weaker proposition, even though the relative risk reduction is identical at 50 percent in both cases. This reciprocal relationship is the reason the absolute figure is the one that drives bedside and policy decisions, and it is why a small absolute risk reduction always produces a large, sobering number needed to treat. You can compute this directly with our tool for turning absolute differences into a treatment count or read the longer explainer on how the number needed to treat is derived and interpreted.
Why the relative figure cannot give a treatment count
Notice that you cannot compute a number needed to treat from a relative risk reduction alone, because the relative figure has discarded the baseline. You must reattach the control event rate first, recover the absolute difference, and only then take the reciprocal. This is one more reason that a relative figure presented without its baseline is incomplete: it is mathematically incapable of answering the question clinicians most want answered.
How this matters when reporting pooled results
Meta-analysis adds a further wrinkle. Pooling is almost always done on a ratio measure, usually the risk ratio or the odds ratio, because ratios are more stable across studies with different baseline risks and because they are combined cleanly on the natural logarithm scale. That means the pooled estimate that comes out of the analysis is a relative measure by construction, and the relative risk reduction follows directly as one minus the pooled risk ratio. The absolute effect, however, depends on the baseline risk, which differs from one population to another.
- Report the relative pooled estimate with its confidence interval, because that is what the synthesis actually estimates and what transfers across settings.
- Translate to an absolute effect by applying the pooled risk ratio to a chosen baseline risk, ideally one representative of the population the reader cares about, so the absolute risk reduction and the number needed to treat can be quoted honestly.
- State the assumed baseline explicitly, because the same pooled relative effect yields different absolute benefits at different baselines, exactly as the worked example showed.
- Show both figures on the summary of findings, since GRADE evidence profiles expect a relative effect and an anticipated absolute effect side by side.
The practical discipline is to never let the relative number stand alone in the abstract or the plain-language summary. Pair it with an absolute effect and a number needed to treat keyed to a stated baseline, so a reader cannot mistake a proportional change for a guarantee of a large benefit. If you are building the underlying synthesis, the comparison of how the risk ratio is computed from raw counts clarifies where the relative figure comes from, the discussion of how the odds ratio and risk ratio differ explains which ratio you should be pooling, and the overview of how a meta-analysis combines evidence across studies sets the wider context. Once the counts are extracted, you can turn them into a published figure with the forest plot builder on the home page, which shows each study's effect and the pooled result together.