"Fifty percent higher risk" is the most misleading sentence in health reporting, and it is usually true. If an outcome affects 2 people in 10,000 and something raises it to 3 in 10,000, that is a 50 percent relative increase and a 0.01 percentage point absolute increase. Both numbers are correct. One of them tells you what would happen to a room full of people and the other does not. This absolute risk reduction calculator computes both from the same event counts and puts them next to each other, because seeing them apart is how the misreading happens.
Arb Digital publishes this as a statistics tool, not a medical one. It takes four numbers you supply — events and totals in two groups — and returns the arithmetic: each group's event rate, the absolute difference between them, the relative difference, and the number needed to treat. It has no opinion about any drug, exposure or intervention, and it uses no real trial as an example anywhere on this page.
What This Absolute Risk Reduction Calculator Does
It computes six related quantities from two groups. The control event rate is events divided by group size in the comparison group. The experimental event rate is the same in the treated group. The absolute risk reduction is the first minus the second, expressed in percentage points. The relative risk is the second divided by the first, and the relative risk reduction is one minus that, expressed as a percentage. The number needed to treat is one divided by the absolute risk reduction, rounded up to a whole person.
It also restates all of it as natural frequencies — "so many in 10,000" rather than "0.3 percent" — because that format is consistently easier to reason about. If the treated group does worse rather than better, the tool switches its own language: the absolute change becomes a risk increase and the count becomes the number needed to harm.
Its sibling tool, our relative risk calculator, takes the same inputs and concentrates on the ratio measures — relative risk and relative risk reduction with the interpretation those need. The split is deliberate: that page owns the ratio, this page owns the difference and the count. Both report the other measure so neither can be read in isolation. If you want plain proportion arithmetic without the epidemiological framing, our percentage change calculator and percentage difference calculator do that.
How to Use It
- Enter events and totals for the control group. The total is everyone in the group, including those who had the event, not just those who did not.
- Do the same for the treated or exposed group. The two group sizes do not have to match, because everything is computed as a rate.
- Choose a natural-frequency base. Pick one large enough that both rates come out as whole numbers of people rather than fractions.
- Type in the follow-up period. It only changes the wording, but a risk without a time window is not a risk. Five years and five weeks are entirely different claims.
- Read the hero and the grid together. The headline is absolute; the relative figure sits in the grid alongside it, deliberately never on its own.
The Formulas and How They're Calculated
Write CER for the control event rate and EER for the experimental event rate. Then:
CER = control events ÷ control total. EER = treated events ÷ treated total. Absolute risk reduction (ARR) = CER − EER. Relative risk (RR) = EER ÷ CER. Relative risk reduction (RRR) = 1 − RR, which is also ARR ÷ CER. Number needed to treat (NNT) = 1 ÷ ARR, conventionally rounded up.
A peer-reviewed guide to these measures, Understanding number needed to treat (NNT): A practical guide for anaesthesia and critical care clinicians, sets out the same definitions: absolute risk reduction as the control rate minus the treated rate, and number needed to treat as its reciprocal, answering how many people must be treated for one additional person to get the desired outcome.
Work the defaults through. Thirty events in 10,000 control participants gives CER = 0.003, or 0.3 percent, or 30 in 10,000. Twenty events in 10,000 treated participants gives EER = 0.002, or 0.2 percent, or 20 in 10,000. ARR = 0.003 − 0.002 = 0.001, which is 0.1 percentage points. RR = 0.002 ÷ 0.003 = 0.667, so RRR = 1 − 0.667 = 33.3 percent. NNT = 1 ÷ 0.001 = 1,000.
Those two sentences describe the same result: "a third lower risk" and "1,000 people treated for one event avoided". Neither is wrong. Only one of them lets you picture what happened.
Why the Relative Figure Dominates Reporting
Relative measures have one genuine advantage: they are often more stable across populations with different baseline risks, which makes them useful for combining studies. They also have one enormous disadvantage: they are unbounded above and completely detached from how common the outcome is. A relative risk reduction of 50 percent is the same number whether the baseline was 40 in 100 or 4 in 100,000, and the two situations are nothing alike.
This is not a hypothetical concern in the literature. A commentary in the peer-reviewed literature, Understanding and Communicating Risk: Assessing Both Relative and Absolute Risk Is Absolutely Necessary, argues that one cannot fully understand risk without reporting both, and shows that identical odds ratios produce dramatically different relative risks depending on how common the outcome is. Reporting one without the other is incomplete rather than merely imprecise.
The practical test is simple. Whenever you meet a relative figure, ask "out of how many?" until you get an absolute number with a time window attached. If the answer is not available, the claim cannot be evaluated, no matter how large the percentage sounds.
What the Number Needed to Treat Actually Counts
NNT is the reciprocal of the absolute risk reduction, so it inherits everything that figure depends on. An NNT of 1,000 does not mean the intervention works in one person out of a thousand and does nothing in the rest — that is a common misreading. It means that across a thousand people treated for the study's follow-up period, there was one fewer event than there would have been otherwise.
Three things change NNT without anything about the intervention changing at all. Baseline risk: the same relative effect in a higher-risk group produces a bigger absolute difference and therefore a smaller NNT. Follow-up length: events accumulate over time, so a five-year NNT is smaller than a one-year NNT for the same intervention. And the outcome chosen: NNT for a mild outcome and NNT for a severe one are not comparable quantities even in the same study.
This is why comparing NNTs across studies is unreliable unless the baseline risk, the time horizon and the outcome definition all match. It is also why an NNT quoted without a time period should be treated as incomplete information.
When the Difference Goes the Other Way
If the treated group has more events than the control group, the absolute risk reduction is negative. The tool relabels it: the magnitude becomes an absolute risk increase and its reciprocal becomes the number needed to harm, meaning how many people would be exposed for one additional event to occur.
The arithmetic is identical and the interpretation is a mirror image, but the reporting asymmetry is worth naming. Benefits tend to reach the public as relative figures, which makes them sound large, while harms more often appear as absolute figures, which makes them sound small. Computing both for both directions is the only way to compare them on the same footing, and that is the whole reason this page shows four measures at once rather than one.
One boundary the tool does not cross: it produces point estimates only, with no confidence interval. Two studies can produce the same ARR with completely different precision depending on their size. If you need an interval around a proportion, our confidence interval calculator handles that, and the p-value calculator covers significance testing. Neither of those substitutes for the other.
Sanity Checks Worth Running
Three checks catch most input errors. First, both event counts must be no larger than their group totals; a rate above 100 percent means the totals and events have been swapped. Second, natural frequencies should come out as sensible whole numbers — if 10,000 gives you 0.4 people, move to 100,000. Third, ARR multiplied by the base should equal the difference between the two natural frequencies, which is an easy arithmetic check you can do in your head.
A fourth check is conceptual rather than numerical. Ask whether the two groups are actually comparable. All of these measures assume the only meaningful difference between the groups is the thing being studied. In a randomised trial that assumption is designed in; in observational data it usually is not, and the same arithmetic then describes an association rather than an effect. The formulas cannot tell the difference, and neither can this page. For conditional probability arithmetic more generally, our conditional probability calculator and odds probability converter cover the neighbouring ground.
Arb Digital builds calculators and content that show their working and refuse to flatter the numbers.
Browse All Free Tools Talk To Our TeamCommon Mistakes to Avoid
- Quoting a relative figure alone — without the baseline rate, a percentage change says nothing about how many people are affected.
- Dropping the time window — a risk figure without a follow-up period is not interpretable, and NNT shrinks as follow-up lengthens.
- Reading NNT as "it only works in one person" — it counts events avoided across a group, not responders within it.
- Comparing NNTs from different studies — baseline risk, follow-up and outcome definition all have to match for the comparison to mean anything.
- Treating an association as an effect — the formulas run identically on observational data, and cannot tell you whether the groups were comparable.
Related Free Tools From Arb Digital
Work with ratio measures using the relative risk calculator, put an interval around a proportion with the confidence interval calculator, test significance with the p-value calculator, handle conditional probabilities with the conditional probability calculator, convert between odds and probabilities with the odds probability converter, or run plain proportion arithmetic with the percentage change calculator. The free online tools hub lists every statistics tool we publish.
Frequently Asked Questions
It is the control group's event rate minus the treated group's event rate, expressed in percentage points. If 0.3 percent of one group had an event and 0.2 percent of the other did, the absolute risk reduction is 0.1 percentage points.
Absolute risk reduction is the difference between the two rates; relative risk reduction is that difference divided by the control rate. A drop from 0.3 to 0.2 percent is 0.1 percentage points absolute and 33 percent relative, from the same data.
Divide one by the absolute risk reduction expressed as a proportion, then round up. An absolute risk reduction of 0.001 gives a number needed to treat of 1,000.
Because relative measures ignore how common the outcome is. Halving a risk of 2 in 10,000 removes 1 case per 10,000 people, which is a 50 percent relative change and a 0.01 percentage point absolute change.
That across 1,000 people treated over the study's follow-up period, there was one fewer event than expected. It does not mean the intervention worked in one person and failed in the other 999.
The absolute risk reduction is negative, and it is then reported as an absolute risk increase. Its reciprocal is the number needed to harm: how many people would be exposed for one additional event to occur.
Events accumulate over time, so the same intervention shows a larger absolute difference and a smaller number needed to treat over five years than over one. A risk figure without its time window cannot be interpreted.
No. It returns point estimates only. Two studies can produce the same absolute risk reduction with very different precision, so an interval requires a separate calculation.
This page performs statistical arithmetic on numbers you enter, for general educational purposes. It is not medical advice, it makes no claim about any treatment, exposure or outcome, and decisions about care are for a qualified clinician who knows the individual and the underlying evidence.
