What is the number needed to treat?

The number needed to treat is the number of patients who must receive a treatment instead of the comparator for one extra person to benefit. It is the reciprocal of the absolute risk reduction, so a treatment that cuts an event rate by 10 percentage points has a number needed to treat of 10.

How do you calculate the number needed to treat?

Compute the event rate in each arm, subtract the treatment rate from the control rate to get the absolute risk reduction, then take one divided by that difference and round up to a whole person. A control rate of 30 percent and treatment rate of 20 percent gives a number needed to treat of 10.

What is the number needed to harm?

The number needed to harm is how many patients must be treated for one extra person to experience an adverse event. It is calculated the same way as the number needed to treat but from the absolute risk increase in harms, and reporting both lets readers weigh benefit against harm.

Why does the number needed to treat depend on baseline risk?

Because it is built on the absolute risk reduction, the same relative effect produces a different number needed to treat in high-risk versus low-risk populations. A value from one trial does not transfer cleanly elsewhere, so always report the baseline event rate and time horizon it assumes.

What is a good number needed to treat?

There is no universal threshold, because a good number needed to treat depends on the seriousness of the outcome, the cost, and the harms of the treatment. A number needed to treat of 10 means ten people are treated for one to benefit, which is excellent for a cheap, safe preventive but poor for an expensive or toxic therapy, so the value is only meaningful next to the matching number needed to harm and the outcome it prevents.

How do you report a confidence interval for the number needed to treat?

Calculate the confidence interval of the absolute risk reduction first, then invert each limit to get the interval for the number needed to treat. When the absolute risk reduction is not significant and its interval spans zero, the number needed to treat interval runs from a number needed to treat for benefit, through infinity, to a number needed to treat for harm, following Altman's convention, and should be reported that way rather than as a single point estimate.

Can you calculate a number needed to treat from a meta-analysis?

Yes. The usual approach is to pool a relative measure such as the risk ratio, choose one or more plausible baseline risks for the target population, apply the pooled ratio to each baseline to get an absolute risk reduction, then take the reciprocal. Because the same relative effect gives very different numbers needed to treat across baselines, reporting it at a low, typical, and high baseline risk is more honest than a single average value.

Number Needed to Treat: How to Calculate It

The number needed to treat is the number of patients who must receive a treatment, instead of the comparator, for one extra person to benefit. It is calculated as the reciprocal of the absolute risk reduction, so a treatment that cuts an event rate from 20 percent to 10 percent has a number needed to treat of 1 divided by 0.10, which is 10. Because it speaks in whole people rather than ratios, it is one of the most intuitive ways to express how much an intervention actually helps.

Why an absolute measure changes the conversation

Ratios such as the risk ratio and odds ratio describe relative change but hide the baseline. A treatment that “halves the risk” sounds dramatic whether the risk falls from 2 percent to 1 percent or from 50 percent to 25 percent, yet the real-world value differs enormously. The number needed to treat restores that context by building on the absolute risk reduction, which is why clinicians and guideline panels lean on it. The contrast with relative measures is drawn out in odds ratio versus risk ratio.

How to calculate the number needed to treat

First compute the event rate in each arm, then the absolute risk reduction as the control rate minus the treatment rate. The number needed to treat is one divided by that difference, rounded up to a whole person. If the control event rate is 30 percent and the treatment rate is 20 percent, the absolute risk reduction is 10 percentage points and the number needed to treat is 10. The number needed to treat calculator does this from raw event counts and returns a confidence interval too.

A worked example with the confidence interval

A clean numeric example fixes the method. In a trial, 30 of 200 control patients have the event (a control rate of 0.15) and 18 of 200 treated patients do (a treatment rate of 0.09). The absolute risk reduction is 0.15 minus 0.09, which is 0.06, so the number needed to treat is 1 divided by 0.06, which rounds up to 17. The interval around it is derived by taking the confidence interval of the absolute risk reduction and inverting its limits: if that interval runs from 0.01 to 0.11, the number needed to treat runs from 1 divided by 0.11 (about 9) to 1 divided by 0.01 (about 100). Reporting “17, 95 percent confidence interval 9 to 100” is honest; reporting a bare 17 hides how much uncertainty surrounds it.

When the confidence interval crosses zero benefit

A notorious quirk arises when the absolute risk reduction is not statistically significant, so its confidence interval spans zero. Inverting a limit that passes through zero sends the number needed to treat to infinity and out the other side into a number needed to harm. The correct expression, following the convention of Altman, is a number needed to treat for benefit at one end, through infinity, to a number needed to treat for harm at the other. Quoting only the tidy point estimate in this situation, as if the effect were certain, is one of the most common reporting errors and is why the absolute measure should always travel with its interval.

Number needed to harm

The mirror image is the number needed to harm: how many patients must be treated for one extra person to experience an adverse event. It is computed the same way, from the absolute risk increase in harms. Reporting both lets a reader weigh benefit against harm directly, which a single effect size cannot do on its own.

Number needed to treat in a meta-analysis

A meta-analysis usually pools a relative measure, then converts the pooled risk ratio to a number needed to treat at a chosen baseline risk. This matters because the same relative effect implies a very different number needed to treat for a high-risk versus a low-risk population. Reporting it at one or more plausible baseline risks, rather than a single average, is the honest approach and pairs well with the certainty of evidence rating that accompanies the absolute effect.

Converting a pooled risk ratio to a number needed to treat

The mechanics are worth spelling out. Pick a plausible baseline risk for your population, call it the control event rate. Multiply it by the pooled risk ratio to get the predicted treated risk, subtract that from the baseline to get the absolute risk reduction, then invert. With a pooled risk ratio of 0.75 and a baseline risk of 0.20, the treated risk is 0.15, the absolute risk reduction is 0.05, and the number needed to treat is 20. Drop the same relative effect onto a low-risk population with a baseline of 0.04, and the absolute risk reduction falls to 0.01, pushing the number needed to treat to 100. One pooled ratio, two wildly different clinical pictures, which is precisely why a single average figure can mislead. The contrast between relative and absolute thinking is the same one drawn out in choosing between odds and risk ratios.

Watch the baseline risk

Because the number needed to treat depends on baseline risk, a value lifted from one trial does not transfer cleanly to a different population. When you read or report one, always note the baseline event rate it assumes. This is the same care that interpreting a forest plot demands, where the pooled relative effect is only half the story. A robust review reports the number needed to treat at a low, typical, and high baseline risk so a reader can locate their own patients, and confirms the figure survives a robustness check on the underlying pool.

Common mistakes when using the number needed to treat

Several errors recur. The first is computing it from an odds ratio as if it were a risk ratio; an odds ratio overstates the effect for common outcomes, so deriving the absolute risk reduction from it inflates the benefit. The second is quoting a number needed to treat without its time horizon: a value of 20 over five years is a very different proposition from 20 over five weeks. The third is comparing numbers needed to treat across different baseline risks or different outcomes, which is meaningless because the denominator changes. The fourth is presenting a non-significant result as a clean point estimate, ignoring that its interval spans benefit and harm. Avoiding all four comes down to one habit: report the absolute measure with its interval, its baseline, and its time frame, never on its own.

How to report it

State the number needed to treat with its confidence interval, name the baseline risk and time horizon it applies to, and present it next to the relative effect rather than instead of it. A confidence interval that crosses from benefit into harm should be reported as such rather than collapsed to a tidy point estimate. Combined with a clear risk-of-bias judgement, an absolute measure like this turns a statistical result into something a clinician and patient can actually use.

Number Needed to Treat: Calculation and Interpretation