Number needed to treat (NNT) tells you how many people need to receive a treatment for one additional person to benefit. The formula is simple: NNT = 1 / ARR, where ARR is the absolute risk reduction. The real skill is knowing how to get the ARR correctly and how to interpret the final number, so let’s walk through the full process.
The Core Formula
Every NNT calculation starts with two numbers from a clinical trial: the event rate in the control group (CER) and the event rate in the treatment group (EER). The event rate is just the proportion of people who experienced the outcome. If 30 out of 200 people in the control group had a heart attack, the CER is 30/200 = 0.15, or 15%.
From there, you calculate the absolute risk reduction:
ARR = CER − EER
Then take the inverse:
NNT = 1 / ARR
That’s it. The NNT is always a whole number (rounded up), because you can’t treat half a person. If your calculation gives you 33.3, the NNT is 34.
A Worked Example
Suppose a trial tests a new blood pressure drug. In the control group, 12% of patients have a stroke over five years. In the treatment group, 8% do.
- CER: 0.12
- EER: 0.08
- ARR: 0.12 − 0.08 = 0.04
- NNT: 1 / 0.04 = 25
This means you’d need to treat 25 people with the drug for five years to prevent one additional stroke. Notice the time frame: five years. That detail matters, and we’ll come back to it.
Why You Must Use Absolute Risk, Not Relative Risk
The single most common mistake in calculating NNT is plugging in the relative risk reduction instead of the absolute risk reduction. These two numbers can look dramatically different.
Take a drug that cuts the risk of a rare event from 0.2% to 0.1%. The relative risk reduction is 50%, which sounds impressive. But the absolute risk reduction is only 0.1%, giving an NNT of 1,000. You’d need to treat a thousand people for one to benefit. If you mistakenly used the 50% figure, you’d calculate an NNT of 2, which would be wildly wrong.
Relative risk reduction stays constant regardless of baseline risk, but NNT changes with it. A treatment that halves your risk of something means very different things depending on whether your starting risk is 40% or 0.4%. Always use the absolute numbers.
NNT Depends on Time
An NNT is meaningless without a time frame attached. A drug with an NNT of 50 over one year is very different from one with an NNT of 50 over ten years. Longer trials generally produce lower (better) NNTs because more events accumulate in the control group, widening the gap between treated and untreated patients.
When trials report time-to-event outcomes with varying follow-up periods, the calculation gets more complicated. Researchers typically estimate survival probabilities at specific time points using statistical models, then derive an NNT for each point. The key takeaway for you: always check the follow-up duration before comparing NNTs from different studies. A five-year NNT and a one-year NNT for different drugs aren’t comparable without adjustment.
What Counts as a “Good” NNT
Lower is better. An NNT of 1 would mean every single person treated benefits, which is rare outside of certain acute treatments. An NNT of 100 means you’re treating 99 people who won’t benefit to help 1 who will. Whether that tradeoff is worth it depends on what you’re preventing and what the treatment costs in terms of side effects, money, and inconvenience.
Real-world NNTs span a huge range. Statins for primary prevention of cardiovascular disease illustrate this well. A population-based cohort study found that the five-year NNT varied enormously based on a patient’s baseline risk: 470 for people in the lowest risk category (under 5% risk), dropping to 204 for moderate risk, 75 for higher risk, and 62 for the highest risk group (10–19.9% risk). The same drug, but wildly different NNTs depending on who’s taking it. This is why baseline risk is so central to the calculation and to clinical decision-making.
For context, treatments considered highly effective often have NNTs in the single digits or low teens. Preventive treatments for common conditions typically land in the 20–100 range. NNTs in the hundreds aren’t necessarily useless, but they need to be weighed against the severity of the outcome being prevented and the burden of treatment.
Confidence Intervals for NNT
Like any statistic derived from a study, the NNT has uncertainty around it. You can calculate a confidence interval by first finding the 95% confidence interval for the ARR, then taking the reciprocal of each boundary and reversing their order. If the ARR’s confidence interval runs from 0.02 to 0.06, the NNT confidence interval runs from 1/0.06 to 1/0.02, or roughly 17 to 50.
If the confidence interval for the ARR crosses zero (meaning the treatment might not work at all), the NNT confidence interval becomes difficult to interpret because it passes through infinity. This is a signal that the treatment effect isn’t statistically significant, and the NNT shouldn’t be taken at face value.
Number Needed to Harm
The same math works in reverse for side effects. Number needed to harm (NNH) tells you how many people need to receive a treatment before one additional person experiences an adverse effect. The formula is identical, just applied to harms instead of benefits:
NNH = 1 / ARI
Here, ARI is the absolute risk increase: the rate of the adverse event in the treatment group minus the rate in the control group.
Comparing NNT and NNH gives you a quick sense of a treatment’s tradeoff. If a drug has an NNT of 189 for preventing heart attacks but an NNH of 104 for major bleeding, you’re more likely to cause a bleed than prevent a heart attack in any given patient. That doesn’t automatically rule out the drug (a heart attack may be more devastating than a bleed), but it frames the decision clearly. Both numbers depend on baseline risk and follow-up duration, so compare them cautiously across different studies.
Putting It All Together
Here’s a quick checklist for calculating and interpreting NNT correctly:
- Get the raw event rates. Find the proportion of events in the treatment group (EER) and control group (CER) from the study data.
- Subtract to get the ARR. CER minus EER. Use absolute percentages, not relative ones.
- Invert. Divide 1 by the ARR. Round up to the next whole number.
- Attach a time frame. State the NNT alongside the study’s follow-up duration.
- Check the confidence interval. Take the reciprocal of the ARR’s confidence interval boundaries to see how precise the estimate is.
- Consider baseline risk. The same treatment will have a much lower NNT in high-risk patients than in low-risk ones. An NNT from one population doesn’t automatically apply to another.
NNT converts abstract probabilities into a concrete, human-scale number. It answers the most practical question in medicine: how many people do we need to treat before one of them is better off for it?