Number Needed to Treat: A Useful Summary With Important Limits

Number Needed to Treat (NNT): A Quick Definition

Number needed to treat, or NNT, is the average number of people who would need to receive a treatment for one additional person to get the specific benefit a study measured, compared with not receiving it, over a set period of time. An NNT of 20 means roughly 1 in every 20 treated people gets that extra benefit — the rest would likely have had the same outcome either way, or the treatment simply didn’t help them.

NNT is popular because it turns an abstract percentage into a number of people, which is easier to picture than a relative risk reduction. But an NNT only means something next to the exact study population, outcome definition, and time frame it came from. Change any one of those, and the same treatment can produce a very different NNT.

Terms to Know Before You Read an NNT

Baseline risk (comparator group risk): How likely the outcome was in the group that did not get the treatment being studied. This number moves the NNT even when the treatment’s relative effect stays the same.
Absolute risk reduction (ARR): The straightforward difference in outcome rates between the treated and untreated groups — for example, 20 percent minus 12 percent equals an 8 percentage-point difference. NNT is calculated directly from this number.
Relative risk reduction (RRR): How much smaller the risk is in the treated group, expressed as a percentage of the untreated group’s risk. RRR tends to look similar across different risk levels even when ARR — and therefore NNT — does not.
Confidence interval (CI): A range showing the uncertainty around a study’s estimate. An NNT is reported as one number, but it usually comes with a wide range of plausible values worth checking.
Time frame: The length of follow-up the NNT applies to. An NNT over 1 year is not the same statistic as an NNT for the same treatment over 5 years.

Why the Same Treatment Can Produce Very Different NNTs

A treatment’s relative effect often stays fairly stable across different groups of patients. Its absolute effect — and therefore its NNT — does not. Because NNT is built from the absolute risk difference, it depends heavily on how likely the outcome already was for the people being treated.

The Cochrane Handbook, a core methods reference for interpreting clinical trial results, illustrates this with oral anticoagulants (blood thinners) used to prevent stroke in atrial fibrillation. Among higher-risk patients with a prior stroke or TIA — roughly a 12 percent yearly stroke rate — the treatment prevented a substantially higher number of strokes per 1,000 patients than it did among lower-risk patients with only about a 2 percent yearly rate. Same treatment, same relative effect, very different number of people who needed to be treated to prevent one stroke — because the starting risk was so different between the two groups.

This is the central limit of NNT: it’s a comparative measure tied to one specific population’s baseline risk, not a fixed property of the treatment itself. A person whose own risk profile looks nothing like the study population may see a real-world NNT that differs meaningfully from the one reported in the paper.

What Else Shifts an NNT

How the outcome was defined. An NNT for “any improvement” on a symptom scale is a different question — and a different number — than an NNT for “complete resolution” of that same symptom. It’s inaccurate to label a result as the NNT for a whole condition when it was calculated for one specific cutoff on a scale.
The length of follow-up. Outcomes measured over 1 year versus 5 years produce different event rates, which changes the absolute risk difference the NNT is built from.
The study population’s baseline risk. As shown above, higher-risk patients generally produce a lower — more favorable-looking — NNT than lower-risk patients, for the identical treatment.
How the statistic was pooled. When an NNT comes from a meta-analysis of several trials, the calculation depends on an assumed baseline risk chosen by the researchers, not necessarily the risk of any one reader.

A Field Guide: Questions to Ask Before You Trust an NNT

What outcome was actually counted? Look for the specific event or improvement threshold the NNT measures, not just the general condition it’s associated with.
Over what time period? An NNT with no attached time frame is missing a required piece of context.
What was the baseline risk of the comparison group? Without this, you can’t meaningfully compare the NNT to a different population.
Does the confidence interval cross into “no meaningful difference”? A wide interval means the true number could be considerably higher or lower than the quoted figure.
How closely does the study population match the group you’re thinking about? Age, disease severity, and other risk factors can shift a real-world NNT away from the published one.
Was the NNT from a single trial or pooled across several? Pooled NNTs from meta-analyses carry an extra layer of assumptions about baseline risk that a single-trial NNT doesn’t.

A Worked Example

Suppose a trial finds that 20 percent of an untreated group develop a bad outcome, compared with 12 percent of a treated group. The absolute risk reduction is 8 percentage points. Dividing 100 by 8 gives an NNT of about 13 — roughly 1 in 13 treated people experiences the extra benefit, over whatever time frame the trial covered, in a population with that same 20 percent baseline risk.

Now change only the baseline risk: in a lower-risk population where just 6 percent of untreated people would have the outcome instead of 20 percent, an identical relative risk reduction produces a much smaller absolute risk reduction — which pushes the NNT considerably higher. This is exactly why you shouldn’t automatically apply an NNT from one study population to another.

What This Means for Reading Study Results

NNT is a useful shorthand, not a personal commitment. An NNT of 15 doesn’t mean any one person has “a 1-in-15 chance” of benefiting — it describes an average across a group matching the study population, over the study’s time frame, for the outcome the study defined. The most careful use of an NNT pairs it with its baseline risk, time frame, exact outcome definition, and confidence interval, treating all four as one package rather than optional extras.

This is general research literacy information, not personalized medical guidance, and isn’t a substitute for talking with a qualified health professional about how a specific study’s NNT might apply to an individual situation. For an urgent health decision, contact a healthcare provider or local emergency services rather than relying on published statistics alone.

Common Questions About Number Needed to Treat

Is a lower NNT always better?

Generally yes for benefit — a lower NNT means fewer people need treatment for one to benefit. But you have to compare it against the “number needed to harm” for the same treatment, and against how serious the outcome being prevented actually is.

Can you compare NNTs between two different studies?

Only carefully. If the two studies had different baseline risks, different outcome definitions, or different follow-up periods, their NNTs aren’t measuring the same thing even if the number itself looks similar.

Does NNT apply to an individual patient?

Not directly. NNT is a population-level average. An individual’s own risk factors can make their personal likelihood of benefiting higher or lower than the study’s headline number.

Why do some sources report NNT as a range instead of one number?

Because NNT is derived from a confidence interval around the treatment effect. Reporting a range is more honest about the uncertainty than presenting a single figure as exact.