Number Needed to Harm, and How to Read It Against Number Needed to Treat

Number needed to harm is the count of people who would have to take a treatment before one of them experiences a particular harm from it. If a drug causes one extra case of a side effect for every 200 people who take it, its number needed to harm for that side effect is 200. The figure has a twin, number needed to treat, which counts how many people take the treatment before one gains the benefit being measured. Read side by side, the two turn a vague "helps some, hurts some" into plain counts you compare. This is general education, not medical advice, so use it to read the evidence, then decide with a qualified clinician who knows your situation.

I have spent years on both sides of these numbers, producing them and reading them. My doctoral research at the Lund University Diabetes Centre concerns the genetics of type 2 diabetes, where the entire language is risk, and I co-authored a meta-analysis in Diabetes Care that pooled effect sizes across many studies. Pooling teaches you quickly that a benefit reported without its matching harm is only half an answer.

What number needed to harm actually measures

Number needed to harm is one divided by the absolute risk increase, the extra chance of a bad event caused by the treatment compared with not taking it. A small number means harm is common, because few people were treated before one was hurt. A large number means harm is rare. The intuition runs opposite to how the figure reads, so slow down: 20 is worse than 2,000.

Number needed to treat works the same way on the upside. It is one divided by the absolute risk reduction, the extra chance of avoiding a bad outcome thanks to the treatment. A number needed to treat of 25 means that for every 25 people treated, one avoids the outcome the study tracked. The other 24 took the treatment without showing that gain, which is expected.

Both numbers are the inverse of an absolute difference, so both inherit whatever that difference depends on: the baseline risk of the population, how long people were followed, and which outcome was counted. A number needed to harm means nothing until you know the harm, the comparison, and the timeframe attached to it.

A neutral worked example

Let me use invented round numbers so nothing rides on a real product. Imagine a treatment studied in 1,000 people against a comparison group of 1,000, watched for three years. In the comparison group, 50 have a serious event the treatment is meant to prevent; in the treated group, 30 have it. That is an absolute risk reduction of 2 percent, a number needed to treat of 50.

Now the other ledger. Suppose the treatment causes a troubling side effect in 25 of the treated people, against 5 in the comparison group. That is an absolute risk increase of 20 in 1,000, again 2 percent, so the number needed to harm is also 50.

Here the two counts match, 50 against 50, which makes the trade-off vivid. For every person spared the serious event, one acquires the side effect. Whether that is a good bargain depends on how the two outcomes compare in weight. One avoided stroke against one mild rash is an easy call; one avoided rash against one serious bleed is the opposite. The numbers set up the question without answering it.

Why the two numbers belong together

A benefit figure shown without its harm figure invites a lopsided decision, and a harm figure shown alone invites needless fear. The honest summary names both counts in the same breath, over the same timeframe, in the same population. A number needed to treat measured over five years cannot be laid against a number needed to harm measured over six months, and both should come from comparable groups, ideally the same trial, or the ledger is rigged before you read it.

Weighting matters as much as counting. A number needed to harm of 50 for a fleeting, reversible problem is not the equal of the same figure for something lasting. People reasonably value outcomes differently, which is why this calculation ends at a conversation rather than a verdict. The arithmetic narrows the choice; your priorities and your clinician close it.

How to translate a benefit and harm claim in a minute

You do not need statistics training for this. When a treatment is described, ask for the benefit as an absolute number first: out of 1,000 people treated for a stated time, how many gain? Then ask the same about harm, and how badly. Be wary of mixed framing, where the upside arrives as a large relative figure and the downside as a small absolute one. A treatment that "halves" a risk while "rarely" causing a problem has paired two scales that cannot be compared. Put both as counts out of the same denominator, and the lopsided impression flattens.

One caveat keeps these numbers from being oversold. Many harms surface only after a treatment is used widely, so an early number needed to harm can look reassuring simply because the harm had not yet been seen. A figure of "none observed" is not "none exists." Time and larger populations revise these numbers.

Where this matters most

Two situations deserve extra care. The first is prevention in people who feel well, where the benefit is often a small absolute reduction spread across many people, pushing the number needed to treat high and making even a modest harm count loom larger. The second is treatments taken for a long time, where small annual harms accumulate while the benefit may plateau.

None of this makes side effects a reason to refuse useful treatment, nor a measured benefit a reason to ignore real harm. The error is reading one number without its partner. So keep the habit simple. For any treatment, ask how many people are helped and how many are hurt, over the same time, in people like you, and how much the help and the harm each weigh. A recommendation that answers all four earns your trust. One that offers only the cheerful half is unfinished.