A note from Teo
Number needed to treat, explained with one division
Number needed to treat tells you how many people take something for one of them to benefit. How to work it out from any study, and what it can't tell you.
If I could put one number next to every health headline, it would be this one. The number needed to treat, or NNT, answers a plain question: how many people have to take this for one of them to benefit?
How to work it out
You need the two numbers from the paper: the risk without the treatment and the risk with it, over the same length of time. Then:
- Subtract to get the absolute difference. Say 10 in 100 people had a heart attack without the pill and 7 in 100 with it. The difference is 3 in 100.
- Divide 100 by that difference. 100 divided by 3 is about 33.
About 33 people take the pill for the length of the study so that one of them avoids a heart attack. The other 32 take it, and for that outcome nothing changes. (Made-up numbers, real arithmetic.)
If the paper gives risks per 1,000 people, divide 1,000 instead. The idea is the same.
Why it's so useful
A relative number ("cuts heart attacks by 30%") sounds the same whatever the starting risk. The NNT doesn't. It shrinks when a treatment helps a lot of people and grows when it helps very few. It also makes you ask the next question: and what happens to the other 32?
Its twin: number needed to harm
Treatments have side effects, and you can count those the same way. If 5 in 100 people on a pill get a side effect and 2 in 100 on the placebo do, the difference is 3 in 100, and the number needed to harm is about 33. One extra person in 33 gets the side effect.
Put the two next to each other and you have most of a decision on one line: one person helped for every so many treated, one person harmed for every so many treated. That's a conversation to have with a clinician, not a sum to do alone.
What the NNT can't tell you
- Whether it's worth it for you. An NNT of 50 can be well worth it for a serious outcome and a gentle treatment. An NNT of 5 can be a poor deal if the harms are worse than the thing prevented. That weighing needs someone who knows your history.
- Anything without a time span. An NNT of 33 over five years is not an NNT of 33 over one year. Always say "over how long".
- Anything about a different group of people. The number comes from the people in the study. If they were younger, healthier or different from you in ways that matter, the number may not travel.
Three common mistakes
- Using the relative number. You can't get an NNT from "30% lower" alone. You need both absolute risks.
- Mixing time spans. Take both risks from the same follow-up period.
- Mixing outcomes. The NNT for "any heart problem" and the NNT for "death from a heart problem" are different numbers. Check which outcome the paper counted.
A question to take to an appointment
"Out of 100 people like me, how many does this help, and how many does it harm, over what time?" Good clinicians like this question. Some will have the numbers ready. If you want it printed on a card, it's one of my Ask-Your-Doctor Cards.
For the arithmetic on your fridge, The Two Numbers is free.
This is education about reading research, not medical advice. Take your questions to someone who can examine you.
