But how do you know how likely you are to benefit from a medical treatment or procedure?
One statistic that can help is called the Number Needed to Treat, or NNT.
The NNT tells us the number of people we need to give a drug (or other intervention) to in order for just one person to receive a benefit (or, to prevent just one adverse outcome).
How to calculate NNT
To calculate the NNT, you first have to find out the absolute risk reduction, or ARR. That’s the amount that your risk is reduced by the treatment compared with people who didn’t get it.
The ARR is not a number most people are used to seeing. Studies, news reports, and other media messages are much more likely to focus on a different number, known as the “relative risk reduction,” or RRR, that can be misleading.
Here’s a classic example: This Pfizer ad makes it look like taking their drug, Lipitor, will reduce your chances of having a heart attack by a whopping 36%. But that’s the relative risk reduction. It tends to exaggerate the benefit. (That’s why you’ll often see relative numbers featured in advertisements.)
This 36% number comes from a randomized trial called ASCOT-LLA published in The Lancet in 2003. It showed that 1.9% of people taking Lipitor suffered a heart attack, while 3.0% of the placebo group had one. The relative risk reduction, or RRR, is the ratio of the two risks and is calculated by subtracting the Lipitor heart attack rate (1.9) from the placebo group rate (3.0) and dividing the difference (1.1) by the placebo group rate (3.0). This equals 36%.
But the absolute risk reduction, or ARR, is calculated by simply subtracting the two risks, so 3.0% – 1.9% = 1.1%.
In reality, Lipitor reduced the risk of heart attack from 3% to about 2%, and this 1% difference is the number that people care about. But the Lipitor ad is more interested in promoting than informing, which is why it describes this difference as a “36%” reduction rather than a more helpful and accurate 1% reduction.
So, let’s calculate the NNT using the ARR of 1%, and see how it reframes the drug’s benefits in a more user-friendly way. The NNT is simply the inverse of the ARR; it can be calculated by taking 100 and dividing it by the ARR (1).
100/ 1 = 100
How NNT helps
This means that 99 people need to take the drug, pay for it, run the risk of side effects, and stand no chance of benefit. Of course, no one knows going in who will be that lucky 1 out of 100 who does benefit.
This is the the power of NNT. It gives a sense of scale to discussions regarding potential harms and benefits. In the Lipitor example, if all you read about was the relative risk reduction of 36% highlighted in headlines and advertisements (a likely scenario), your response might be: “Wow! I can cut my risk of a heart attack by over one-third!”
But if you were lucky enough to read some thoughtful news coverage that included the absolute risk reduction of just 1% you might think: “Hmm, that’s a far cry from 36%. I’m going to ask my doctor what she thinks.”
And if you were armed with the NNT number of 100 — realizing you probably won’t be that lucky one person out of 100 who actually benefits from the drug — you might not hesitate to say: “I don’t like those odds at all; especially given the costs and risks.”
It’s important to point out that these decisions are personal, and different people may make different decisions about treatment based on the same information. Furthermore, different people have different baseline risk profiles and different risk tolerance. This means clinical decisions should not be based on NNT alone. It’s just one piece of information that needs to be interpreted in a clinical context and under medical supervision.
Looking at how NNT is calculated it becomes clear that the ideal NNT would be one, because it would mean that all who were treated benefited. But as you might guess, a NNT=1 is rarely, if ever, seen. So what sort of NNT’s are we looking for?
An NNT of 5 or less (≤5) was probably associated with a meaningful health benefit … (while) … an NNT of 15 or more (≥15) was quite certain to be associated with, at most, a small net health benefit.
NNT in the real world with real patients
“That’s where it really helps: when you’re trying to frame benefits or harms for patients who are trying to make treatment decisions,” says Dan Mayer, MD. Mayer is a retired emergency room physician who taught evidence-based medicine at Albany Medical College for 22 years, and is the author of Essential Evidence-Based Medicine.
The classic example comes from the 1980’s in trying to help patients decide between the two clot-busting drugs: tPA (the new drug) and streptokinase (the old drug). Studies found an absolute improvement in the tPA group of 1 percent compared with streptokinase. So the NNT was 100. But tPA cost around $2,000 and streptokinase about $25. So you’d have to treat 100 people at a cost of $200,000 — vs $2,500 for streptokinase — to benefit just one patient. That kind of cost effectiveness information is important not just for patients, but also our health care system.
And Mayer says he’s found the NNT useful in helping patients in another way.
Many new drugs not only have a high NNT and a high price tag, but we usually know less about them. So it helps patients decide this: Do I want to risk taking this highly advertised drug — with a high NNT of a 100 or 1,000 — that hasn’t been all that well studied yet, or the doctors don’t have much experience with? Or, do I want to go with the cheaper, better studied, old standby? That decision is made easier by knowing the NNT.
Closely related to NNT: NNH & NNS
Number Needed to Harm (NNH): The number of people who, if they received the intervention in question, would lead to just one person being harmed.
With NNH, instead of looking at desirable outcomes, you are comparing the absolute risk increase of bad outcomes. The absolute risk increase, or ARI, is the risk of a bad outcome in the treatment group minus the risk of bad outcome in control group.
For example, let’s say a drug increases the risk of a stroke from 10% to 40%:
- ARI = 40% – 10% = 30%
- NNH is 100/ARI … or … 100/30 = 3.
In other words, 3 people would need to take the medication for just one person to have the bad outcome of a stroke.
Number Needed to Screen (NNS): The number of people who need to be screened (for a given duration) to prevent one death or one adverse event.
This is based on the absolute risk reduction (ARR), or how much the risk decreases with a given screening technique.
An example (from the video below) is using CT scans to screen smokers who are at high risk for cancer. Study findings showed a 0.5% absolute risk reduction of death from the CT scanning. So as before, the NNS can be calculated by taking 100 and dividing it by the ARR:
- NNS = 100/ARR = 100/0.5= 200
That means 200 people would need to be scanned — and exposed to radiation, and potentially other harms from biopsies and follow-up procedures — in order to prevent one lung cancer death.
NNT in the real world with real news stories
Here are two examples of news coverage we’ve reviewed in which knowing the NNT made the treatment being discussed much easier to understand.In the spring of 2017, the biopharmaceutical company, Amgen, released a cholesterol-lowering drug called Repatha to much fanfare. Even the New York Times claimed the new drug “has the potential to improve the health and longevity of millions of Americans.”
Amgen claimed it reduced the risk of heart attacks and strokes by 15-20%. But that was the relative risk reduction (RRR). The absolute risk reduction was actually 1.5%. So the NNT was 67 (100 ÷ 1.5) – meaning 1 patient in 67 will benefit from the $14,000/year drug.Later in 2017, the FDA approved a new shingles vaccine (“Shingrix”) made by GlaxoSmithKline. An Associated Press story wrote that the vaccine was about 90% effective in preventing shingles, as well as the nerve pain that lingers after an episode (post-herpetic neuralgia).
But as we reported, the NNT told a very different story:
“About 34 patients would need to be given Shingrix over three years to prevent one case of shingles, and about 260 would have to receive the vaccine to prevent one case of post-herpetic neuralgia.”
Other opinions on the NNT
Many studies show that communicating risk information in only relative terms usually misleads patients into overestimating the benefits of therapies. There also has been some work showing that using NNT alone can be misinterpreted by patients.
Hilda Bastian, PhD is a research scientist, blogger, and cartoonist who thinks the NNT is not an intuitive statistical concept to grasp. In this 2015 opinion piece she argues the NNT requires too much “cognitive gymnastics” for many patients, and it might be easier for doctors and patients to simply communicate in terms of natural frequencies, event rates, or simply comparing relative and absolute risk numbers.
Others point out that interventions with a high NNT can deliver important public health benefits. In the Lipitor example above, for example, although 1 out of 100 chances might not be favorable odds for any individual patient, the cumulative impact of the drug across a population of millions may still be substantial.
For those who want to learn more about other limitations of the NNT, try this article from the Journal of the Canadian Medical Association.
More resources on NNT:
theNNT.com – provides several illustrative cases of NNT, NNH, and NNS.
Students4BestEvidence: a very clear introduction to the NNT complete with graphics.
If you’re someone who learns better by visualizing a concept, here is a very nice video by Aaron Carroll, MD of Healthcare Triage, explaining NNT.