Number needed to treat is the most useful statistic in medicine and almost nobody selling anything uses it. It answers the question a price tag implies: how many people have to buy this for one of them to get the thing they bought it for?
The pooled answer
Twenty-one randomized trials covering 99,599 patients and eight different drugs in the class, with a mean follow-up of 2.4 years. [1] The review found high-certainty evidence that these drugs reduced major adverse cardiovascular events, with an incidence rate ratio of 0.87, 95% CI 0.83 to 0.91, and a number needed to treat of 66. All-cause death came in at IRR 0.88, 95% CI 0.84 to 0.92, NNT 121. Cardiovascular death at IRR 0.87, 95% CI 0.81 to 0.92, NNT 170.
Alongside those, serious adverse events fell 9%, heart attacks 15%, heart failure 15%, acute kidney failure 9% and infections 10%. Gastrointestinal disorders rose 63% and gallbladder disorders 26%. Stroke and pancreatitis showed no difference — findings consistent with what the gastrointestinal meta-analysis found separately.
What an NNT of 66 actually means
Sixty-six people take the drug for about two and a half years, and one major cardiovascular event that would otherwise have happened does not. The other sixty-five get no cardiovascular benefit from it. They may get other things — weight loss, better glycemic control — but not the event prevention, because there was no event coming for them in that window.
That is not a criticism. An NNT of 66 for a hard cardiovascular endpoint is genuinely good, better than plenty of accepted therapies. It is simply what prevention looks like from the inside: almost everyone pays and almost nobody is the one who benefits, and there is no way to know in advance which you are.
Why your number is worse
This is the part that matters most and it is the part that never travels with the statistic. NNT is not a property of a drug. It is a property of a drug applied to a population, and it moves with how much risk that population carries.
These trials enrolled people with type 2 diabetes, established cardiovascular disease, chronic kidney disease or heart failure. They had a lot of events available to prevent. A forty-year-old with a BMI of 32, normal blood pressure and no diabetes has far fewer, so the same relative risk reduction removes far less absolute risk, and the NNT rises accordingly — possibly into the many hundreds or beyond. Nobody has run the trial that would measure it, which means anybody quoting 66 to a low-risk buyer is quoting a number that does not describe them.
What to do with it anyway
Use it as a unit rather than as a value. When a seller or a study implies that a drug prevents heart attacks, the useful question is how many people had to take it for one heart attack not to happen, and over how long. Most claims cannot answer that, and the ones that can usually look more modest once they do — which is the same discipline as asking what a cardiovascular result costs rather than whether it exists.
If you are buying for weight loss, buy for weight loss. The weight evidence is strong, direct, and measured in people like you. The cardiovascular case is strong for somebody else. Pricing the two together is how a purchase gets talked into being more than it is, and the honest version of that calculation keeps them apart. The six-month view is the horizon most buyers actually have.