Before you read any further, take a quick test. Answer honestly. No one is watching. Three new drugs, three trials. Which one would you most like to prescribe?
- Drug A reduces mortality by one-third.
- Drug B lowers mortality from 6 percent to 4 percent.
- Drug C raises survival from 94 percent to 96 percent.
If Drug A felt like the winner, congratulations. You’re normal. You’ve also just been fooled, because all three describe exactly the same result.
This isn’t a trick I made up. In a randomized study published in 2011, physicians were shown one of these framings. When the result was given as a relative reduction, about 94 percent rated the new drug more effective. When the identical data were given as absolute survival, about 52 percent did. Patients answered almost exactly the same way. Our training bought us almost no protection.
I’ve practiced medicine for more than forty years. I’m a lipidologist. I’ve stood at the front of the room and clicked to the slide with the big percentage on it. So this isn’t an essay about other doctors. It’s about us.
Real physician voices, twice a week
Free, and one click to unsubscribe.
I know what a great drug looks like
Before insulin, type 1 diabetes was a death sentence. Children wasted away on starvation diets. Insulin didn’t nudge a lab value. It turned a fatal disease into a manageable one.
Antibiotics did the same for bacterial meningitis. Direct-acting antivirals now cure hepatitis C in more than 95 percent of patients, often in 8 to 12 weeks. HIV pre-exposure prophylaxis, taken as prescribed, cuts the risk of sexually acquired HIV by about 99 percent.
Yes, a huge relative effect doesn’t automatically mean a tiny number needed to treat (NNT); that depends on baseline risk. And curing an infection isn’t the same math as preventing a future heart attack. But these treatments change the natural history of disease.
That’s the standard I have in mind when I hear the word breakthrough. Not a slightly prettier Kaplan-Meier curve.
The magic trick, and why it works on us
If I tell you a drug cuts your risk of heart attack by 50 percent, you’re impressed. If I tell you it lowers your five-year risk from 2 percent to 1 percent, you start asking about cost and side effects.
Same result. The first is a 50 percent relative risk reduction. The second is a 1-point absolute risk reduction. NNT: 100 people for five years to prevent one event.
Every one of us was tested on this distinction. We could define it in our sleep. The trouble is that knowing it and feeling it are different things, and the research on physicians is humbling.
In 1992, a study in the Annals of Internal Medicine showed clinicians the same trial results in different formats. Doctors rated the treatment as less effective when they saw absolute numbers or NNT than when they saw relative risk reductions.
In 2017, a systematic review in JAMA Internal Medicine covered 48 studies involving more than 13,000 clinicians. In only 3 of 28 assessments of treatment benefit did a majority of clinicians estimate the benefit correctly. We are remarkably good at recommending treatments whose actual benefits we routinely overestimate.
Sit with that for a second. Every “shared decision” begins with the doctor’s estimate of benefit. If that estimate is inflated, the conversation is rigged before the patient says a word. We call it informed consent. Too often it’s consent informed by a number we don’t actually know.
FOURIER: a stunning lab result, a modest clinical one
In FOURIER, the PCSK9 inhibitor evolocumab lowered LDL cholesterol by about 59 percent. That’s a remarkable biochemical feat.
Over a median 2.2 years, the primary composite endpoint occurred in 11.3 percent of placebo patients and 9.8 percent of evolocumab patients. Absolute risk reduction: 1.5 points. NNT: about 67. Heart attacks and strokes fell. Cardiovascular and all-cause mortality did not significantly change during the trial.
A 60 percent drop in LDL turned into a 1.5-point drop in a composite endpoint over two years.
My cardiology colleagues will object, and they have a point. The trial was short. In the open-label extension, FOURIER-OLE, patients originally randomized to evolocumab had lower cardiovascular death than those who started it later: 3.32 percent versus 4.45 percent (hazard ratio 0.77). Benefit seems to build over time.
I take that seriously. I’d also note it was open-label, industry-funded, and limited to patients who finished the parent trial.
But notice what happens when we state even the favorable result honestly. Cardiovascular mortality was about 3.3 percent versus 4.5 percent over a median five-year extension, a difference of roughly 1.1 percentage points. Both groups received evolocumab during that extension, so this compares earlier with delayed treatment among selected participants, not an ongoing randomized placebo comparison. The benefit may be real. But turning that difference into a definitive NNT of 88 would oversell the evidence. The magnitude still matters.
The longer-follow-up argument doesn’t rescue relative risk. It makes the point stronger: The honest answer depends on the time horizon, so you’d better be able to say it.
Statins: the drug we all prescribe
A 2022 systematic review for the U.S. Preventive Services Task Force found that statins in primary prevention reduced all-cause mortality by an absolute 0.35 points over trials lasting one to six years. NNT: about 286. Fatal or nonfatal heart attack fell by 0.85 points. Real benefits. Statistically significant. Much smaller than most patients imagine.
Now the uncomfortable question. The last time you prescribed a statin for primary prevention, did you tell your patient the estimated absolute benefit for someone with their particular risk profile? Or did you reach for the much more impressive relative risk reduction?
I’m not arguing statins don’t work. They do, and for some patients the benefit is substantial. I’m arguing that effective has become a nearly meaningless word unless someone says how much.
Who paid for your lunch?
Almost no physician believes a sandwich changes how they prescribe. In 2016, researchers writing in JAMA Internal Medicine linked industry payment records to Medicare prescribing for nearly 280,000 physicians. Doctors who got a single sponsored meal, averaging under $20, were more likely to prescribe the brand-name drug being promoted. More meals, and pricier meals, were linked to more prescribing. One of the four drugs studied was rosuvastatin.
The authors were careful to call it an association, not proof of cause. Fine. But I’d ask you to notice your own reaction. If your first thought was “that’s other doctors, not me,” you’ve just repeated the exact thinking the framing studies describe.
The drug rep doesn’t need to lie to you. The relative risk reduction is true. The rep only needs you to remember the big number and forget to ask for the small one.
Now follow the money
Imagine a drug that lowers an important event by 1 absolute point over five years. NNT: 100. At $6,000 a year, treating 100 people for five years costs $3 million to prevent one event, on average. At $5 a month, the same benefit costs $30,000.
This is a hypothetical, not a cost-effectiveness analysis. It ignores downstream savings and quality of life. And preventing a fatal heart attack matters far more than preventing a short hospital stay.
Of course, someone will point out that a 1 percent absolute reduction, applied to a million people, prevents 10,000 events. They’re right. Population health matters. But so does the experience of the individual patient who pays for and takes that medication every day. A public health benefit doesn’t eliminate our obligation to explain the modest probability of individual benefit. Nor does it mean that any price is justified.
That’s exactly why a 1 percent absolute benefit isn’t automatically worthless. A 1 percent reduction in death can be well worth pursuing, especially when a drug is cheap and safe. But 1 percent isn’t a breakthrough just because the p-value is impressive. Statistical significance tells us something about the evidence against chance alone. It tells us almost nothing about whether the benefit is large enough to matter to the person taking the drug.
We started treating probabilities
We no longer just treat disease. We treat the probability of disease. There’s nothing wrong with that. Prevention is one of medicine’s noblest goals. But risk is not illness. A patient with meningitis has a disease that demands treatment today. A healthy 55-year-old with a 7 percent ten-year cardiovascular risk has a probability.
Both can produce equally forceful prescriptions. And when industry finds a way to lower a biomarker tied to that probability, we sometimes act as if the clinical question has already been answered.
It hasn’t. Patients don’t live in their lipid panels. They live in a world where drugs cost money, cause side effects, and require years of adherence. They deserve to know what those sacrifices are likely to buy.
A 30-second audit
Here’s my challenge. Think of the last five preventive prescriptions you wrote for a chronic condition. For each one, can you state:
- That patient’s baseline risk
- The absolute risk reduction
- Over how many years
- The NNT
- The annual cost to the patient
- Whether the trial endpoint was death, a hard clinical event, or a composite
If you can’t, you weren’t sharing a decision. You were passing along a slogan. You’re not alone. The evidence says most of us can’t. But “everyone does it” is an excuse we’d never accept from a patient.
A great drug, redefined
A great drug substantially reduces the chance of an outcome patients actually care about. Its benefit is large enough to notice, not just large enough to publish. It’s reasonably safe, reasonably affordable, and proven on something more meaningful than a lab number.
Not every worthwhile drug needs an NNT of two. Some inexpensive treatments with NNTs of 50 or 100 are perfectly reasonable.
But we should stop giving every statistically significant result the same applause. We should stop treating a composite endpoint as if it were a mortality benefit. Preventing a hospitalization, a nonfatal heart attack, and a death all matter, but they’re not interchangeable. And we should stop letting relative risk do the talking while absolute risk waits in the supplementary appendix.
Thirty percent of what?
How much benefit is enough? Not physicians, societies, drug companies, or insurers should decide that alone. Patients should. Our job is to tell them the truth in words they can understand.
“Take this for five years, and your chance of a heart attack falls from 2 percent to 1 percent.” That’s an honest conversation. “This cuts heart attacks by 50 percent.” Also true. Hardly the whole story.
Medicine has become brilliant at measuring tiny differences between groups. We’ve become just as brilliant at making them sound enormous, and we’ve fooled ourselves along the way.
So the next time someone announces a drug that cuts risk by 30 percent, ask the two questions I’ve been asking for decades. Thirty percent of what? And how many people must take it, for how many years, at what cost, to prevent one bad outcome?
If you can’t answer, neither can your patient. And that should make all of us uncomfortable.
Larry Kaskel is an internist and “lipidologist in recovery” who has been practicing medicine for more than thirty-five years. He runs a concierge practice in the Chicago area and serves on the teaching faculty at the Northwestern University Feinberg School of Medicine. He is also affiliated with Northwestern Lake Forest Hospital.
Before podcasts were even a thing, Dr. Kaskel hosted Lipid Luminations on ReachMD, building a library of more than four hundred shows featuring thought leaders in cardiology, lipidology, and preventive medicine.
He is the author of Dr. Kaskel’s Living in Wellness, Volume One: Let Food Be Thy Medicine, a book that blends evidence-based medicine with practical strategies for a longer healthspan. His current work focuses on rethinking the cholesterol hypothesis and exploring the infectious roots of atherosclerosis. More information is available at larrykaskel.com.
