Playing the Odds
(From 2010, still relevant)
Much of what we do today in medicine isn’t treating diseases, but manipulating risk of disease.
Two people with the same elevated cholesterol value may be treated differently because their other risk factors for heart disease are different. A 65-year-old smoker with diabetes and high blood pressure is statistically more likely to benefit from cholesterol lowering medication than a younger, healthier person. Both these people can lower their heart attack risk by 50%, but, in the case of the younger, healthier person with an already low risk of heart disease, 50% of nothing is nothing. One of the findings of the recent JUPITER study was that lower risk patients could also reduce their heart attack risk by lowering their cholesterol. The question is whether they should be treated, since their risk is already low.
The way I explain this to patients is with lottery tickets and rebate coupons.
“If I buy a megabucks ticket and you buy two, you will have twice the chance of winning that I have, but you probably shouldn’t start spending your money yet” usually gets a nod or a smile.
That example illustrates relative risk. Just like in the example with one or two lottery tickets, relative risk isn’t enough to make a treatment decision. You need to know the absolute risk. For example, who would wear an insulated rubber suit just because it reduces your risk of getting hit by lightning while walking your dog by 60%? Most of us would probably say, “No thank you, I’ll take my chances”.
The Framingham Heart Study provides a simple risk calculator for heart disease. With it, I can show patients their relative and absolute risk of disease in the next ten years. I can then show them the impact of reducing that risk by lowering blood pressure, quitting smoking or treating cholesterol.
Our middle aged diabetic, hypertensive smoker may be facing more than 20% risk of getting a stroke, heart attack or a symptomatic blockage of a coronary artery, while the younger, healthier person may have only a 2% risk of disease in the next ten years.
Which one of these patients to treat for their high cholesterol might be illustrated with a question of when you would rather use a “50% off” coupon – buying a cup of coffee or buying a new car?
Let’s look at the wisdom of treating both the low risk and the high risk person for their high cholesterol in order to reduce their heart attack risk by 50%:
If we treat 100 patients with a 25% 10-year heart attack risk for ten years, only 12 would have a heart attack instead of 25. Treating 100 patients for ten years would prevent 12 heart attacks. You would therefore have to treat 8 patients to prevent one heart attack. We call this the Number Needed to Treat (NNT). An NNT of 8 is a pretty good deal.
For patients with a 2% heart attack risk, we would have to treat 100 of them for ten years in order to avoid one heart attack. An NNT of 100 is clearly very different from an NNT of 8, so “50% risk reduction” really doesn’t tell us much if we don’t know the absolute risk.
Here are some more or less surprising examples of the number needed to treat:
Shingles vaccine doses given in order to avoid one case of shingles: 59.
Ear infections treated with Amoxicillin to avoid one ruptured eardrum: 20.
65-69-year old women treated for osteoporosis to avoid one hip fracture: 88.
Cortisone shots to relieve one sore shoulder: 3.
Aspirin prescriptions to prevent one heart attack: 200.
Prostate cancers treated in order to prevent one death: 18-48 (most men with prostate cancer don’t die from their disease)
Adenomatous colon polyps removed to prevent one colon cancer: 50 (only 2% of “precancerous polyps” actually turn into cancer)
The Number Needed To Treat is not popular with the makers of many of today’s blockbuster drugs. In the case of symptomatic treatment, like heartburn, bladder spasms or pneumonia, patients can more easily judge whether a medication works or not. With risk reduction, we’ll never know ourselves whether we wasted our time and money or not.
As physicians we should not accept claims of relative risk reduction without knowing the absolute risk and the Number Needed To Treat.
I remember people in Sweden talking about a book in the sixties, titled “Hur man ljuger med statistik”. The author, it turns out, was American. Darrell Huff’s “How to lie with statistics”, first published in 1954, is still in print. No wonder; statistics are still being used to trick people, including doctors and patients.