I started writing this weeks’ blog post about a recent article I saw in the online magazine Vox regarding the safety and benefits of statins (the most commonly used type of medication for cholesterol lowering). However, once I started writing I realized that I needed to make a major digression to explain the concept of “Number Needed to Treat”, which in turn doesn’t fully make sense until one understands the concepts of relative versus absolute risk. So I will instead make this weeks’ entry another in my series on how to read and interpret the medical literature and defer my thoughts on the Vox article to a future post.
Let’s start with the difference between absolute and relative risk. Absolute risk is your actual statistical chance of having a particular thing happen to you; relative risk is how big of a change there is in that absolute risk given a particular set of circumstances.
For example, if your risk of catching a particular strain of a virus is 4% this year, and a vaccine could cut that risk to 2%, 4% and 2% would be your absolute risk for catching that virus before and after the vaccine. But since 2% is half of 4%, we would say you reduced your relative risk by 50%.
As a real-life example, let’s take one of the most feared of diseases: ALS, also known as Lou Gehrig’s disease. The risk that an American will develop this disease at some point in their lifetime is about one in four-hundred, or 0.25%. However, we know that some people are at higher risk than others. Airline pilots, for example, have roughly double the lifetime risk of developing ALS as the average person, for reasons that are not fully understood. So we would say that an airline pilot has a 100% relative risk of developing ALS as compared to the general population.
Note that compared to the general public, a pilot’s relative risk of developing ALS is 100% higher, but their absolute level of risk still remains a pretty low 0.5%. (0.5% being double – aka 100% more than – 0.25%).
This difference between relative and absolute risk is a crucial distinction, as it appears all over medicine. To take a few other examples:
The average person’s lifetime risk for developing colon cancer is about 4%. Some studies suggest that high consumption of red meat can increase this risk by about 20%, meaning there is a 20% increased relative risk for colon cancer in red meat eaters. This raises absolute risk from 4% to 4.8%.
In the JUPITER study, a landmark drug trial that compared the drug Rosuvastatin (brand name: Crestor) to placebo, there was a 44% reduced relative risk for heart attack or stroke in people who took Crestor as compared to placebo. This was based on the fact that amongst those taking a placebo, 1.36% suffered a cardiovascular event, whereas amongst those taking Crestor, only 0.77% did. In other words, if you took Crestor during the study, you dropped your risk for a cardiovascular event from 1.36% to 0.77%, a 44% relative reduction in risk (because 0.77 is 44% less than 1.36)
The use of relative and absolute risk to express study results generates a lot of controversy because it is often pointed out – correctly – that these differences are used to create incorrect perceptions amongst the public. For example, a headline that tells you your steak might increase your risk of colon cancer from 4% to 4.8% would not grab much attention; a headline that tells you that eating steak will increase your risk of this deadly disease by 20% sounds much more compelling. And pharmaceutical companies love to advertise with large relative risk numbers for benefits but smaller absolute numbers when explaining the risks of their products (“Studies show that Dynamite cuts the risk for Leprosy by 60%, while only causing 1% of patients to spontaneously combust!”)
There are in fact good uses for both types of statistics, which is why they exist, but that gets into a more complex conversation, so let me turn our attention now to the third metric I want to discuss today: Number Needed to Treat, aka NNT.
NNT simply means, on average, how many patients need to be treated with a particular intervention for one person to benefit. It is a way of quantifying a change in absolute risk, and provides a very useful and practical shorthand for how likely something is to provide actual benefit to a patient.
For example, the NNT for penicillin and strep throat is 3, meaning that if I have three patients with strep throat and give all three of them penicillin, I can expect to cure one of them as a result. It’s important to note here that that is not because only one out of three patients with strep throat who get Penicillin recover. Rather it’s because all three patients will get better, but had we left all three patients alone, two of them would have gotten better on their own. Ergo, only one of the three patients actually benefited from the antibiotic.
There are a few factors that go into calculating NNT. One is how effective the treatment itself is. Another, as the above example indicates, is how well people will do if they are just left alone and not treated. But a third is how much baseline absolute risk the person being treated has.
To emphasize this last point, let’s imagine a drug that cuts the risk for a heart attack by 50%. (What type of risk reduction is that? That’s right – a RELATIVE risk reduction).
Now imagine that we administer this drug to people in their 20s. What would the NNT be? The answer is about 100,000 (that’s a hundred thousand). Why? Because heart attacks in this age group are vanishingly rare. Only about one person in 50,000 will suffer a heart attack during their 20s, so we’d expect an untreated group of 100,000 people at this age to contain about two heart attacks. If we administered this medication which eliminated half of all heart attacks, we’d then expect to see just one person have a heart attack, instead of two. So in this scenario, our 50% relative risk reduction translates to a one in one-hundred thousand absolute risk reduction, or an NNT of one-hundred thousand.
By contrast, let’s imagine the same scenario in a group of people in their 70s. This age group suffers much more heart disease. About 10% of people, or one in ten, will suffer a heart attack at some point in their 70s. Therefore if we give our hypothetical drug to twenty people, we would expect to prevent one heart attack, for an NNT of 20. Here the 50% risk reduction translates to a one in twenty absolute risk reduction or an NNT of 20.
Putting all three of our terms together then, two out of every twenty people will have a heart attack between ages 70-80, for a 10% absolute risk. If we had a hypothetical drug that cut that risk in half, we would instead see 5% of people, or one in twenty, have a heart attack during that decade. Rephrased, the drug:
would cause a 5% absolute risk reduction, from 10% to 5%
would cause a 50% relative risk reduction, from 10% to 5%
would save one in 20 people from having a heart attack, for an NNT of 20, because now for every twenty people only one would have a heart attack instead of two.
I know this can all get a bit confusing to the uninitiated, so don’t worry if you walk away from this feeling like you have not completely grasped it. But it is helpful to just have a basic understanding of these three terms, as they really help to better put into context the risks and benefits of various medical treatments. They will also pave the way for my discussion of that Vox article in a future post.
