What is the Number Needed to Treat?

In science and particularly medicine, the public and patients can often be sold on statistics that are based on poor research methods and even poorer interpretation of those data by physicians. I am by no means the expert on science, although my PhD was in basic science and I tend to be a bit critical. However, I want to share with you how to not get bamboozled by statements like "it cut the number of heart attacks in half!" next time you are in a medical conversation.

In science we are concerned with showing that there is some difference when we compare two treatments. For example, if I increase blood flow (and oxygen) to a muscle, does it now use more oxygen than before? If so, does it now do more work than before? Often times we report this as "statistically significant" when we determine that "when there is no difference between treatments, the percentage of times we would accidentally say there is a difference is less than 5%". Read that sentence again - it is important.

This is often expressed in research as a "p value" of less than 0.05 or a "risk" of some sort that is above 1.0. When you see "relative risk" in a paper comparing two things, the risk is reported as 1.2-1.9 or some interval. A risk of 1.0 means there is no difference, so anything above that means there is an increased risk.

Example:
Two groups were examined, both with LDL cholesterol above 220. One group took a statin and the other did not. The no-statin group had an increased relative risk of death from cardiovascular disease over the next 10 years (RR 1.9-2.2). This means the no statin group had double the risk! Also, the statin group reduced LDL to 160, while the non statin group did not.

Seems impressive right? This means you should take the statin. Right? However, this is where the public and some physicians get tripped up. What if the risk of dying in that 10 years without the statin in the over 220 LDL group was 1 in 4 million? And by taking the statin your risk becomes 1 in 2 million, would you take it? Probably not. Even though you are accepting DOUBLE the risk! Now what if statins caused side effects in 1 in 10 or 1 in 20 patients? Now would you still take it? This is where our very important concept of "number needed to treat" and "number needed to harm" comes in.


Number needed to treat (NNT) is just what it sounds like - the number needed to treat to prevent 1 event (in this example it is death). If the NNT is 50, that means 50 people need to get the drug to prevent 1 death! NN to Harm (NNH) is the same concept - how many people do you need to treat to harm one patient?? In our example, if the NNT for 10 years was say 500 and the NNH was 10, would you take it?? Likely not - but this still gets reported in Time Magazine or in your local newspaper or even (gasp) by your local physician as "this drug cuts your risk of dying from a cardiovascular disease in half! you'd be stupid not to take it!"

Don't be fooled! Dig deeper! Next time you are speaking with your doc or reading a new report in Time, ask, "what was the NNT for this treatment?? And the NNH?" Then and only then can you accurately assess whether you are interested in the treatment or not. Be well!