Thursday, July 10, 2008

Number Needed to Treat

The Number Needed to Treat (NNT) is defined as “the number of patients who need to be treated during a specific period to achieve 1 additional good outcome.” (1) Conversely, it can also be considered to be the number of patients who need to be treated to prevent 1 additional bad outcome. The NNT is actually a measure that can help a reader of a scientific article understand whether a piece of research has practical value. And it arises in part because, while the practical interpretation of scientific research is complex, the move toward evidence-based medicine requires physicians and others to be able to interpret what they read, and journal editors and researchers understand this.

One change we are seeing is that one cannot ignore the size of an effect and focus only on statistical significance. (2) This is why we are seeing greater and greater use of confidence intervals (CI) rather than p values, because the CI provides information about both size and significance. The other change we are seeing is a move toward understanding that absolute changes in risk are more important than relative changes in risk. Bear with me here.

In the reference I cite for this entry (reference #2), Simon notes that he was told that snoring increases the risk of a stroke by 3-fold if left untreated. But, he asks not unreasonably, what was the risk before he knew that, given he is 42, in good health and has no other risk factors? Three times nothing is still nothing; 3 times some very small risk is still very small. Does he need to therefore have treatment?

Better would be some measure of absolute risk (noting that above, that 3-fold increase was a relative risk increase). And this is where number needed to treat comes in. The NNT is the average number of patients a doctor would have to treat to have 1 additional event occur. Here are a couple of examples: for prevention of stroke by taking low-dose aspirin daily, the NNT is 102; for preventing infection following a dog bite by administering antibiotics, the NNT is 16 (or, for every 16 dog bites treated with antibiotics, you would see one fewer infection on average). A perfect NNT would be 1, which is often seen in areas such as orthopedic surgery (ie, for every broken radius treated by surgery, you see 1 less broken radius- which makes perfect sense).

The NNT is calculated from a 2x2 contingency table, and is 1/(absolute risk reduction). As an example, say that among patients treated with detuned ultrasound (a placebo intervention), 17.9% of them later suffered chronic back pain, while in the group that received spinal manipulation only 1.3% later suffered chronic back pain. This is an absolute risk difference of 16.6% (17.9%-1.3%). Doing the math here shows that we have an NNT of 6, so for every 6 people who get the SMT, you will see 1 less case of chronic pain.

This is just a brief introduction to a complex and interesting measure, one that has a great deal of significance when you begin to calculate in costs of treatment, rarity of disease, and a host of other factors.

1. Guyatt G, Rennie D, Meade MO, Cook DJ. Users’ guides to the medical literature: a manual for evidence-based practice. New York, NY; McGraw Hill, 2008; p.793

2. Simon S. Number needed to treat., accessed July 10, 2008


Henry Mueller said...

After reading your BLOG today it came to mind that a brief explanation of the various statistical abbreviations and the significance of there values would benefit the college community in better understanding the significance of the literature.

Anonymous said...

Thanks, Henry- in future I will do just that.