Increasingly people are accessing healthcare information in order to make decisions for their own health. A 2010 Pew poll found that 80% of internet users will do so for health care information. This presents a huge potential benefit, but also a significant risk.
Daniel Levitin talks about the need for public information literacy, something we also discuss frequently here on SBM. If you are accessing the internet to inform your health care decisions, then you need to know how to determine the legitimacy and trustworthiness of the websites you are visiting. There is a big difference between NaturalNews (a crank site full of misinformation and conspiracy theories) and Nature News (an outlet for one of the most prestigious science journals in the world).
Even when you can discriminate between good and bad health information websites, the challenge remains to properly interpret the scientific information to which you now have access.
Part of the problem is that the original source of scientific information is the published technical literature, which is designed for optimal communication among experts. It is extremely difficult to properly interpret a technical paper unless you have a fair degree of expertise in the field. Even as a physician it can be challenging for me to fully understand a research study in a medical field far from my specialty of neurology.
Most of the public, therefore, has to rely on experts to interpret that data for them and present it in a way that is digestible and meaningful. Properly communicating complex science to a non-expert general public is a skill set unto itself, and presents many challenges.
Unfortunately, often that interpretation is being done by non-experts, by journalists who many not even have a specialty in science journalism. Even worse, the information could be presented by biased, commercial, or agenda-driven sources.
As one example, often reporting of health information does not properly distinguish between relative risk and absolute risk. A headline might declare that a new treatment reduces the risk of a bad health outcome by 50%. That sounds impressive. But this is a relative risk reduction, and may reflect an absolute risk reduction of 0.2% to 0.1%, or an absolute risk reduction of 0.1% (that sounds less impressive).
Number needed to treat
This brings us to a specific method for representing the health effects of a treatment or the effectiveness of a diagnostic procedure – the number needed to treat (NNT). Daniel Levitin mentioned above advocates for the NNT as a way of communicating health information to the public, and he has a point.
There is also a group of physicians, the NNT group, who advocate for using the NNT to communicate health information. They have a useful website that translates healthcare statistics into NNT.
Put simply, the NNT is the number of patients who need to take a treatment in order for one person to receive a specific benefit. This may put statistical information into a context that makes it easier for the general public, and physicians also, to understand.
For example, a few years ago I wrote about a Cochrane review of statin drugs for primary prevention of heart disease. The review found:
What they found was that total mortality had a relative risk reduction of 17%, risk of heart attacks was reduced by 28%, and strokes by 22%.
If the information were only stated in this way, even identifying the numbers as relative risk, the average person might walk away thinking the effects are fairly significant. However, you could also represent the data by stating that the NNT was 1 in 1,000. The data show that for primary prevention you have to treat 1,000 patients for one year in order to prevent one death. This still may be worth is, as death is a pretty significant negative outcome, but it helps understand what the effect of the drugs truly is.
Levitin also uses statins as an example. One popular statin has an NNT to reduce cholesterol of 1 in 300 – only one person is 300 will actually have their cholesterol reduced by the drug. However, the number needed to harm is 20 – one person in 20 being treated will have side effects (muscle pain and cramping, for example).
For another example, let’s take a look at using warfarin instead of aspirin for atrial fibrillation in order to prevent strokes and blood clots. The NNT for strokes was 60, while the NNT for systemic blood clots was 360. Meanwhile, 1 in 167 had a fatal hemorrhage while 1 in 25 had a bleed requiring hospitalization. In other words, for every three strokes you prevent, you cause one death and put seven people in the hospital with a bleed.
The NNT is also a good way to look at cost effectiveness. This is unfortunately increasingly important as we try to rein in rising health care costs. No one wants to think in these terms, but it can be useful to consider how much it costs to treat those 300 people for one year and the 15 who will have side effects in order to reduce the cholesterol in one person. The NNT makes it a little more intuitive to see what we are getting for our health care dollars.
Conclusion: NNT provides useful context for medical decision-making
The NNT is a legitimate and highly useful way to present medical statistics. It is perhaps more intuitive to understand than relative or absolute risk reduction. It might be useful for medical journals to require scientists to provide NNT data for their studies. This would also make it easier for journalists and science communicators to provide the NNT as well.
We advocate an approach to systematic reviews that distills information into, in effect, one number: the NNT. This is simple to remember and directly supports efforts to work with patients to make the best possible clinical decisions for their care.
Almost two decades later, the NNT is gaining ground but has not achieved widespread adoption. Habits are often slow to change. Perhaps the explosion of health care information online will provide a further incentive to adopt the NNT as a standard format for communicating health statistics to the public.