There’s an awful lot of numbers in medicine. Numbers of patients. Numbers related to all the years of training and numbers of this and that case. There are numbers for all the machines that go bing. There is a finite number of times you can try to look like you find a senior doctor’s one standard joke funny.
There’s also lots of numbers to attach to diseases and treatments. They’re the ones you’re forced to learn along the way – incidence and prevalence, mortality rates, complication rates. Then there’s the number that’s been getting a little coverage of late – the NNT or number needed to treat.
The headlines of recent pieces are a little breathless – “Can This Treatment Help Me? There’s a Statistic for That” or “This Man’s Simple System Could Transform American Medicine”. Part of you clicks on these links expecting the magic number to come with a special one-time limited bonus edition blender to turn the printout of your number into an easily digestible health drink.
And it would be pretty great to have the sort of number that would tell you whether you needed a pill for your complaint. Or surgery. Or acupressure delivered by a particularly well trained capybara. (Yeah, you spotted the odd one out. Does anyone ever really “need” surgery?)The thing is that amongst the text (especially the NY Times one) is a pretty sensible acknowledgement – of course it’s not a magic statistic for everything. In fact I know plenty who would say you can’t use it to give any more than a vague “yeah, it’s probably pretty good.” It’s certainly not useful enough to hand over all future medical decisions to Wayne your local shopping centre tax accountant any time in the future.
You Need More Numbers
If you actually go on to read the articles in a bit more detail (assuming you can get past the little “segregation animations” on the NYT version) they make some points that are needed to put this in perspective.
The first is that your particular NNT only applies to that particular situation. So your 1 in 2000 aspirin preventing heart attack stat applies to that particular prevention scenario. But if you apply aspirin to the patient who has had a heart attack before, the NNT is more like 1 in 77 (to prevent a non-fatal heart attack). So it’s kind of important to know that the NNT you’re talking about actually applies to the particular situation you’re assessing.
More importantly, the NNT is based on research in specific patient populations. One of the biggest challenges in medicine applying clinical research to the patient in front of you. And there’s plenty of patient’s who don’t fit the clinical trial box. A little too old. The wrong extra diseases. The trial was done in an entirely different socio-economic group. The patient doesn’t quite fit? Then the NNT isn’t quite the NNT.
On top of that research studies tend to idealise conditions. In particular, compliance with the intervention is pretty likely to be higher than the real world. Not taking the medications as much as in the trial? The NNT just got weaker again.
And if you’re looking at numbers you don’t get to do it in isolation. You need to look at the potential for bad stuff to happen from the treatment – in this case the number needed to harm (NNH). If you go back to the aspirin example, 1 in 3333 have a major bleeding event when being given the little white pill. And each time there’s the NNT, there’s the NNH you have to consider to weigh up whether the treatment is worth it. (Happily, the NY Times article which really is good mentions they’ll get to that in the next article).
So if your kid has an acute middle ear infection you’d like antibiotics right? Well 1 in 20 do get a reduction in pain once more than 24 hours have passed but there is no change in infection complications or recurrence. 1 in 9 get diarrhoea. Still keen?
There are Some Things Not About the Number At All
There’s another issue that’s pretty central to the practice of medicine too. It’s sort of about patients. And using evidence-based medicine isn’t just about applying numbers and statistics. Evidence-based medicine is a process that seeks the best available information to guide the best course of action for the particular patient being considered, within the context they find themselves.
It’s worth taking the time to consider a broader view of what is required for good practice of evidence-based medicine from this recent piece in the BMJ by Trisha Greenhalgh. In it she highlights the need for the patient to be involved in “real shared decision making” (amongst lots of other issues). This requires the clinician to find out “what matters to the patient – what is at stake for them”. Only in doing that can the dialogue about treatment be appropriately informed.
That’s why the idea of a “stat for that” doesn’t quite work out so simply. There might be numbers that provide information about whether you should adopt a treatment. But there’s not an easily accessible number that says much about what is at stake for the patient. That’s where the people working in health come in.
And maybe capybaras if there’s actually a way to practice evidence-based acupressure.
More? The actual site that provides summaries of lots and lots of NNTs is www.thennt.com
Also, just after I posted this the follow-up NY Times piece on NNH appeared (and chose some of the same numbers as their examples coincidentally). Here’s the link and it is a good read just like the first NNT one that triggered this post.