What practicing evidence-based medicine looks like in the messy clinical world and why EBM calls for more, not less, art in the practice of medicine
I’ve been hearing a lot about intermittent fasting as an approach to weight loss lately. This comes from physicians, patients, and celebrities. I never pay attention to this kind of chatter because, as I have written, I think weight loss advice (outside of GLP-1s and bariatric surgery) arises from a data free zone. However, I’ve realized that intermittent fasting serves as a good example from which to discuss the complexities of actually practicing medicine according to evidence-based medicine (EBM).
Sensible Medicine is a reader-supported publication. If you appreciate our work, consider becoming a free or paid subscriber.
Intermittent fasting is what it sounds like. There are several different ways to do intermittent fasting, but essentially you choose times during which you eat and fast. Most people start by extending their evening fast to midday, leaving about 8 hours during which they eat.
EBM is much misunderstood and, these days, often maligned, equating it to slavish devotion to the latest RCT in a high impact factor journal. What EBM really is a three-legged stool supporting your practice. It is the process of integrating clinical experience and expertise with the best available evidence from systematic research. I can’t even count the number of talks I given in which this slide appears.
So how does EBM apply to intermittent fasting? Intermittent fasting comes with a beautiful mechanistic explanation about how it would work (leg #1). Weight loss occurs when energy outlay exceeds energy input but in a way that this deficit does not cause hormonal mechanisms to slow metabolism. There is an entire genre of literature, both studies in medical journals and lay books that outline this mechanism AND prove that it there is some truth to the theory. Intermittent fasting limits caloric intake – if you are eating for 8 hours of the day rather than sixteen you are likely to consume fewer calories. Also, during fasts insulin levels fall, enabling the body to access stored fat. But, of course, most medical interventions that should work don’t when they are studied in actual humans with the rich behavioral and physiologic complexities that we are only beginning to understand. This is pretty much the thesis of Ending Medical Reversal.
How about clinical experience (leg #2)? Intermittent fasting has an impressive following.[i] If we are not going to call all the people practicing intermittent fasting delusional and all the doctors recommending it hucksters, we must assume that it is having some positive effect in some people. Now, I am aware of the placebo effect (see chapter in Ending Medical Reversal).[ii] Doctors are amazingly susceptible to convincing themselves what they recommend works and people are equally apt to feel better when they take something or do something that they believe will work. However, weight is an objective measure. Few people will swear by a weight loss strategy that is ineffective.
Now what about the data (leg #3). As far as I can tell, there have been two nice studies, RCTs, of intermittent fasting to induce weight loss. One was published in the NEJM and one in JAMA IM. Both were negative.
How should you look at a negative study? First it could be a negative study, the treatment didn’t work. Second it could be underpowered. A difference exists between the intervention group and the placebo one but the study only had the numbers to detect a difference larger than the actual difference. I think most people understand power but forget that having greater power might make a smaller difference statistically significant but it will not change the magnitude of the difference. In the NEJM study, the intermittent fasting group lost 8.0 kg while the control group lost 6.3 kg. Power or no power, that is still only 1.7 kg over a year.
Third the study could have myriad problems by which it fails to adequately test the question: wrong patients, poorly delivered intervention, poorly chosen controls… This is where the treatment evangelist often hangs his hat. “Well you didn’t use this dose” or “You enrolled the wrong patient.” We always have to remember that the burden of proof is with the person trying to sell a new therapy.
The last possibility with a negative study is that the study was just wrong. The following graphs were famously outlined in JPA Ioannidis’ Why Most Published Research Findings Are False. This has to do with the alpha and beta error we accept when we design a study. Alpha error, generally set at 5% in biomedical research, is the chance of finding a difference when one does not exist and beta error, generally set at 20%, if the chance of finding no difference when one exists.[iii]
This table shows where alpha and beta error fit in. The columns represent truth, does a difference really exist between the intervention and the placebo. The rows are what we discover in our study. Thus, we may have a true positive study, one that finds a difference when one exists or all of the other possibilities.
This next table is the important one. Here we have populated the table with numbers. Think of this as showing the results of 1000 studies. The likelihood, “prestudy probability,” of the treatment being effective, in this example is 50%, thus there are 500 in each column. Given that you know the alpha and beta error. The beauty of this is that knowing your pre-study probability and your alpha and beta error, you can fill in all the 2X2 table. You see that if your study is negative, there is an 83% chance that that is a true negative and a 17% chance that it is a false negative.[iv]
Let’s get back to intermittent fasting and what to tell a patient. We have an intervention with a sensible mechanism. Most people will buy that it should work. We have clinical experience among doctors and some patients that it works. We have negative studies. The studies might tell us that it doesn’t work or that it works a little – just below the level of detection of the studies as they are powered. The study might also be wrong. But if your pre-study probability is 50% before you read the NEJM study it is probably 25% before you read the JAMA IM study, having been influenced by the former. In that case, the chance that this negative study is wrong has fallen to only 9%.
Take this as an example of the complexity of practicing evidence-based medicine.
So where do I come down on intermittent fasting. It is probably harmless. It probably has a small effect in most people (as well as a large effect in a few and no effect in a few). If someone is looking for a way to lose weight, short of a GLP1-RA or bariatric surgery, it seems a reasonable thing to try. It does not, however, to call for evangelism.
[i] To be honest, I’m a bit worried about the invective I am likely to absorb from them after posting this article.
[ii] I recognize that I am risking looking like a huckster here.
[iii] 1-beta, usually 80%, is the power of a study.
[iv] This assumes that everything else about the study is flawless and there is no publication bias…