No, Intermittent Fasting Won’t Kill You

You probably saw the headlines that intermittent fasting increases cardiovascular mortality by 91%. Given the popularity of intermittent fasting, it was bound to be a click-generator. There are a few reasons why you should selectively ignore this story. In brief, intermittent fasting probably won’t double your chance of dying early. But there is at least one good reason to review this study in detail. It’s a great example of how important random chance is to the practice of medicine.

AHA Epi Lifestyle Conference

In case you missed it, the study in question was presented at the American Heart Association (AHA) Epidemiology and Prevention/Lifestyle conference in Chicago. Researchers analyzed two separate datasets to study the long-term implications of the diet that has become popular with the likes of Elon Musk, Jennifer Aniston, and Rishi Sunak. While we tend to call this diet trend “intermittent fasting,” we should probably refer to it more accurately as “time-restricted eating” as they do in the research. Many different versions of intermittent fasting have come and gone over the years, while the current flavor is to eat only at specific times of the day, hence the refinement to “time-restricted eating.”

Information about eating patterns was collected from the National Health and Nutrition Examination Survey (NHANES). As part of the survey, participants completed two separate food questionnaires about what they ate over the past 24 hours. Deaths were recorded in the Centers for Disease Control and Prevention’s National Death Index database, and researchers had, on average, 8 years’ worth of follow-up data on just over 20,000 participants.

Those who ate their daily meals within an 8-hour window (ie, they fasted for more than 16 hours a day) had more cardiovascular mortality than the control group, which ate during a 12- to 16-hour window (ie, they ate fairly consistently throughout the day). Comparing these two groups, the 16:8 dieters had a 91% increased risk for cardiovascular death.

Much of the early reporting was based on a press release put out by the AHA. Anyone wanting to look at the actual data had to find the conference abstract. Unfortunately, the numbers in the press release and the abstract didn’t match up. The press release quoted a 91% increase in cardiovascular mortality whereas the conference abstract posted online documented a hazard ratio of 1.96 (95% CI, 1.23-3.13).

A minor mathematical quibble before proceeding. “Risk” and “hazard” are not synonymous terms, and a hazard ratio of 1.96 does not actually mean that the risk for cardiovascular mortality has increased by 96%. However, setting that aside, the AHA did eventually upload a PDF of the poster presentation and it does appear as if the press release had the “correct” numbers. Or more likely, because the abstract was obviously submitted far in advance of the conference, it contained initial results that were subsequently revised and reanalyzed in the weeks, days, if not hours before the poster presentation. Unfortunately, this is common and there are often discrepancies between what is submitted, what is presented, and what is ultimately published.

Many people have since commented on how the press coverage was based on unpublished, non–peer-reviewed results. It is fair to criticize media outlets for highlighting such preliminary research but also the AHA for promoting it. There is nothing wrong with poster presentations and preliminary research, but there should have been more circumspection before presenting it for public consumption.

Diet Quality and Multiple Analyses

Beyond these surface-level objections were some more substantive issues. The study was, by nature, observational and therefore prone to confounding. Researchers adjusted for relevant variables, but as we have learned multiple times in the past, residual confounding is always a potential issue. Also, researchers didn’t have information on the quality of the study participants’ diet. If someone breaks their 16-hour fast with ultraprocessed junk food, the type of food might be more relevant to long-term cardiovascular health than when it was eaten. People may have engaged in intermittent fasting because they had shift work or an irregular sleep schedule, which are independent cardiovascular risk factors. It’s also possible that people adopted time-restricted eating as a dietary strategy precisely because they were at high risk for cardiovascular disease, an association that could be explained by reverse causation. Suffice it to say, many other external factors might be at play here.

Also, dietary patterns were assessed using two 24-hour questionnaires from NHANES. Dietary questionnaires are notoriously unreliable and subject to the poor recall of participants. But here the problem is more marked. Dietary patterns, especially those subject to the whims of pop culture, will wax and wane in popularity. Compliance with most diets is a problem. And using a single time point (actually, the average of two time points) to assess diet, which by nature changes and fluctuates with time, will not be a true reflection of someone’s overall diet.

One of the most important factors (and the one seldom discussed) is the play of random chance. Until the poster was uploaded for all to see, the press release did not make clear exactly how many analyses were done in this project. There were 36 statistical analyses in this paper.

Participants were not divided into an intermittent-fasting group and a control group; they were divided into subgroups based on eating duration (< 8 hours, 8-10 hours, 10-12 hours, the reference group of 12-16 hours, and > 16 hours). Each subgroup was tested against three outcomes: cardiovascular mortality, cancer mortality, and all-cause mortality. Finally, the researchers looked at the overall patient population, the subgroup of people with preexisting cardiac disease, and the subgroup of people with cancer. In this multiplicity of analyses, most were negative. Fasting for more than 16 hours was associated with higher cardiovascular (but not overall or cancer-related) mortality. This finding generated all the headlines.

But eating over more than 16 hours per day (which I suppose means you stop eating only to sleep) was associated with substantially less cancer mortality, if you had a preexisting diagnosis of cancer. No one seems to have mentioned this, probably because gorging yourself throughout the day is unlikely to be a fruitful pathway to disease prevention.

To summarize, most analyses were negative. There were isolated signals of harm and benefit with the extreme subgroups of the patient population, whereas there was no effect for the majority of the patient population that had fairly conventional dietary patterns.

You might be tempted to concoct some elaborate explanation or justification for why this might be true. But we should acknowledge that multiple hypothesis testing is probably at play. For those who remember the ISIS-2 trial, it demonstrated the benefits of aspirin use in the setting of an acute myocardial infarction. Importantly, those benefits varied by astrological sign (the benefits of aspirin did not extend to Geminis and Libras). This analysis was included deliberately to demonstrate the dangers of multiple hypothesis testing because if you do enough analyses, you will get a positive result purely by chance.

Hold the Presses

We are not great at identifying the impact of random chance in medical research. When we see an unexpected result, we tend to think “groundbreaking,” not “outlier.” And this is probably the case here.

Notwithstanding all the other critiques about the preliminary nature of the analysis, the inaccuracy of dietary questionnaires, and the issue of residual confounding, this is fundamentally a null study. Will the statistically significant results survive the drafting of a manuscript and peer review? Can they be replicated by other groups in other datasets? These are the relevant questions that should have been asked before the media coverage brought this study to the attention of the general public.

There is nothing wrong with poster presentations, preliminary research, and conference abstracts, but they probably shouldn’t make the six o’clock news. We can wait to be certain before we announce them to the world.

Christopher Labos is a cardiologist with a degree in epidemiology. He spends most of his time doing things that he doesn’t get paid for, like research, teaching, and podcasting. Occasionally he finds time to practice cardiology to pay the rent. He realizes that half of his research findings will be disproved in 5 years; he just doesn’t know which half. He is a regular contributor to the Montreal Gazette, CJAD radio, and CTV television in Montreal and is host of the award-winning podcast The Body of Evidence.