This week, a single egg is said to raise our risks for diabetes… not

By this evening, nearly 200 news stories have reported that a new Harvard study had found eggs raise risks for developing type 2 diabetes. Medical professionals could even earn continuing medical education credits by reading a MedPage Today article teaching that men eating seven or more eggs a week “were 58% more likely to develop type 2 diabetes than those who did not eat eggs, and women were 77% more likely to become diabetic if they ate at least an egg a day.” As the story made its way around the world, it became increasingly embellished, just like that game of telephone you may remember from childhood.

Readers of the Daily Mail in the UK read that doctors are now warning people to eat fewer egg and urging their diabetes patients to limit eggs to three a week. Australia’s Herald Sun even [incorrectly] claimed the study had shown that eggs make diabetes worse in those who already have diabetes.

The source of this latest egg scare was a study published online ahead of print publication in the journal, Diabetes Care. It was a data dredge using a massive database of nearly 60,000 male and female medical professionals. The authors from Brigham and Women's Hospital and Harvard Medical School in Boston looked for correlations between egg consumption and the onset of type 2 diabetes.

It was a null study. Try as they might, the authors were unable to find a single tenable link between the widely varied amounts of eggs these professionals had been eating and their risks for developing type 2 diabetes over 11-20 years of follow-up.

That’s not what the conclusion in the study’s abstract said, though. It stated: “These data suggest that high levels of egg consumption (daily) are associated with an increased risk of type 2 diabetes in men and women.” This must have misled medical reporters who didn’t understand epidemiological studies and hazard ratios, or who hadn’t read the actual study.

If they had, they would have realized in an instant that these correlations weren’t worth reporting as increased risks. Before revealing the punchline that illustrates how spurious these correlations were, like countless others that data dredges dredge up, we probably should pretend to take this study seriously.

Overview

The authors began by noting that the results of epidemiological studies have been inconsistent and suggested eggs are associated with both higher and lower risks for type 2 diabetes. Epidemiological studies and rat experiment have reported higher fat diets enriched with egg yolks were associated with higher fasting blood glucose levels, they noted, but a randomized clinical trial of fat people showed no effects.

Rather than conduct a randomized clinical trial to test this suggested link, the authors set out to conduct a large epidemiological study. They used the databases of two large clinical trials that had been designed to study aspirin or vitamins in the prevention of heart disease and cancer among mostly Caucasian adults: the Physicians’ Health Study (PHS-I) of 22,071 middle-age male physicians; and the Women’s Health Study (WHS) of 39,876 middle-age female healthcare professionals.

The PHS study was a randomized trial of aspirin and beta-carotene for the primary prevention of heart disease and cancer that was launched in 1982. All of the doctors were followed using periodic mail-in questionnaires. The trial was stopped early, but the database has become a rich source of more than 300 epidemiological papers reporting various correlations, such as finding that moderate alcohol consumption (one or more drinks a day) was associated with nearly half the risk for type 2 diabetes compared to nondrinkers over 12 years of observation.

The Women’s Health Study, begun in 1993, had been a randomized controlled trial of aspirin and vitamin E for the primary prevention of heart disease and cancer. It had similarly followed the women using questionnaires mailed out each year. Its database, compiled from those questionnaires, has also become a rich source of countless epidemiological studies. For example, one found no correlation between the women consuming the highest and lowest amounts of sugars and their risk for type 2 diabetes.

For this paper, the Boston authors eliminated participants in the databases who had preexisting diabetes, and missing data on egg consumption or other significant confounding factors (smoking, alcohol, BMI, exercise, hypertension, and fruits and vegetables), leaving 20,703 men and 36,295 women for their statistical analysis.

Dietary assessments had been done in both clinical trials, although the dietary information collected from the women had been more detailed and comprehensive, allowing good estimates of energy intake and nutrients, according to the authors. Diet information for the men was more limited. Egg consumption was self reported by the men using a simple abbreviated semi-quantitative food frequency questionnaire. The men were asked how often, on average, they had eaten one egg during the past year, choosing from 7 different amounts per day, week or month. This information was obtained at enrollment and 2, 4, 6, 8, and 10 years later. Among women, information on egg consumption was self-reported using a 131-item validated food frequency questionnaire at baseline. Women were asked to report their average consumption of one egg over the past year, with 9 possible responses. Because of the prospective design, where the participants had reported dietary intakes prior to developing diabetes, the authors noted that this lessened risks for recall bias.

To assess incidences of type 2 diabetes that occurred during the subsequent 11 to 20 years of follow-up (female healthcare professionals may find this interesting):

Follow-up and ascertainment of type 2 diabetes cases was completed in March 2007. Because all men were physicians, self-report was deemed sufficient. Among the female health professionals, self-reports of type 2 diabetes were validated using American Diabetes Association criteria, by obtaining additional information using telephone interview, supplemental questionnaire, or review of medical records from treating physicians. Overall, the positive predictive value for type 2 diabetes validation was 91%.

The men and women eating the most and least eggs differed in a variety of ways. The men eating the most eggs, for example, also consumed the most whole milk and fruits and vegetables, were almost 3 times more likely to smoke, fewer had high cholesterol, more had hypertension, while there was no difference in BMI or exercise levels. [Will the next data dredge report that whole milk or fruits and vegetables are linked to diabetes?]

According to the authors, “the large sample size, the long duration of follow ups, the repeated and standardized methods for data collection in both cohorts, the robustness of the findings in sensitivity analyses are major strengths of this paper.”

The findings

To describe their statistical methods briefly, after adjusting for traditional diabetes risk factors (such as age and family history for diabetes), they compared the odds ratios associated with type 2 diabetes between no eggs and the various categories of egg consumption (using Cox proportional hazard models). The odds varied from 9% to 58% among the men and -3% to 77% among the women (95% confidence interval).

They also reported that they found no correlation to the development of type 2 diabetes associated with high carbohydrate diets, high cholesterol levels or BMI.

Those odds ratios might sound relevant, but as we know, they aren’t tenable correlations or beyond what will crop up in a computer model by random change, statistical error and confounding factors. This was a null study. Null studies are good news. They show us that there isn't anything to worry about because there's not a credible link to even suggest our diet causes diabetes.

But what would have convinced readers of that and provided proof of how data dredges can pull up countless meaningless correlations that don't deserve our concern was found in the appendix available online. Among the various combinations examined, the odds ratios were haphazard and not related to the egg consumptions. The correlations went up, down, up, down and up as egg consumption increased.

Here’s the punchline. Want to know the biggest correlation these authors found? That women (fat and thin) eating high carbohydrate diets had about a 40% lower risk for type 2 diabetes associated with eating 5-6 eggs per week. Of course, this was an insignificant correlation, too, but interestingly, it didn’t make the news. No one would believe that eating all of those eggs and carbs could possibly be good for them. :-)

Instead, the findings that were reported appeared to support what is popularly believed to be unhealthy. The conclusions stated:

In this large prospective study, we have demonstrated that daily consumption of at least one egg is associated with an increased risk of type 2 diabetes in both men and women, independent of traditional risk factors for type 2 diabetes.

It didn’t really, but it made for a good news story.

© 2008 Sandy Szwarc