What Eggsactly Are We Asking Here? Unscrambling the Epidemiology of Eggs, Cholesterol, and Mortality

HomeCirculationVol. 145, No. 20What Eggsactly Are We Asking Here? Unscrambling the Epidemiology of Eggs, Cholesterol, and Mortality Free AccessEditorialPDF/EPUBAboutView PDFView EPUBSections ToolsAdd to favoritesDownload citationsTrack citationsPermissions ShareShare onFacebookTwitterLinked InMendeleyReddit Jump toFree AccessEditorialPDF/EPUBWhat Eggsactly Are We Asking Here? Unscrambling the Epidemiology of Eggs, Cholesterol, and Mortality Deirdre K. Tobias, ScD Deirdre K. TobiasDeirdre K. Tobias Correspondence to: Deirdre K. Tobias, ScD, 900 Commonwealth Ave, 3rd Floor, Boston, MA 02215. Email E-mail Address: [email protected] https://orcid.org/0000-0003-0020-9552 Division of Preventive Medicine, Department of Medicine, Brigham and Women’s Hospital and Harvard Medical School, Boston, MA. Nutrition Department, Harvard TH Chan School of Public Health, Boston, MA. Search for more papers by this author Originally published16 May 2022https://doi.org/10.1161/CIRCULATIONAHA.122.059393Circulation. 2022;145:1521–1523This article is a commentary on the followingAssociations of Dietary Cholesterol, Serum Cholesterol, and Egg Consumption With Overall and Cause-Specific Mortality: Systematic Review and Updated Meta-AnalysisEstablishing dietary guidelines relies on consensus across various sources of evidence, and substantial heterogeneity can be problematic. Critics of nutritional epidemiology often point to examples of inconsistent dietary exposure/outcome associations as proof against the validity and utility of observational cohorts. For example, a commentary once plotted the results from several cohort studies for associations of a variety of foods, including eggs, with cancer risk to illustrate the extent of conflicting conclusions derived from nutritional epidemiology.1 It seemed almost completely random whether a study concluded eggs were protective or harmful. Although between-study differences in measurement errors and statistical adjustment for confounders can lead to meaningful heterogeneity, inconsistencies between their specific causal questions implicitly asked by the analytic frameworks may have an even greater effect.Article, see p 1506In the study by Zhao et al,2 the authors revisit the hypothesis “that impairment of cholesterol metabolism can be involved in the development of chronic diseases including cardiovascular disease (CVD) and cancer.” They comprehensively evaluate dietary cholesterol from all sources, eggs (as their population’s top dietary cholesterol source, ≈44%), and serum cholesterol in relation to all-cause and cause-specific mortality. In brief, the authors analyzed data from the ATBC study (Alpha-Tocopherol, Beta-Carotene Cancer Prevention) cohort of Finnish male smokers enrolled in the 1980s with demographic, cardiovascular, and health history, self-reported diet at baseline, and long-term follow-up for deaths and other events. They observed that the men who consumed higher amounts of cholesterol in their diet at baseline were more likely to die over the next 30 years, with each additional 300 mg/d associated with a 10% higher risk of death from all causes (hazard ratio [HR], 1.10 [95% CI, 1.06–1.13]). Likewise, men who ate more eggs at baseline had a 6% greater risk of death from all causes for each egg/d (HR, 1.06 [95% CI, 1.04–1.09]). Findings were comparable but modest for deaths from cardiovascular causes (total dietary cholesterol HR, 1.13 [95% CI, 1.08–1.18]; eggs HR, 1.09 [95% CI, 1.05–1.12]) although moderately attenuated for cancer deaths (total dietary cholesterol HR, 1.06 [95% CI, 1.01–1.12]; eggs HR, 1.04 [95% CI, 1.00–1.08]). Their analyses were carefully adjusted for potential confounding risk factors. Thus, the authors reasonably conclude that “greater consumption of dietary cholesterol and eggs were associated with an increased risk of overall and CVD mortality.”Eggs Compared With What?Had these results for eggs and mortality risk come from a randomized dietary intervention trial, one might immediately ask, “What did the control group eat in place of eggs?” because it is reasonable to question whether contrasting eggs versus cinnamon rolls or steel cut oatmeal with fruit would give different magnitudes, and possibly even directions (ie, benefit versus risk) of eggs’ relative effect on mortality risk. The same can be asked from observational analyses; in other words, cholesterol and eggs compared with what? It is helpful to imagine that the results were from a controlled feeding trial. We can define the hypothetical dietary interventions that reflect the variables included in a multivariable-adjusted regression model.First, consistent with feeding studies, Zhao et al appropriately controlled for total energy intake in their models; otherwise, simply adding eggs to the intervention group’s diet (and making no adjustments to the control diet) results in meaningful increases in daily calorie intake, subsequent weight gain, and metabolic consequences in one intervention group only, obscuring any effects of the eggs per se on the mortality rate.3 Second, given the reality that these are observational data, the authors are very much aware of the fact that men who self-select to habitually eat more eggs and other sources of dietary cholesterol will have diets that differ in many other ways, and they address this potential confounding by adjusting for percentage of energy from carbohydrate, protein, and fatty acids.We then interpret the multivariable-adjusted models as the contrast between men with higher intake of cholesterol-containing foods or eggs versus men consuming an equal amount of energy from cholesterol-free (Table, Experiment 1) or nonegg (Table, Experiment 2) foods, matched overall for their total energy and macronutrient ratios. When the substitution food(s) is not explicitly defined in the hypothesis or the model, the comparator diet is technically the statistical average of mortality risk across all nonintervention food calories, weighted by their prevalence in the population’s diet, and some clues to what this may consist of can be found in Table 1 in the article by Zhao et al. The authors model egg intake again, but with further adjustment for total dietary cholesterol. This now creates a substitution interpretation between eggs and all other cholesterol-containing foods consumed in this population (Table, Experiment 3). Thus, we can improve this interpretation for eggs and mortality as a substitution for butter, milk, sausage, and other foods.Table. Interpreting Dietary Exposure and Comparators From Nutritional Epidemiology Regression ModelsExperimentIntervention foodsComparator foodsOverall (background) dietsAll-cause mortality hazard ratio (95% CI)Cardiovascular disease mortality hazard ratio (95% CI)Cancer mortality hazard ratio (95% CI)1Cholesterol-containing foods contributing 300 mg/dEqual calories of cholesterol-free foods contributing 0 mg/dMatched for total energy and macronutrient ratios1.10 (1.06–1.13)1.13 (1.08–1.18)1.06 (1.01–1.12)21 egg/dEqual calories of any nonegg foodMatched for total energy and macronutrient ratios1.06 (1.04–1.09)1.09 (1.05–1.12)1.04 (1.00–1.08)31 egg/dEqual calories of nonegg cholesterol-containing food (butter, milk, sausage, fish, pork, cheese, beef, other)Matched for total energy and macronutrient ratios0.91 (0.84–0.99)0.92 (0.81–1.04)0.93 (0.81–1.08)With this useful feeding trial framework, we can conclude, as the authors did, that men eating high-cholesterol foods and eggs had a higher risk of death than the men consuming more calories from the typical background diet. However, when the model explicitly contrasted eggs against higher amounts of dairy, processed meats, and other high-cholesterol foods, men eating more eggs and fewer of these other products actually had a lower risk of death. The comparator diet clearly matters very much, even reversing conclusions altogether from eggs being harmful to beneficial. Collectively, it appears that, in this population, eggs themselves are not cause for alarm, but rather the other cholesterol-containing foods popular at baseline among these men (ie, butter, milk, and sausage).Meta-Analysis Illustrates the Effect of Substitutions in Nutritional EpidemilogyAs we earlier saw with the authors’ thorough cohort analysis, the impact of multivariable adjustment for interpretation of nutritional exposures cannot be understated. Eggs “compared with what” mattered very much for the direction of egg’s association with mortality. The importance of understanding the cohort-specific food substitutions is visible in the forest plot of the updated meta-analysis. In their publication, Zhao et al include an extensive updated meta-analysis, reporting the updated relative risk for each additional egg per day with CVD mortality from 21 prospective cohorts is 1.05 (0.96–1.13). However, they note that this combined estimate is overshadowed by significant between-study heterogeneity (I2=83.9%). Unfortunately, none of the individual studies themselves modeled an explicit substitution for egg with some other food, and multivariable adjustment for dietary factors, even total energy, varied drastically from cohort to cohort. This precludes Zhao et al from being able to specify the comparator diet(s) in their meta-analysis of egg intake. Further, the global representation of cohorts implies that those participants with higher intake of eggs are likely being contrasted with a range of different background diets. In fact, Zhao et al report that the relationship between eggs and CVD varied significantly by geographical region, showing no relationship for higher egg intake and CVD in Asia, but a positive association in the United States, further supporting our speculation that differences in the comparator diets may explain heterogeneity for the associations of eggs with CVD end points.ConclusionsStatistical regression models of observational data are ultimately agnostic to whether the combination of dietary variables results in a meaningful substitution, and clarifying the comparator food(s) of interest can help overcome the confusion from past nutritional epidemiology research. As such, there has been a notable shift in recent years toward generating hypotheses that include well-defined substitution exposures,4 and analytic frameworks are emerging to accommodate this.5 We also caution that meta-analyses may be improved by appropriately comparing apples with apples, not only for the exposure and comparator foods but for other critical study features, including the outcome of interest, follow-up time, and population characteristics. These measures may hopefully mitigate between-study heterogeneity and improve consistency across the evidence base. Overall, we are optimistic that the heterogeneity that is commonplace across nutritional epidemiology research, as the authors highlight herein, can be tamed with careful attention to well-defined substitutions.After unscrambling the evidence, what do these data allow us to conclude for guidelines around intake of dietary cholesterol and eggs for all-cause or CVD mortality? This inclusive analysis by Zhao et al suggests that the answer depends on what the comparator foods are. On the individual level, if we chose to reduce dietary cholesterol then we are ultimately increasing intake of some other food. Likewise, choosing to eat more eggs means inherently skipping something we could have eaten instead. If the foods we eat more or less of have their own relationship with mortality or CVD death, whether harmful or preventive, then eggs may look better or worse in contrast. In the ATBC cohort, eggs fared better than butter and meat but worse than the average background diet of these Finnish men. This suggests that other aspects of these foods, rather than their dietary cholesterol content per se, may be related to CVD risk, although studies that intervene on the nutrient level are warranted. Practically, such explicit substitutions between foods are often below our conscious decision-making. When making efforts to eat less of a certain food or nutrient, it is important to maintain awareness of what we are inadvertently eating more of, because increasing even less healthful options, in this case meat or butter, may have important implications for long-term health.Article InformationSources of FundingDr Tobias receives grant funding from the National Institutes of Health and American Diabetes Association for research not related to this article.Disclosures None.FootnotesThe opinions expressed in this article are not necessarily those of the editors or of the American Heart Association.For Sources of Funding and Disclosures, see page 1523.Circulation is available at www.ahajournals.org/journal/circCorrespondence to: Deirdre K. Tobias, ScD, 900 Commonwealth Ave, 3rd Floor, Boston, MA 02215. Email [email protected]harvard.eduReferences1. Schoenfeld JD, Ioannidis JP. Is everything we eat associated with cancer? A systematic cookbook review.Am J Clin Nutr. 2013; 97:127–134. doi: 10.3945/ajcn.112.047142CrossrefMedlineGoogle Scholar2. Zhao B, Gan L, Graubard BI, Mannisto S, Albanes D, Huang J. Associations of dietary cholesterol, serum cholesterol, and egg consumption with overall and cause-specific mortality: systematic review and updated meta-analysis.Circulation. 2022; 145:1506–1520. doi: 10.1161/CIRCULATIONAHA.121.057642LinkGoogle Scholar3. Willett W. Nutritional Epidemiology. Oxford Scholarship Online; 2013.Google Scholar4. Gardner C. “Instead of what,” and repeated 4-year interval change regarding red meat and T2D: increasing causal inference in nutritional epidemiology through methodological advances.Am J Clin Nutr. 2021; 113:497–498. doi: 10.1093/ajcn/nqaa385CrossrefMedlineGoogle Scholar5. Ibsen DB, Laursen ASD, Würtz AML, Dahm CC, Rimm EB, Parner ET, Overvad K, Jakobsen MU. Food substitution models for nutritional epidemiology.Am J Clin Nutr. 2021; 113:294–303. doi: 10.1093/ajcn/nqaa315CrossrefMedlineGoogle Scholar Previous Back to top Next FiguresReferencesRelatedDetailsRelated articlesAssociations of Dietary Cholesterol, Serum Cholesterol, and Egg Consumption With Overall and Cause-Specific Mortality: Systematic Review and Updated Meta-AnalysisBin Zhao, et al. Circulation. 2022;145:1506-1520 May 17, 2022Vol 145, Issue 20 Advertisement Article InformationMetrics © 2022 American Heart Association, Inc.https://doi.org/10.1161/CIRCULATIONAHA.122.059393PMID: 35576317 Originally publishedMay 16, 2022 Keywordsmortalitycholesterol, dietaryEditorialsdiet, food, and nutritioncardiovascular diseasesnutritional sciencesPDF download Advertisement SubjectsCardiovascular DiseaseDiet and NutritionEpidemiology