Orlich MJ, Singh PN, Sabaté J, Jaceldo-Siegl K, Fan J, Knutsen S, Beeson WL, Fraser GE. Vegetarian Dietary Patterns and Mortality in Adventist Health Study 2. JAMA Intern Med. 2013;173(13):1230-1238. doi:10.1001/jamainternmed.2013.6473
Some evidence suggests vegetarian dietary patterns may be associated with reduced mortality, but the relationship is not well established.
To evaluate the association between vegetarian dietary patterns and mortality.
Prospective cohort study; mortality analysis by Cox proportional hazards regression, controlling for important demographic and lifestyle confounders.
Adventist Health Study 2 (AHS-2), a large North American cohort.
A total of 96 469 Seventh-day Adventist men and women recruited between 2002 and 2007, from which an analytic sample of 73 308 participants remained after exclusions.
Diet was assessed at baseline by a quantitative food frequency questionnaire and categorized into 5 dietary patterns: nonvegetarian, semi-vegetarian, pesco-vegetarian, lacto-ovo–vegetarian, and vegan.
Main Outcome and Measure
The relationship between vegetarian dietary patterns and all-cause and cause-specific mortality; deaths through 2009 were identified from the National Death Index.
There were 2570 deaths among 73 308 participants during a mean follow-up time of 5.79 years. The mortality rate was 6.05 (95% CI, 5.82-6.29) deaths per 1000 person-years. The adjusted hazard ratio (HR) for all-cause mortality in all vegetarians combined vs nonvegetarians was 0.88 (95% CI, 0.80-0.97). The adjusted HR for all-cause mortality in vegans was 0.85 (95% CI, 0.73-1.01); in lacto-ovo–vegetarians, 0.91 (95% CI, 0.82-1.00); in pesco-vegetarians, 0.81 (95% CI, 0.69-0.94); and in semi-vegetarians, 0.92 (95% CI, 0.75-1.13) compared with nonvegetarians. Significant associations with vegetarian diets were detected for cardiovascular mortality, noncardiovascular noncancer mortality, renal mortality, and endocrine mortality. Associations in men were larger and more often significant than were those in women.
Conclusions and Relevance
Vegetarian diets are associated with lower all-cause mortality and with some reductions in cause-specific mortality. Results appeared to be more robust in males. These favorable associations should be considered carefully by those offering dietary guidance.
The possible relationship between diet and mortality remains an important area of investigation. Previous studies have identified dietary factors associated with mortality. Those found to correlate with reduced mortality include nuts,1- 4 fruit,5,6 cereal fiber,2 polyunsaturated fatty acids (PUFAs),2 ω-3PUFAs,3 green salad,7 Mediterranean dietary patterns,8- 11 “healthy” or “prudent” dietary patterns,10,12,13 plant-based diet scores,14 plant-based low-carbohydrate diets,15 and vegetarian diets.4,16,17 Associations with increased mortality have been found for a high glycemic load,2 meat,6,7 red meat,18,19 processed meat,18,19 eggs,7 potatoes,5 increased energy intake,20 and animal-based low-carbohydrate diets.15
Vegetarian dietary patterns may contain many of the above-listed foods and nutrients associated with reduced mortality while having reduced intakes of some foods associated with increased mortality. Vegetarian dietary patterns have been associated with reductions in risk for several chronic diseases, such as hypertension,21,22 metabolic syndrome,23 diabetes mellitus,24,25 and ischemic heart disease (IHD),17,26 which might be expected to result in lower mortality. Vegetarian diets represent common, real-world dietary patterns and are thus attractive targets for study.
Previous studies of the relationship between vegetarian dietary patterns and mortality have yielded mixed results. In the first Adventist Health Study, a study of 34 198 California Seventh-day Adventists,27 vegetarian dietary patterns were associated with reduced all-cause mortality and increased longevity.4,17 In contrast, the European Prospective Investigation into Cancer and Nutrition–Oxford (EPIC-Oxford) cohort study did not show an all-cause mortality advantage for British vegetarians (among 47 254 vegetarian and nonvegetarian participants),28 and pooled results have shown reductions only for IHD mortality.16
Our objective, in light of the potential benefits of vegetarian diets and the existing uncertainty in the literature, was to evaluate the possible association of vegetarian dietary patterns with reduced mortality in a large American cohort including many vegetarians.
Adventist Health Study 2 (AHS-2) is a cohort of 96 469 Seventh-day Adventist men and women recruited at churches in the United States and Canada between 2002 and 2007.29 Butler et al29 provided a detailed explanation of the cohort formation and characteristics. Written informed consent was obtained from all participants upon enrollment. The study was approved by the institutional review board of Loma Linda University.
Exclusions were applied in the following order: missing data for questionnaire return date, birth date, sex, or race (n = 1702); age younger than 25 years (n = 434); estimated energy intake (not including write-in items) less than 500 kcal/d or more than 4500 kcal/d; improbable response patterns (eg, identical responses to all questions on a page) or more than 69 missing values in dietary data (n = 4961); non-US residents (n = 4108); or history of a specific prior cancer diagnosis (except nonmelanoma skin cancers) or of cardiovascular disease (CVD) (coronary bypass, angioplasty/stent, carotid artery surgery, myocardial infarction, or stroke; or angina pectoris or congestive heart failure treated in the past 12 months) (n = 11 956). After exclusions, there remained an analytic sample of 73 308.
Mortality data through December 31, 2009, were obtained from the National Death Index. International Statistical Classification of Diseases, 10th Revision (ICD-10) codes for the underlying cause of death were used for causal classification. Unnatural causes of death (ICD-10 letters U, V, W, X, and Y) were considered as censoring events. Deaths associated with IHD were identified as ICD-10 I20-25; CVD deaths, as those starting with the letter I; and cancer deaths, as those starting with the letter C. Noncardiovascular, noncancer deaths were identified as all natural deaths not classified as CVD or cancer deaths. Infectious disease deaths were identified as those starting with the letters A or B; neurologic deaths, the letter G; respiratory deaths, the letter J; renal deaths, the letter N; and endocrine deaths, the letter E. Stroke deaths were identified using the code I60-69; diabetes mellitus deaths, E10-14; and renal failure deaths, N17-19.
Usual dietary intake during the previous year was assessed at baseline by a self-administered quantitative food frequency questionnaire of more than 200 food items. Dietary patterns were determined according to the reported intake of foods of animal origin. Thus, vegans consumed eggs/dairy, fish, and all other meats less than 1 time/mo; lacto-ovo–vegetarians consumed eggs/dairy 1 time/mo or more but fish and all other meats less than 1 time/mo; pesco-vegetarians consumed fish 1 time/mo or more but all other meats less than 1 time/mo; semi-vegetarians consumed nonfish meats 1 time/mo or more and all meats combined (fish included) 1 time/mo or more but no more than 1 time/wk; and last, nonvegetarians consumed nonfish meats 1 time/mo or more and all meats combined (fish included) more than 1 time/wk. For some analyses, the 4 vegetarian categories (vegan, lacto-ovo–vegetarian, pesco-vegetarian, and semi-vegetarian) were combined as “vegetarian.” The food frequency questionnaire was previously validated against six 24-hour dietary recalls for intake of nutrients30 and selected foods/food groups.31 Validity correlations for red meat, poultry, fish, dairy, and eggs were 0.76, 0.76, 0.53, 0.86, and 0.64, respectively, in whites and 0.72, 0.77, 0.57, 0.82, and 0.52, respectively, in blacks.31 Mean duration of adherence to dietary patterns was calculated for respondents to a follow-up questionnaire in which participants were asked to characterize their consumption of meat and dairy products at that time and in previous decades.
Other variables, all measured at baseline, were as follows (Table 1 footnotes for category specification): sex (dichotomous), race (dichotomous), geographic region (6 levels), personal income (4 levels), educational level (4 levels), marital status (dichotomous), smoking (8 levels), alcohol use (5 levels), exercise (ie, “vigorous activities, such as brisk walking, jogging, bicycling, etc, long enough or with enough intensity to work up a sweat, get your heart thumping, or get out of breath”) (5 levels), sleep (3 levels), menopausal status of women (dichotomous), hormone therapy in postmenopausal women (dichotomous), dietary energy (7 levels: <1000 kcal, 1000-1499 kcal,1500-1999 kcal, 2000-2499 kcal, 2500-2999 kcal, 3000-3999 kcal, and ≥4000 kcal), body mass index (calculated as weight in kilograms divided by height in meters squared) (9 levels: <18, 18 to <20, 20 to <23, 23 to <25, 25 to <27, 27 to <30, 30 to <35, 35 to <40, and ≥40). Race was included as a potentially important covariate. Participants self-identified their race/ethnicity in 1 or more of 21 categories. Those self-identifying as black/African American, West Indian/Caribbean, African, or other black were categorized as black for this analysis and all others were categorized as nonblack.
Baseline descriptive statistics were calculated according to the 5 dietary-pattern categories. Means and percentages were adjusted for age, sex, and race by direct standardization using the entire analytic sample as the standard distribution. Age-sex-race standardized mortality rates were computed by dietary pattern. Analyses of mortality were performed using Cox proportional hazards regression with attained age as the time variable and left truncation by age at study entry. Covariates were selected on an a priori basis as likely confounders based on prior studies and suspected relationships. Menopausal status and hormone therapy were represented in models as nested covariates (ie, sex + [sex × menopause] + [sex × menopause × hormone therapy]). Covariates were tested for possible interaction with the diet variable and for suspected interactions between selected covariates. The Cox proportional hazards assumption was evaluated using Schönfeld residuals, log(−log) plots, and attained-age interaction terms. Significant nonproportionality of hazards was present for race and marital status, so attained-age interaction terms for these variables were retained in the models. Residual methods were used to evaluate possible outliers and influential data points; no data points required removal. Multiple imputation of missing values was done for the small amount of missing data in the dietary variables used to calculate vegetarian status and for all covariates; a guided multiple-imputation approach was used when possible,32 as we have evidence that many of the missing dietary data are true zeroes.33 Analyses were performed using commercial software (SAS, version 9.3; SAS Institute, Inc). Guided multiple imputation was performed using R, version 2.13.1 software34 and the Hmisc package.35
Among the 73 308 individuals in our analytic sample, 5548 (7.6%) were vegans, 21 177 (28.9%) were lacto-ovo–vegetarians, 7194 (9.8%) were pesco-vegetarians, 4031 (5.5%) were semi-vegetarians, and 35 359 (48.2%) were nonvegetarians. Table 2 presents characteristics of the participants at baseline according to the 5 dietary patterns. Percentages and means were age-sex-race standardized as appropriate. Vegetarian groups tended to be older, more highly educated, and more likely to be married, to drink less alcohol, to smoke less, to exercise more, and to be thinner. The proportion of blacks was highest among pesco-vegetarians and lowest in lacto-ovo–vegetarians. Of postmenopausal women, far fewer vegans were receiving hormone therapy. Mean reported duration of adherence to current dietary pattern (not included in Table 2) was 21 years for vegans, 39 years for lacto-ovo–vegetarians, 19 years for pesco-vegetarians, 24 years for semi-vegetarians, and 48 years for nonvegetarians.
The mean (SD) follow-up time was 5.79 (1.31) years. During this time, there were 2570 deaths among 73 308 participants, and the overall mortality rate was 6.05 (95% CI, 5.82-6.29) deaths per 1000 person-years. Table 3 gives the age-sex-race standardized mortality rates by dietary pattern. Vegans, lacto-ovo–vegetarians, and pesco-vegetarians had significantly lower mortality rates compared with nonvegetarians.
Table 1 reports the comparison of multivariate-adjusted risk of death for all vegetarians combined with that for nonvegetarians. Vegetarians had 0.88 (95% CI, 0.80-0.97) times the risk of all-cause mortality of nonvegetarians. In men, the hazard ratio (HR) was 0.82 (95% CI, 0.72-0.94) and in women, 0.93 (0.82-1.05). Significantly reduced risk in both sexes combined was also seen for other mortality (ie, non-CVD, noncancer) (HR, 0.85; 95% CI, 0.73-0.99) but not clearly for IHD mortality (0.81; 0.64-1.02), CVD mortality (0.87; 0.75-1.01), or cancer mortality (0.92; 0.78-1.08). For men, CVD mortality (0.71; 0.57-0.90) and IHD mortality (0.71; 0.51-1.00) achieved significance, and other mortality had a notable but nonsignificant reduction (0.83; 0.66-1.04). In women, there were no significant reductions in these causal categories of mortality, although the effect estimates for IHD mortality, cancer mortality, and other mortality were moderately less than 1.0. Results (not included in table) for stroke were, for both sexes combined, HR, 1.10 (95% CI, 0.82-1.47); for men, 0.83 (0.52-1.31); and for women, 1.27 (0.89-1.80).
Table 4 reports the comparison of the multivariate-adjusted risk of death for 4 categories of vegetarians compared with nonvegetarians. Pesco-vegetarians had significantly reduced risk in both sexes combined for all-cause mortality (HR, 0.81; 95% CI, 0.69-0.94), IHD mortality (0.65; 0.43-0.97), and other mortality (0.71; 0.54-0.94); in men for all-cause mortality (0.73; 0.57-0.93), CVD mortality (0.66; 0.44-0.98), and other mortality (0.60; 0.39-0.93); and in women for IHD mortality (0.51; 0.26-0.99). Lacto-ovo–vegetarians had significantly reduced risk in both sexes combined for all-cause mortality (HR, 0.91; 95% CI, 0.82-1.00) and in men for CVD mortality (0.77; 0.59-0.99). Vegans had significantly reduced risk in both sexes combined for other mortality (HR, 0.74; 95% CI, 0.56-0.99) and in men for all-cause mortality (0.72; 0.56-0.92), IHD mortality (0.45; 0.21-0.94), and CVD mortality (0.58;0.38-0.89).
Table 5 presents the results of multivariate-adjusted Cox analyses for several more-specific categories of mortality within the broad “other” mortality of Table 1 (ie, non-CVD, noncancer mortality), comparing all vegetarians with nonvegetarians. In men and women combined, vegetarians had a significantly reduced risk of renal mortality (HR, 0.48; 95% CI, 0.28-0.82) and endocrine mortality (0.61; 0.40-0.92); in men, vegetarians had reduced risk of renal mortality (0.42; 0.19-0.91) and endocrine mortality (0.48; 0.25-0.92); and in women, nonsignificant reductions for both renal mortality (0.57; 0.28-1.19) and endocrine mortality (0.76; 0.44-1.30). Forty of 67 renal deaths were associated with renal failure (for both sexes combined, HR, 0.26; 95% CI, 0.12-0.57; for women, 0.39; 0.13-1.17; and for men, 0.21; 0.07-0.63). Sixty-seven of 104 endocrine deaths were associated with diabetes mellitus (for both sexes combined, HR, 0.53; 95% CI, 0.32-0.89; for women, 0.78; 0.41-1.48; and for men, 0.27; 0.11-0.66).
A sensitivity analysis in which body mass index was added to the model generally had only a modest effect on the results. Overall HRs for vegetarians were then 0.90 (95% CI, 0.82-0.98) for both sexes combined, 0.83 (0.72-0.96) for men, and 0.95 (0.84-1.06) for women. The adjustment for body mass index did not consistently move results toward the null. Mortality results adjusted for body mass index affected statistical significance in the following instances. For all vegetarians combined compared with nonvegetarians: IHD mortality in men (HR, 0.77; 95% CI, 0.54-1.10), endocrine mortality in both sexes combined (HR, 0.71; 95% CI, 0.46-1.09), and diabetes mortality in both sexes combined (HR, 0.65; 95% CI, 0.38-1.11). For specific vegetarian dietary patterns compared with nonvegetarians: vegans, all-cause mortality in both sexes combined (HR, 0.84; 95% CI, 0.72-1.00) and IHD mortality in men (0.50; 0.24-1.06); lacto-ovo–vegetarians, all-cause mortality in both sexes combined (0.92; 0.84-1.02) and CVD mortality in men (0.81; 0.63-1.05); pesco-vegetarians, IHD mortality in both sexes combined (0.69; 0.45-1.05), other mortality in both sexes combined (0.77; 0.60-1.00), CVD mortality in men (0.68; 0.45-1.04), and other mortality in men (0.65; 0.43-1.00). Additional adjustment by dietary energy intake resulted in negligible changes. Formal tests for interaction of the diet variable (vegetarian vs nonvegetarian) with sex revealed significant interaction for CVD mortality (P = .01), but no significant interaction for all-cause mortality or other categories of mortality.
These results demonstrate an overall association of vegetarian dietary patterns with lower mortality compared with the nonvegetarian dietary pattern. They also demonstrate some associations with lower mortality of the pesco-vegetarian, vegan, and lacto-ovo–vegetarian diets specifically compared with the nonvegetarian diet.
Some associations of vegetarian diets with lower cardiovascular mortality and lower noncardiovascular, noncancer mortality were observed. Vegetarian diets have been associated with more favorable levels of cardiovascular risk factors,17,22- 25,36,37 and nutrient profiles of the vegetarian dietary patterns suggest possible reasons for reduced cardiovascular risk, such as lower saturated fat and higher fiber consumption.38 Analysis within the non-CVD, noncancer category revealed notable reductions in mortality with underlying cause classified as endocrine or renal (diabetes mellitus and renal failure, in particular). These apparent protective associations seem consistent with previously published findings showing an association of vegetarian diets with reduced risk of incident diabetes25 and of prevalent diabetes, hypertension, and metabolic syndrome.21,23,24
No significant associations with reduced cancer mortality were detected. The heterogeneous nature of cancer may obscure specific diet-cancer associations in analyses of combined cancer mortality, and lack of significance may reflect insufficient power to detect weaker associations at early follow-up. Early analyses of vegetarian dietary patterns and cancer incidence in AHS-2 demonstrated significantly reduced risks of female-specific and gastrointestinal cancers.39
Effects were generally stronger and more significant in men than women. Previous studies40- 42 among Adventists have demonstrated effect modification by sex of the association of vegetarian diets with reduced ischemic heart disease mortality. It is possible that within dietary groups the diets of men and women differ in important ways; however, a recent evaluation38 of the nutrient profile of the dietary patterns in this cohort did not reveal striking differences. Alternatively, the biological effect of dietary factors on mortality may be different in men and women. Future analysis will evaluate possible effect modification by sex for particular foods or nutrients, which may suggest sex-specific mechanisms.
Strengths of this study include the large number of participants consuming various vegetarian diets; the diverse nature of this cohort in terms of sex, race, geography, and socioeconomic status, enhancing generalizability; the low use of tobacco and alcohol, making residual confounding from these unlikely; the shared religious affiliation of the cohort, which may lead to greater homogeneity across several possible unmeasured confounders, enhancing internal validity; and precise dietary pattern definitions based on measured food intake rather than self-identification of dietary patterns.
This analysis is limited by relatively early follow-up. If dietary patterns affect mortality, they may do so with moderate effect sizes, via complex pathways, and with long latency periods. Early follow-up analysis may thus have bias toward the null, and true associations may remain undetected. Observed mortality benefits may be affected by factors related to the conscious lifestyle choice of a vegetarian diet other than dietary components. Potential for uncontrolled confounding remains. Dietary patterns may change over time, whereas the analysis relies on a single measurement of diet at baseline. Caution must be used in generalizing results to other populations in which attitudes, motivations, and applications of vegetarian dietary patterns may differ; dietary pattern definitions used may not reflect some common uses of these terms.
Further study of the possible association with mortality of specific foods and nutrients that characterize the different diet-pattern groups is a major future goal of the AHS-2 study. Later follow-up may yield more statistically robust results; allow direct comparisons between vegetarian groups and enable subgroup analysis, particularly by race/ethnicity; and allow for analysis by more specific causes of mortality.
The lack of similar findings in British vegetarians28 remains interesting, and this difference deserves careful study. In both cohorts, the nonvegetarians are a relatively healthy reference group. In both studies, the nutrient profiles of vegetarians differ in important ways from those of nonvegetarians, with vegetarians (especially vegans) consuming less saturated fat and more fiber.38,43 It appears that British vegetarians and US Adventist vegetarians eat somewhat differently.44 For instance, the vegetarians in our study consume more fiber and vitamin C than those of the EPIC-Oxford cohort: mean dietary fiber in EPIC-Oxford vegans was 27.7 g/d in men and 26.4 g/d in women compared with 45.6 g/d in men and 47.3 g/d in women in AHS-2 vegans; mean vitamin C in EPIC-Oxford vegans was 125 mg/d in men and 143 mg/d in women compared with 224 mg/d in men and 250 mg/d in women in AHS-2 vegans.38,43 Individuals electing vegetarian diets for ethical or environmental reasons may eat differently from those who choose vegetarian diets primarily for reasons of perceived superiority for health promotion. We believe that perceived healthfulness of vegetarian diets may be a major motivator of Adventist vegetarians. More important, other large cohort studies have linked increased red and processed meat consumption to higher mortality,18,19,45 and our findings build on this work by demonstrating reduced mortality in those consuming low-meat dietary patterns. Notably, the findings of the present study are similar to those of prior North American Adventist cohorts, demonstrating a consistent association over several decades and replicating prior results in a population with greater geographic and ethnic diversity.46
In conclusion, in a large American cohort, we found that vegetarian dietary patterns were associated with lower mortality. The evidence that vegetarian diets, or similar diets with reduced meat consumption, may be associated with a lower risk of death should be considered carefully by individuals as they make dietary choices and by those offering dietary guidance.
Corresponding Author: Michael J. Orlich, MD, Adventist Health Studies, School of Public Health, 24951 N Circle Dr, NH 2031, Loma Linda University, Loma Linda, CA 92350 (firstname.lastname@example.org)
Accepted for Publication: February 23, 2013.
Published Online: June 3, 2013. doi:10.1001/jamainternmed.2013.6473
Author Contributions: Drs Orlich and Fraser had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Orlich, Jaceldo-Siegl, and Fraser.
Acquisition of data: Jaceldo-Siegl, Knutsen, Beeson, and Fraser.
Analysis and interpretation of data: Orlich, Singh, Sabaté, Fan, and Knutsen.
Drafting of the manuscript: Orlich, Singh, and Jaceldo-Siegl.
Critical revision of the manuscript for important intellectual content: Singh, Sabaté, Fan, Knutsen, Beeson, and Fraser.
Statistical analysis: Orlich, Singh, Fan, Beeson, and Fraser.
Obtained funding: Fraser.
Administrative, technical, and material support: Knutsen and Beeson.
Study supervision: Jaceldo-Siegl and Fraser.
Conflict of Interest Disclosures: Dr Orlich reports receiving a small honorarium from the Northern California Conference of Seventh-day Adventists to partially defray travel expenses for a speaking engagement at which he gave an overview and update of Adventist Health Studies research and a small honorarium from the Southern California Conference of Seventh-day Adventists for a speaking engagement at which he lectured on lifestyle approaches for chronic disease prevention.
Funding/Support: Project support was obtained from National Cancer Institute (NCI) grant 1U01CA152939 (Dr Fraser). Dr Orlich’s research fellowship was supported by grant 2010-38938-20924 from the National Institute of Food and Agriculture (NIFA).
Disclaimer: The views expressed in this article are those of the authors and do not necessarily represent the views of the NCI or NIFA. The ideas and opinions expressed herein are those of the authors, and endorsement by the NCI, NIFA, or their contractors or subcontractors is not intended nor should it be inferred.
Previous Presentation: These findings were presented in a brief oral presentation at the Society for Epidemiologic Research annual meeting; June 28, 2012; Minneapolis, Minnesota.
Additional Contributions: Hanni Bennett, Research Associate, Adventist Health Studies, School of Public Health, Loma Linda University, provided invaluable administrative support for AHS-2.