Nut Consumption and Risk of Cardiovascular Disease and Ischemic Heart Disease Mortality: The Adventist Health Study 2

Abstract

Background

Cardiovascular disease (CVD) and ischemic heart disease (IHD) are leading causes of mortality worldwide. Nut consumption has been associated with reduced risk.

Objectives

To examine the associations between total nut and tree nut intake with CVD and IHD mortality, and to explore the potential benefits of substituting nuts for less heart-healthy foods.

Methods

This prospective cohort study followed 80,529 Adventist participants for an average of 11.1 y. During this period, 4258 CVD deaths, including 1529 IHD deaths, were recorded. Dietary intake was assessed using a validated food frequency questionnaire. We examined total nut intake (combining tree nuts, peanuts, and peanut butter) as well as tree nut intake specifically (including almonds, cashews, walnuts, and mixed nuts). Cox proportional hazards models were used to estimate the hazard ratios (HRs) and 95% confidence intervals (CIs), adjusting for demographic, dietary, and lifestyle factors.

Results

Participants with higher nut intake tended to be older, more educated, and reported healthier lifestyle behaviors. After multivariable adjustment, higher total nut intake was significantly associated with lower CVD (HR: 0.86; 95% CI: 0.79, 0.94) and IHD mortality (HR: 0.81; 95% CI: 0.70, 0.94) when comparing the 90th with the 10th percentile of intake. Stronger inverse associations were observed with tree nut intake and CVD (HR: 0.83; 95% CI: 0.74, 0.92) and IHD mortality (HR: 0.73; 95% CI: 0.62, 0.87). Substitution analysis revealed that replacing unprocessed red meats and processed meats with total nuts and tree nuts was associated with lower risk of CVD and IHD mortality.

Conclusions

Higher intakes of total nuts and tree nuts are associated with significantly lower risk of CVD and IHD mortality. These findings support dietary recommendations to increase nut consumption as part of a heart-healthy diet.

Introduction

Cardiovascular disease (CVD) continues to be the leading cause of death in the United States [1] and globally [2]. Ischemic heart disease (IHD) is a component of CVD with high prevalence [3,4]. Improving lifestyle habits and diet quality may prevent CVD and IHD. Nuts have been suggested as an important component of a healthy diet in the United States [5] and other countries [6,7] because they are nutrient-dense foods rich in unsaturated fatty acids, proteins, fibers, phenolic compounds, vitamins, and minerals [8,9]. These nutrient components have been reported to have multiple mechanistic roles that may contribute to reduced risk of CVD and IHD mortality [10,11].

Interest in nuts for CVD prevention emerged in 1992 after Fraser et al. [12] reported a lower risk of fatal coronary artery disease and myocardial infarction among frequent nut consumers in the Adventist Health Study. Subsequent intervention studies [13] demonstrated significant reductions in total and LDL cholesterol from consuming specific types of nuts, such as walnuts and almonds. Since then, considerable evidence drawn from epidemiological studies [[14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24]] and clinical trials [25] has consistently reported that higher nut intake is associated with lower risk CVD and IHD mortality. These findings are further supported by meta-analyses of prospective cohort studies [[26], [27], [28]], which highlight the existence of a dose-response relationship, where greater consumption of nuts is associated with lower risk of CVD and IHD mortality.

Most prospective studies of nut consumption report CVD rather than IHD mortality and focus on total nut intake [15,16,[19], [20], [21],24,29,30], which includes a heterogeneous composition of various tree nuts, peanuts, and peanut butter. Fewer studies [20,23,29,31,32] have examined tree nuts specifically. Although several United States-based cohorts have reported significant inverse associations [12,[19], [20], [21],24,29], most international studies have reported null findings [15,16,23,30,31], likely reflecting differences in sample size, event rates, follow-up durations, variability in nut intake, and residual confounding from baseline cardiometabolic risk factors, including higher consumption of red and processed meats, alcohol, and smoking. Few prospective studies [22,[33], [34], [35]] have performed substitution analyses, limiting insight as to how these foods contribute to broader diet types. To address these gaps, we prospectively examined total nut and tree nut consumption in relation to CVD (including cerebrovascular) mortality, and specifically IHD mortality, in the Adventist Health Study 2 (AHS-2), a large North American cohort characterized by low prevalence of smoking, alcohol, and meat consumption, and wide variation in nut intake. Because peanuts are botanically classified as legumes [36] with distinct micronutrient and phytochemical profiles, and peanut butter is often consumed with added sodium, sugars, and oils, we analyzed tree nuts separately to assess whether the associations with CVD and IHD mortality differ from total nuts, while also allowing for comparison with prior studies.

Methods

Study population

A detailed description and explanation of the AHS-2 cohort was provided by Butler et al. [37]. In brief, over 96,000 Seventh-Day Adventist males and females were recruited in churches throughout the United States and Canada between 2002 and 2007. Before the start of the study, all pertinent documents, procedures, and instruments were reviewed and approved by the Loma Linda University Institutional Review Board (# 5220197) and have been approved and renewed annually thereafter. Informed consent was obtained from all participants at enrollment.

Of the 96,354 participants in AHS-2, linkage with the National Death Index was possible for 92,361. Among these United States male and female participants with available mortality data, we applied the following exclusions: subjects with reported age <30 (n = 686); having cancer at baseline (except for nonmelanoma skin cancer; n = 7090); energy intake <500 kcal/d or >4500 kcal/d [38]; extreme BMI (in kg/m²) <16 or >50, were missing >69 items on the food frequency questionnaire (FFQ) (n = 2354), or were missing information on age, sex, race, or BMI (n = 1702). A sample of 80,529 remained for analysis after exclusions were applied.

Mortality data

Deaths were identified through linkage with the National Death Index through December 2015. The International Classification of Diseases, 10th Revision codes for the underlying cause of death were used. CVD deaths were identified as codes I00-I99 and IHD deaths as I20-I25.

Assessment of nut intake

Participants completed a self-administered semiquantitative FFQ (https://adventisthealthstudy.org/sites/adventisthealthstudy.org/files/2023-04/ahs-2.pdf) of >200 food items at baseline. Prior to completing the FFQ, participants were instructed to carefully review the guidelines for each section. In Part B, titled “Your Eating Habits,” participants were specifically advised to “Include your use of foods in mixed dishes…, when answering.” The FFQ included various food categories such as specific types of fruits (including seasonal varieties), vegetables, grains, eggs, dairy products, meats, and a range of beverages.

The FFQ contained 6 predefined questions that specifically assessed the frequency of nuts and seeds. These were not open-ended but structured with fixed response options. The reminder to “Include use in mixed dishes” was reiterated in the heading of this section. Each item had 8 frequency categories that ranged from “never or rarely” to “4 or more per day” and 3 additional answer choices for serving sizes (“standard,” “½ or less,” and “1½ or more”). A standard serving size for nuts and peanut butter was 28 g or 1 ounce, and 1 tablespoon or 15 g, respectively. Previous validation of the FFQ was performed using multiple 24-h dietary recalls among a sample of 1011 White and Black participants [39]. For White participants, the deattenuated validity correlations for total nuts and seeds, tree nuts (walnuts, cashews, and almonds), peanuts, and peanut butter were 0.58, 0.37, 0.40, and 0.59, respectively. For Black participants, the correlations were 0.47, 0.27, 0.27, and 0.41, respectively. For our analysis, the total nuts variable included almonds, cashews, walnuts, mixed nuts, peanuts, and peanut butter, whereas the tree nuts variable only included cashews, almonds, walnuts, and mixed nuts. We included mixed nuts because according to the Food and Drug Association Code of Federal Regulations Title 21 [40], this food consists predominantly of 4 or more of the shelled tree nuts (almonds, walnuts, Brazil nuts, cashews, filberts, pecans, etc.), and may contain peanuts. The nut and seed variables were represented as pseudo-continuous (g/d) resulting from the product of the $\text{frequency}\times \text{portion}\text{size}\times \text{food}\text{weights}$ [41], then log-transformed and energy adjusted using the residual method [42].

Assessment of covariates

All covariates were measured at baseline and selected a priori as likely confounders based on prior studies and suspected relationships. The following variables were chosen for the analytic models: age (continuous; attained age as the time variable in proportional hazards analyses), sex (dichotomous, male compared with female), education (3 levels; up to high school, some college, and bachelor’s degree or higher), smoking [6 levels: never smoked, stopped smoking over 30 y ago (quit years 30+), stopped smoking over 20 years ago (quit years 21–30), stopped smoking over 10 y ago (quit years 11–20), stopped smoking over 5 years ago (quit years 6–10), and stopped smoking >1 y ago (quit years ≤5) or current smoker combined as 1 level], alcohol intake was separated into wine (5 intake levels: never drank, past user that drank no >3 glasses/mo, past user that drank ≥1 glass/wk, current user that drinks no >3 glasses/mo, and current user that drinks ≥1 glass/wk), and beer or wine coolers and liquor (5 intake levels; never drank, past user that drank no >3 drinks/mo, past user that drank ≥1 drink/wk, current user that drinks no >3 drinks/mo, and current user that drinks ≥1 drink/wk), physical activity (continuous; min/wk of “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”), hours of sleep (3 levels: ≤5 h, 6–8 h, and ≥9 h), BMI (log-transformed and splined), prevalent diabetes (dichotomous; having been treated for Type I or Type II diabetes mellitus in the last 12 mo), having been treated for hypertension in the last 12 mo (dichotomous), having been treated for hypercholesterolemia in the last 12 mo (dichotomous), prevalent CVD (dichotomous; history of myocardial infarction and/or stroke, or having been treated for angina pectoris or congestive heart failure in the last 12 mo), ever having previous CVD surgery (dichotomous; includes coronary bypass, angioplasty/stent, and carotid artery surgery), menopausal status of females (dichotomous; premenopausal compared with postmenopausal), hormone replacement therapy in postmenopausal females (3 levels; never used, current user, and previous user), medications taken for cholesterol and blood pressure (dichotomous; defined as taken for ≥2 y during the last 5 y), and aspirin use (dichotomous; defined as currently taking at least once a day within the last 5 y). Subjects that self-identified at least in part as Black (i.e., African American, West Indian/Caribbean, African, or Other Black) were categorized as Black, and all others as non-Black.

Dietary covariates included the following: total energy intake (continuous), intake of whole grains, fruits, vegetables, cheese, dairy, eggs, unprocessed red meats, processed meats, fish, and poultry (all as continuous). We also included coffee and tea as continuous variables. All continuous dietary variables were log-transformed for nonnormal data, energy adjusted using the residual method [42], and splined using 4 knots at quantiles 5, 35, 65, and 95 [43]. We evaluated multicollinearity using correlation analyses and variance inflation factors (VIFs). Correlations with nut variables were weak (|r| < 0.25), and VIFs ranged from 1.0 to 1.7, indicating multicollinearity was unlikely to bias our estimates.

Dietary supplements (vitamin E, vitamin D, folate, magnesium, and fish oil) were explored and entered into the model as separate covariates. No appreciable change in the effect estimates was observed and the dietary supplements were left out of the final model for simplicity (Supplementary Table 1).

Where possible, guided multiple imputation was performed to fill in the missing data in key dietary variables [44]. Where this was not possible, multiple imputation using chained equations was performed for missing data (e.g., for nondietary variables) [45]. In both cases, 5 rounds of imputation were used.

Statistical analysis

Attained-age Cox proportional hazards regression with left censoring was performed to assess the relationship between total nut and tree nut consumption and CVD and IHD mortality. The proportional hazards assumption was evaluated using log(-log) plots and tested using Schoenfeld residuals and attained-age interaction terms. Violations with race, history of CVD, and history of diabetes were found and the attained-age interaction terms were included in the final models. To address the challenges of imputing interaction terms, age was utilized as the time variable to allow the baseline data from different ages to inform the imputation process, under the assumption that interactions remained stable over the follow-up period [46]. We assessed for outliers using deviance residuals and observed no significant change >10% with the effect estimates.

The linearity of the relationship between the exposure variables and mortality outcomes was assessed using 4-knot restricted cubic spline regressions with knots located at the 5th, 35th, 65th, and 95th percentiles [43]. The likelihood ratio test was used to assess the fit between the log-linear and splined models. Although no significant differences in model fit were found, we selected the spline models because of flexibility and robustness at the tails of the distributions. We also conducted analyses using quintiles of intake (g/d). The linear trends across the quintiles were tested by assigning the median intake to each quintile and analyzing them as continuous variables. Estimates were calculated using 5 imputed datasets; the average estimates were obtained, and the standard errors were calculated using Rubin’s rules [47]. For each analysis, 3 models were used to show the effect of including additional covariates. Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated and reported.

We performed a priori subgroup analyses by age (≤65 y compared with >65 y), sex (male compared with female), race (Black compared with non-Black), type of diet [vegetarian (includes semivegetarian, pescovegetarian, lactoovovegetarian, and vegan) compared with nonvegetarian] and BMI (3 levels; ≤24.9, 25–29.9, and ≥30). Interaction tests for each exposure variable by age, sex, race, diet type, and BMI were performed. We also conducted subgroup analyses by CVD status, comparing participants with prevalent CVD, diabetes, or history of CVD-related surgery (included coronary bypass, angioplasty/stent, and carotid artery surgery) to those without. To assess the robustness of our associations against potential unmeasured or uncontrolled confounding, we also performed sensitivity analysis and reported the E-values for the estimates and upper confidence limits for the main results [48].

We investigated how replacing total nuts and tree nuts with selected protein sources and energy-dense foods impact the risk of CVD and IHD mortality. Cox proportional hazards regression was performed using the specified substitution model via the partition method where all food groups were entered simultaneously [49,50]. Models were adjusted for total energy intake. For this analysis, the substitution unit of 1 serving/d was chosen. The standard serving size for unprocessed red meat was 85 g, 45 g for processed meat (includes ground beef, sausage, salami, hotdogs, and bologna), 85 g for poultry (includes chicken and turkey), 85 g for fish (includes white fish and salmon), 50 g for legumes (includes refried, navy, garbanzo, pinto, lima, lentils, and soybean), 28 g for snacks (chips and popcorn), 85 g for desserts (doughnuts/pastries, cookies, cake, ice cream, and frozen yogurt), 50 g for eggs, 28 g for cheese, and 113 g for fried potatoes (French fries, hash browns, and potato wedges) [[51], [52], [53], [54], [55]]. The results can be interpreted as the estimated effects on CVD mortality and IHD mortality when increasing a serving (28 g) of total nuts or tree nuts per day while simultaneously decreasing the serving size of the other foods [55]. This approach has been previously performed and reported [50,55].

Analyses were performed using Statistical Analysis Software package version 9.4 (SAS Institute Inc., Microsoft). All statistical analyses were 2-sided, and statistical significance was considered if P <0.05.

Results

Table 1 presents the demographic characteristics of this cohort. Participants with the highest intake of nuts, in comparison with those with the lowest intake, were older, more educated, more avoidant of alcohol and smoking, and had lower intakes of unprocessed red meats, processed meats, and eggs. Conversely, they had higher intakes of vegetables, fruits, and whole grains. They also exhibited lower BMI and engaged in more exercise.

TABLE 1.

Baseline characteristics across quintiles of nut intake among Adventists (n = 80,529).

Baseline characteristics	Quintiles of total nut1 intake2 (g/d)
	Overall		Q1		Q2		Q3		Q4		Q5
			(n = 16,104)		(n = 16,130)		(n = 16,090)		(n = 16,099)		(n = 16,106)
Age (y), mean (SD)	57.5	(14.0)	55.1	(14.2)	55.8	(13.8)	57.5	(13.9)	58.9	(13.9)	60.3	(13.7)
Female, n (%)	52,107	(64.7)	11,333	(70.4)	10,805	(67.0)	10,483	(65.2)	10,127	(62.9)	9359	(58.1)
Black, n (%)	20,913	(26.0)	6699	(41.6)	5101	(31.6)	3672	(22.8)	2775	(17.2)	2666	(16.6)
Education (graduate degree and above), n (%)	31,495	(39.1)	5041	(31.3)	5878	(36.4)	6477	(40.3)	6942	(43.1)	7157	(44.4)
Exercise3 (min/wk), mean (SD)	81.4	(96.2)	65.1	(90.3)	73.6	(91.8)	80.5	(93.2)	88.1	(96.4)	99.7	(104.8)
Sleep (h/d), n (%)
≤5	8249	(10.2)	2303	(14.3)	1904	(11.8)	1499	(9.3)	1225	(7.6)	1318	(8.2)
>8	4549	(5.6)	949	(5.9)	835	(5.2)	925	(5.7)	926	(5.8)	914	(5.7)
Smoking status, n (%)
Past user	13,432	(16.7)	3020	(18.8)	2863	(17.8)	2627	(16.3)	2415	(15.0)	2507	(15.6)
Current user	2380	(3.0)	820	(5.1)	556	(3.4)	394	(2.4)	310	(1.9)	300	(1.9)
Alcohol, n (%)
Wine (current user, at least ≥1 glass/wk)	1240	(1.5)	248	(1.5)	241	(1.5)	240	(1.5)	263	(1.6)	248	(1.5)
Beer/liquor (current user, ≥1 drink/wk)	1393	(1.7)	388	(2.4)	308	(1.9)	262	(1.6)	218	(1.4)	217	(1.3)
Menopausal status (yes)4, n (%)	39,917	(49.6)	8100	(50.3)	7885	(48.9)	8010	(49.8)	8141	(50.6)	7781	(48.3)
Current hormone replacement therapy users5, n (%)	15,176	(18.8)	3358	(20.9)	3023	(18.7)	2932	(18.2)	2968	(18.4)	2895	(18.0)
Cardiovascular disease history, n (%)	6062	(7.5)	1272	(7.9)	1198	(7.4)	1198	(7.4)	1186	(7.4)	1208	(7.5)
History of cardiovascular disease surgery, n (%)	2906	(3.6)	559	(3.5)	564	(3.5)	586	(3.6)	595	(3.7)	602	(3.7)
Diabetes, n (%)	6788	(8.4)	1482	(1.8)	1411	(1.8)	1311	(1.6)	1255	(1.6)	1329	(1.7)
Hypertension, n (%)	21,440	(26.6)	5004	(31.1)	4454	(27.6)	4241	(26.4)	3967	(24.6)	3774	(23.4)
High cholesterol, n (%)	20,084	(24.9)	4109	(25.5)	3983	(24.7)	4062	(25.2)	4019	(25.0)	3911	(24.3)
Medications (current users), n (%)
Aspirin	10,959	(13.6)	2087	(13.0)	2137	(13.2)	2191	(13.6)	2239	(13.9)	2305	(14.3)
Blood pressure	12,135	(15.1)	2852	(17.7)	2574	(16.0)	2389	(14.8)	2242	(13.9)	2078	(12.9)
Cholesterol	8053	(10.0)	1701	(10.6)	1727	(10.7)	1604	(10.0)	1549	(9.6)	1472	(9.1)
BMI (kg/m2), mean (SD)	27.0	(5.5)	28.2	(5.9)	27.7	(5.7)	27.0	(5.4)	26.4	(5.3)	25.9	(5.0)
Total energy (kcal), mean (SD)	1962.8	(778.2)	1977.4	(816.5)	1908.9	(774.8)	1971.9	(781.7)	1987.5	(736.3)	1954.4	(755.6)
Supplements (current users), n (%)
Vitamin D	9751	(12.1)	1596	(9.9)	1711	(10.6)	1916	(11.9)	2169	(13.5)	2359	(14.6)
Vitamin E	24,548	(30.5)	4073	(25.3)	4453	(27.6)	4972	(30.9)	5348	(33.2)	5702	(35.4)
Folate	8057	(10.0)	1496	(9.3)	1484	(9.2)	1604	(10.0)	1647	(10.2)	1826	(11.3)
Fish oil	46,026	(57.2)	9256	(57.5)	9209	(57.1)	9239	(57.4)	9198	(57.1)	9124	(56.6)
Magnesium	7876	(9.8)	1063	(6.6)	1297	(8.0)	1534	(9.5)	1812	(11.3)	2170	(13.5)
Dietary variables (g/d), mean (SD)
Fruits	328.8	(289.7)	235.8	(259.0)	284.5	(279.0)	319.6	(268.2)	365.8	(275.0)	438.5	(321.0)
Vegetables	284.7	(182.3)	215.0	(159.8)	252.7	(165.9)	282.1	(168.7)	311.6	(172.5)	362.3	(205.7)
Whole grain	174.5	(146.7)	125.2	(132.8)	145.3	(130.6)	172.9	(137.5)	198.0	(145.3)	231.1	(160.5)
Red meat	3.7	(10.5)	5.1	(12.5)	4.4	(11.0)	3.6	(9.8)	3.0	(9.7)	2.5	(9.0)
Processed meat	4.1	(9.6)	5.7	(11.6)	5.0	(10.5)	4.0	(8.8)	3.2	(8.3)	2.6	(7.8)
Fish	11.0	(19.7)	11.6	(20.7)	11.9	(19.1)	10.8	(18.7)	10.2	(19.2)	10.5	(20.6)
Poultry	9.4	(18.3)	12.6	(21.7)	10.6	(18.5)	8.9	(16.8)	7.6	(16.5)	7.0	(17.0)
Coffee	70.8	(179.0)	80.1	(192.2)	79.4	(188.0)	71.8	(178.3)	65.1	(168.6)	57.6	(165.4)
Tea	32.6	(104.6)	29.4	(101.6)	31.7	(102.9)	31.4	(97.1)	32.5	(101.3)	37.8	(118.4)
Dairy	149.2	(205.6)	141.4	(202.2)	155.2	(203.9)	158.1	(207.5)	150.2	(204.5)	141.3	(209.2)
Eggs	10.2	(16.8)	10.3	(17.8)	10.7	(16.6)	10.6	(16.8)	9.8	(15.1)	9.6	(17.7)
Tree nuts6	10.4	(15.3)	1.1	(1.1)	3.5	(2.1)	6.5	(4.3)	11.9	(7.2)	29.0	(24.5)
Additional dietary variables used for substitution models (g/d), mean (SD)
Snacks	3.2	(6.3)	2.6	(6.7)	3.2	(6.8)	3.2	(5.7)	3.3	(5.9)	3.6	(6.1)
Sweets	47.0	(53.6)	43.7	(57.2)	48.2	(55.4)	48.2	(50.8)	47.4	(50.2)	47.7	(53.7)
Cheese	17.1	(24.6)	12.9	(21.2)	16.0	(22.6)	18.4	(24.2)	18.8	(25.3)	19.5	(28.4)
Fried potatoes (hashbrowns, French fries, etc.)	11.6	(20.1)	11.7	(20.8)	12.6	(20.6)	11.9	(20.2)	10.9	(19.0)	10.6	(19.9)
Legumes	63.6	(66.7)	42.2	(55.8)	53.6	(57.1)	63.1	(61.3)	72.4	(67.1)	86.7	(80.0)
Vegetarian status, n (%)
Nonvegetarian	38,754	(48.1)	10,050	(62.6)	8923	(55.2)	7995	(49.7)	6227	(38.7)	5559	(34.5)
Vegetarian7	41,775	(51.9)	5995	(37.4)	7253	(44.8)	8098	(50.3)	9882	(61.3)	10,547	(65.5)

¹Total nuts included: almonds, cashews, walnuts, mixed nuts, peanuts, and peanut butter.

²Quintiles based on percentiles of the energy-adjusted g/d intake of total nuts (median quintiles are as follows, Q1 = 2.24, Q2 = 6.38, Q3 = 12.22, Q4 = 20.85, Q5 = 39.08).

³Exercise defined as “vigorous activities, such as brisk walking, jogging, bicycling, etc., long enough or with enough intensity to work up a sweat, and get your hear thumping or get out of breath.”

⁴Calculated among women only.

⁵Current hormone replacement therapy users among postmenopausal women only.

⁶Tree nuts included: almonds, cashews, walnuts, and mixed nuts.

⁷Vegetarian status included: vegan, lactoovovegetarian, pescovegetarian, and semivegetarian.

We documented 4258 deaths from CVD, and of those, there were 1529 deaths due to IHD during an average follow-up time of 11.1 y. Overall, the consumption of nuts was significantly inversely associated with the risk of CVD and IHD mortality when adjusted for age, sex, race, and total energy intake (model 1; Table 2), with associations attenuating but remaining significant in fully adjusted models that included medical, lifestyle, and dietary covariates (model 3; Table 2). Comparisons between the 90th percentile and 10th percentile of intake revealed significant inverse associations with CVD mortality for both total nuts (including peanuts and peanut butter) (HR: 0.86; 95% CI: 0.79, 0.94) and tree nuts (HR: 0.83; 95% CI: 0.74, 0.92). Stronger significant inverse associations were observed for IHD mortality for both total nuts (HR: 0.81; 95% CI: 0.70, 0.94) and tree nuts (HR: 0.73; 95% CI: 0.62, 0.87). Figure 1 presents the spline plots derived from the fully adjusted multivariable model of the intake of total nuts and tree nuts with risk of CVD and IHD mortality. No evidence of nonlinearity was observed in the spline models, as indicated by the P values of 0.73 and 0.33 for total nut intake and 0.21 and 0.81 for tree nut intake with CVD and IHD mortality, respectively.

TABLE 2.

The association between total nut¹ and tree nut² intake and cardiovascular disease mortality and ischemic heart disease mortality among Adventists (n = 80,529)³.

Cardiovascular disease mortality
Total nuts	Quintiles of intake					P-trend4	90th vs. 10th	P value
	Q1	Q2	Q3	Q4	Q5
Median intake (g/d)	2.24	6.38	12.22	20.85	39.08		39.08 vs. 2.24
No. of participants	16,104	16,130	16,090	16,099	16,106
No. of deaths (n = 4258)	820	730	830	894	984
Model 1	1.00 (Reference)	0.92 (0.82, 1.03)	0.89 (0.79, 1.01)	0.80 (0.72, 0.89)	0.78 (0.70, 0.87)	<0.0001	0.79 (0.73, 0.86)	<0.0001
Model 2	1.00 (Reference)	0.98 (0.87, 1.09)	0.98 (0.86, 1.11)	0.89 (0.81, 0.99)	0.90 (0.80, 1.01)	0.0115	0.88 (0.80, 0.96)	0.0051
Model 3	1.00 (Reference)	1.00 (0.90, 1.12)	1.01 (0.87, 1.16)	0.91 (0.82, 1.02)	0.87 (0.76, 1.04)	0.0128	0.86 (0.79, 0.94)	0.0007
Tree nuts
Median intake (g/d)	0.00	2.38	4.89	11.26	25.53		25.53 vs. 0.00
No. of participants	16,045	16,176	16,093	16,109	16,106
No. of deaths (n = 4258)	833	778	795	889	963
Model 1	1.00 (Reference)	0.86 (0.78, 0.96)	0.82 (0.73, 0.91)	0.73 (0.66, 0.80)	0.70 (0.63, 0.78)	<0.0001	0.73 (0.68, 0.79)	<0.0001
Model 2	1.00 (Reference)	0.92 (0.83, 1.02)	0.90 (0.81, 1.01)	0.84 (0.76, 0.92)	0.83 (0.74, 0.92)	<0.0001	0.83 (0.77, 0.91)	<0.0001
Model 3	1.00 (Reference)	0.95 (0.85, 1.05)	0.93 (0.84, 1.04)	0.86 (0.78, 0.95)	0.83 (0.75, 0.93)	<0.0001	0.83 (0.74, 0.92)	0.0005
Ischemic heart disease mortality
Total nuts
Median intake (g/d)	2.24	6.38	12.22	20.85	39.08		39.08 vs. 2.24
No. of participants	16,105	16,094	16,065	16,159	16,106
No. of deaths (n = 1529)	309	272	310	311	327
Model 1	1.00 (Reference)	0.86 (0.73, 1.01)	0.85 (0.71, 1.02)	0.67 (0.57, 0.79)	0.66 (0.56, 0.77)	<0.0001	0.65 (0.56, 0.74)	<0.0001
Model 2	1.00 (Reference)	0.92 (0.78, 1.09)	0.97 (0.81, 1.16)	0.78 (0.66, 0.91)	0.81 (0.69, 0.96)	0.0015	0.80 (0.69, 0.92)	0.0016
Model 3	1.00 (Reference)	0.94 (0.79, 1.11)	0.99 (0.81, 1.20)	0.79 (0.67, 0.93)	0.80 (0.67, 0.95)	0.0010	0.81 (0.70, 0.94)	0.0046
Tree nuts
Median intake (g/d)	0.00	2.38	4.89	11.26	25.53		25.53 vs. 0.00
No. of participants	16,045	16,176	16,093	16,109	16,106
No. of deaths (n = 1529)	341	284	256	336	312
Model 1	1.00 (Reference)	0.78 (0.67, 0.90)	0.75 (0.64, 0.89)	0.65 (0.56, 0.75)	0.59 (0.51, 0.70)	<0.0001	0.60 (0.51, 0.71)	<0.0001
Model 2	1.00 (Reference)	0.84 (0.72, 0.98)	0.85 (0.71, 1.02)	0.78 (0.67, 0.91)	0.74 (0.63, 0.87)	0.0010	0.72 (0.62, 0.82)	<0.0001
Model 3	1.00 (Reference)	0.86 (0.73, 1.00)	0.88 (0.74, 1.05)	0.80 (0.68, 0.94)	0.74 (0.63, 0.88)	0.0014	0.73 (0.62, 0.87)	0.0080

¹Total nut included: almonds, cashews, walnuts, mixed nuts, peanuts, and peanut butter.

²Tree nuts included: almonds, cashews, walnuts, and mixed nuts.

³Hazard ratios (HR) and 95% confidence intervals (CI). All exposure variables and continuous dietary covariates were energy-adjusted, log-transformed, and splined using 4 knots. Model 1 was adjusted for age, sex, race, and total energy intake. Model 2 was additionally adjusted by education, smoking status, hours of sleep, history of cardiovascular disease, history of cardiovascular disease surgery, having been treated for hypertension in the last 12 mo, having been treated for hypercholesterolemia in the last 12 mo, history of diabetes, physical activity, aspirin use, medications taken for cholesterol and blood pressure, and BMI; for women, the model was adjusted for menopausal status and hormone replacement therapy (HRT), and was entered as nested covariates (i.e., sex + sex∗menopausal status + sex∗menopausal status∗HRT). Model 3 was additionally adjusted by fruits, vegetables, whole grains, unprocessed red meats, processed meats, fish, poultry, dairy, eggs, coffee and tea as separate variables, beer or wine coolers and liquor, and wine.

⁴The linear trend test was conducted by entering the categorical variable as a continuous variable in the quintile model, followed by the likelihood ratio test to obtain the P values.

FIGURE 1.

Spline plots obtained from the fully adjusted multivariable analysis of the association between total nut and tree nut consumption and cardiovascular disease (CVD, left panels) mortality and ischemic heart disease (IHD, right panels) mortality. Spline estimates are shown with the reference set a 0.00 g/d intake. Models were adjusted for age, sex, race, total energy intake, education, smoking status, sleep hours, history of cardiovascular disease, history of cardiovascular disease surgery, high blood pressure and high cholesterol, history of diabetes, exercise min/wk, aspirin use, cholesterol and blood pressure medication, and BMI; for women, the model was adjusted for menopausal status and hormone replacement therapy (HRT) and was entered as nested covariates (i.e., sex + sex∗menopausal status + sex∗menopausal status∗HRT); fruits, vegetables, whole grains, red meats, processed meats, fish, poultry, dairy, eggs, coffee and tea as separate continuous variables, beer or wine coolers and liquor, and wine. The solid lines are the point estimate of the hazard ratio (HR), and the dashed lines are the 95% confidence intervals (CI). The P90 line indicates the 90th percentile of the median intake for total nuts (39.08 g/d) and tree nuts (25.53 g/d), respectively.

The subgroup analysis results comparing the 90th and 10th percentiles for intake of total nuts and tree nuts are presented in Supplementary Tables 2 and 3. No significant effect modification across levels of the subgroup variables (i.e., P for interaction) was found.

Supplementary Table 4 presents the E-values for the associations between total nut and tree nut intake with CVD and IHD mortality in the full cohort. The E-values for the estimates (eHR) and upper confidence limits (uCL) for the associations between total nut and tree nut consumption with CVD mortality were eHR of 1.4 (uCL: 2.0) and 1.4 (uCL: 2.1), respectively. For IHD mortality, the eHR (uCL) for total nuts and tree nut consumption were 1.2 (uCL: 1.7) and 1.2 (uCL: 2.4). These E-values suggest that the observed associations could potentially be explained away by an unmeasured confounder of modest strength.

Figure 2 presents the results for substituting a serving of 28 g/d of total nuts or tree nuts for selected food items. We observed a significantly lower risk of CVD and IHD mortality when total nuts and tree nuts were substituted for some of the selected foods. Specifically, when replacing total nuts with unprocessed red meats, we observed a 41% (95% CI: 19%, 57%) and 35% (95% CI: 14%, 63%) reduction in risk of CVD mortality and IHD mortality, respectively. The substitution of total nuts for other food items yielded a 15%–22% lower risk of CVD mortality, and a 17%–29% lower risk of IHD mortality. We observed similar reductions in risk when tree nuts were substituted for the selected foods. Specifically, we observed a 44% (95% CI: 23%, 59%) reduction in risk of CVD mortality, and we observed a 38% (95% CI: 17%, 64%) reduction in risk of IHD mortality when tree nuts were substituted for unprocessed red meat. Substituting tree nuts for processed meat, snacks, sweets, eggs, cheese, and fried potatoes was associated with a 19%–26% lower risk for CVD mortality, and a 20%–31% lower risk for IHD mortality. Substituting legumes was associated with a modest 6% lower risk of CVD mortality for both total nuts (95% CI: 1%, 11%) and tree nuts (95% CI: 0%, 11%), whereas no significant associations were observed for IHD mortality. No significant associations were observed when total nuts and tree nuts were substituted for poultry or fish.

FIGURE 2.

Forest plots presenting hazard ratio (HR) and 95% confidence intervals (CI) illustrating the substitution of 1 standard serving (28 g/d) of total nuts and tree nuts for 1 standard serving of selected foods in relation to cardiovascular disease (CVD, first 2 forest plots) mortality and ischemic heart disease (IHD, last 2 forest plots) mortality among Adventist Health Study 2 participants (n = 80,529). HR and 95% CI were obtained from the fully adjusted multivariable analysis and was adjusted for age, sex, race, total energy intake, education, smoking status, sleep hours, history of cardiovascular disease, history of cardiovascular disease surgery, high blood pressure and high cholesterol, history of diabetes, exercise min/wk, aspirin use, cholesterol and blood pressure medication, and BMI; for women, the model was adjusted for menopausal status and hormone replacement therapy (HRT) and was entered as nested covariates (i.e., sex + sex∗menopausal status + sex∗menopausal status∗HRT); fruits, vegetables, whole grains, red meats, processed meats, fish, poultry, dairy, eggs, coffee and tea as separate continuous variables, beer or wine coolers and liquor, and wine. ∗Total nuts included: almonds, cashews, walnuts, mixed nuts, peanuts and peanut butter. ∗∗Tree nuts included: almonds, cashews, walnuts, and mixed nuts. ∗∗∗Fried potatoes included: hashbrowns and French fries.

Discussion

In this large prospective cohort study of Adventists, a higher intake of total nuts and especially of tree nuts was inversely associated with risk of CVD mortality and IHD mortality in a log-linear fashion.

This study builds upon the seminal research on the inverse association of nut intake and coronary artery disease mortality [12] as observed in the Adventist Health Study 1 (AHS-1) [56]. Our current research in this population is strengthened by a larger sample size, a more detailed assessment of nuts, a more diverse participant pool by geography and race, a longer follow-up duration, a higher number of events, controlling for a broader array of confounders, and adding the more comprehensive outcome of CVD mortality. These differences highlight the evolving methodology and analytical refinements between the 2 studies. Although our overall findings corroborated the inverse association between nut intake and coronary artery disease mortality that was reported in the earlier research [12], we noted smaller effect estimates. Even within AHS-1, incident IHD results (myocardial infarctions) were stronger than mortality outcomes (although the 95% CIs for incident estimates were wider). We also observed consistent inverse associations across vegetarian and nonvegetarian subgroups, although these findings should be interpreted with caution due to potential limitations in power and residual confounding by unmeasured or poorly measured variables. Overall, the consistency of the inverse associations strengthens the evidence supporting the cardiovascular health benefits of nuts.

The broader epidemiological literature [18,20,21,23,24,29,30] often evaluates total nut consumption, which includes tree nuts, peanuts, and peanut butter. Our observations of an inverse association between total nut consumption and CVD and IHD mortality are consistent with several United States cohorts. In a combined analysis of the Nurses’ Health Study (NHS) and Health Professional Follow-up Study (HPFS) [29], consuming ≥5 servings/wk of total nuts (tree nuts and peanuts) was significantly inversely associated with CVD (HR: 0.75; 95% CI: 0.62, 0.84) and IHD mortality (HR: 0.71; 95% CI: 0.63, 0.81). A more recent analysis of the NHS and HPFS cohorts [55] reported a lower risk of CVD mortality (HR: 0.91; 95% CI: 0.88, 0.96) per 0.5-serving/d increase in total nut (tree nuts and peanuts) intake. Similar inverse associations were also observed in the Southern Community Cohort [21]. In contrast, several international cohorts, such as the Golestan Cohort [15], and the Netherlands Cohort [23] have reported nonsignificant inverse associations between total nut intake and CVD and IHD mortality. These differing findings may be due to smaller sample sizes, shorter follow-up periods resulting in fewer events, and limited variability in nut intake.

Fewer studies [20,23,29] have specifically evaluated the association between intake of tree nuts and risk of CVD and IHD mortality. Our findings align with results from the NHS and HPFS cohorts [29], where consuming tree nuts ≥5 times/wk significantly reduced CVD mortality risk (HR: 0.76; 95% CI: 0.67, 0.85). Extending these analyses to include the NHS II cohort, Liu et al. [20] reported that a 0.5-serving/d increase in tree nut (includes walnuts) intake (14 g/d) was associated with a 10% reduction in IHD mortality risk (HR: 0.90; 95% CI: 0.83, 0.98). Additionally, the Prevencion con Dieta Mediterranea trial, the largest clinical trial to date on this topic, reported that a Mediterranean diet enriched with nuts (almonds, peanuts, hazelnuts, pistachios, pine nuts, and walnuts) significantly reduced the risk of CVD mortality [HR: 0.42; 95% CI: 0.24, 0.74 for 1–3 servings/wk (28 g), and HR: 0.45; 95% CI: 0.25, 0.81 for >3 servings/wk compared with nonconsumers] [25]. The trial also reported a significant inverse association between consuming >3 servings/wk of walnuts (HR: 0.53; 95% CI: 0.29, 0.98) and CVD mortality.

Substitution analyses illustrate the potential practical impact of eating nuts in place of certain foods. The notable strong inverse association observed when substituting both total nuts and tree nuts for unprocessed red meat is consistent with findings from other cohort studies. A large prospective cohort study of 121,342 United States participants from the HPFS and NHS reported that replacing 1 serving of unprocessed red meat (85 g/d) and processed red meat (28 g/d) with a serving of nuts (28 g/d) was associated with an 18% and 22% lower risk of CVD mortality, respectively [33]. The substantial reduction in risk associated with replacing unprocessed red meat may reflect the strong adverse associations previously reported for red meat in the AHS-2 cohort [57]. These observations demonstrate the potential of dietary modifications, particularly replacing red and processed meats with nuts, to lower the risk of CVD and IHD mortality. Our results contribute to the existing body of evidence [22,33,34] supporting the favorable role of nuts in reducing the risk of CVD and IHD mortality when substituted for selected food items.

Several mechanisms have been proposed to explain the association between nut intake and CVD and IHD mortality. A meta-analysis of 61 clinical trials [58] reported that consuming a median dose of 56 g/d of tree nuts significantly reduced total cholesterol, LDL-C, apolipoprotein B, and triglycerides without affecting HDL-C. Meta-analyses of clinical trials [59] and cross-sectional studies [60,61] also suggest that nut consumption can reduce proinflammatory cytokines such as IL-6, ICAM-1, C-reactive protein, and TNF-α, which are known contributors to the development of CVD. Additional potential mechanisms include the role of nuts in managing adiposity through satiety effects [62], and the benefits of alpha-linolenic acid in nuts like walnuts for endothelial function by enhancing membrane fluidity and nitric oxide synthesis [63].

Compared with earlier analyses of this population [12], this study benefits from a longer follow-up time, a larger sample size, and a more diverse cohort, which improved the reliability and generalizability of our findings. Additionally, the AHS-2 FFQ’s detailed assessment of nut intake provided greater flexibility in modeling the nuts variable compared with Fraser et al. [12] and other well-regarded cohorts.

This analysis is limited to assessing diet and other covariates at a single time point, whereas intakes may have changed over time. However, most of the AHS-2 participants adhered to their diet type patterns (e.g., vegetarian) for ≥10 y [64], indicating some stability in dietary habits. The generalizability of our study findings may be limited by the unique characteristics of the AHS-2 cohort, which consists predominantly of health conscious and religious participants. However, there are many individuals outside of this cohort that adopt similar health-focused practices. Furthermore, the biological effects of nuts are unlikely to differ substantially between this group and other populations, and the consistency of our results with findings from other cohorts supports the generalizability of our findings. Additionally, the AHS-2 population is diverse in terms of sex, race, and geographical location, which enhances the relevance of our findings. Dietary assessment via FFQ can lead to measurement error, as reflected in the reported validity correlations for nut intake. Such measurement error in the exposure variable generally leads to attenuated results. Although the FFQ included instructions to report foods consumed in mixed dishes, nut intake from these sources may have been reported with limited validity. Although education served as a proxy for socioeconomic status in our models, unmeasured confounding by income may persist given the high cost of nuts. Residual confounding remains a concern despite adjusting for multiple potential confounders. The calculated E-values for the eHRs (range of 1.2–1.4) and uCLs (range of 1.7–2.4) suggest that a modest unmeasured confounder could potentially explain away the observed associations. Although these E-values warrant cautious interpretation, they do not constitute evidence against the presence of an association [48]. The consistency of our results with those reported from other cohorts is reassuring, as confounding is less likely to explain consistent findings in diverse populations. Finally, although we adjusted for total energy intake in our substitution models, substituting foods by weight may introduce residual differences in energy due to varying energy densities [50], potentially influencing the associations.

Although causality has not been established, the potential causal role of nuts in CVD and IHD mortality prevention deserves some consideration. According to Bradford Hill’s criteria for causation [65], several factors support a causal interpretation of the observed association between higher nut intake and reduced risk of CVD and IHD mortality. These factors include the consistency of findings across multiple studies involving various cohorts and geographic locations, the biological plausibility of nut components (such as unsaturated fats, fiber, and other bioactive compounds) in improving lipid profiles and vascular health, and the observed dose-response relationship, where higher nut consumption is associated with progressively lower CVD and IHD risks. Although further prospective studies and clinical trials are needed to strengthen these findings and support causality with greater certainty, the existing evidence supports a cautious interpretation that higher nut consumption likely reduces CVD and IHD mortality risks.

In this large Adventist cohort, we observed a significant inverse association with both total nuts and especially tree nuts and risk of CVD mortality and IHD mortality. The current findings, together with previously published evidence, suggest that higher intake of total nuts and tree nuts may lower the risk of CVD and IHD mortality.

Author contributions

The authors’ responsibilities were as follows – MSS, MJO, GEF, JS: designed research; GEF, MJO, FMB, JS, DJS: conducted research; MSS, DJS: analyzed data; and MSS: wrote the article. MJO, GEF: had primary responsibility for final content; and all authors: read and approved the final manuscript.

Data availability

Data described in the manuscript, code book, and data will be made available upon request pending application and approval by the Adventist Health Study-2 (AHS-2) group. For additional information please see our guidelines: AHS-2 Guidelines for external collaborators.pdf.

Funding

The original study was funded by the National Cancer Institute (NCI), grant 1U01CA152939 (PI: GEF). The funder had no role in the design of the study; in the collection, analyses, or interpretation of the data; in writing the manuscript, or in the decision to publish the results.

Conflict of interest

JS has received through his institution research grant funding, unrelated to this work, from the California Walnut Commission and the Almond Board of California. The other authors declare no conflicts of interest.

Appendix A. Supplementary data

The following is the Supplementary data to this article: