Planetary Health Diet Index and risk of total and cause-specific mortality in three prospective cohorts

Abstract

Background

In 2019, the EAT-Lancet Commission proposed a healthy dietary pattern that, along with reductions in food waste and improved agricultural practices, could feed the increasing global population sustainably. We developed a Planetary Health Diet Index (PHDI) to quantify adherence to the EAT-Lancet reference diet.

Objectives

We aimed to assess associations between PHDI and total and cause-specific mortality in 3 prospective cohorts of males and females in the United States.

Methods

We followed 66,692 females from the Nurses’ Health Study (1986–2019), 92,438 females from the Nurses’ Health Study II (1989–2019), and 47,274 males from the Health Professionals Follow-up Study (1986–2018) who were free of cancer, diabetes, and major cardiovascular diseases at baseline. The PHDI was calculated every 4 y using a semiquantitative food frequency questionnaire. Hazard ratios (HRs) were calculated using multivariable proportional-hazards models.

Results

During follow-up, we documented 31,330 deaths among females and 23,206 among males. When comparing the highest with the lowest quintile of PHDI, the pooled multivariable-adjusted HRs were 0.77 [95% confidence interval (CI): 0.75, 0.80] for all-cause mortality (P-trend < 0.0001). The PHDI was associated with lower risk of deaths from cardiovascular diseases (HR: 0.86; 95% CI: 0.81, 0.91), cancer (HR: 0.90; 95% CI: 0.85, 0.95), respiratory diseases (HR: 0.53; 95% CI: 0.48, 0.59), and neurodegenerative diseases (HR: 0.72; 95% CI: 0.67, 0.78). In females, but not males, the PHDI was also significantly associated with a lower risk of deaths from infectious diseases (HR: 0.62; 95% CI: 0.51, 0.76). PHDI scores were also associated inversely with greenhouse gas emissions and other environmental impacts.

Conclusions

In 3 large United States-based prospective cohorts of males and females with up to 34 y of follow-up, a higher PHDI was associated with lower risk of total and cause-specific mortality and environment impacts.

Introduction

As the world’s population is projected to grow to nearly 10 billion people by 2050, feeding all people a healthy diet sustainably is a global challenge [1,2]. Food systems play vital roles in both human health and the environment [2]. Agriculture and food production are responsible for ∼30% of greenhouse gas (GHG) emissions, 70% of water use, and ≥40% of land use [[3], [4], [5], [6], [7]]. In 2019, the EAT-Lancet Commission proposed a healthy dietary pattern based on a literature review of randomized dietary interventions with disease risk factors as outcomes, prospective cohort studies, and randomized trials with disease outcomes. Along with reductions in food waste and improved agricultural practices, based on global food system modeling, the adoption of this dietary pattern could feed the increasing global population sustainably and reduce GHG emissions by 50% in 2050 [2,8]. The food-based reference diet for generally healthy individuals aged ≥2 y emphasizes high consumption of high-quality plant-based foods (e.g., whole grains, vegetables, fruits, nuts and legumes, and unsaturated plant oils), low to moderate amounts of animal-sourced foods, and low intakes of saturated fats, refined grains, and sugar. Global shifts to this dietary pattern were found to be crucial to avoid severe environmental degradation and in modeling, could potentially prevent ∼11 million premature deaths per year [2,[9], [10], [11]].

In several studies, planetary health diet scores have been developed to quantify adherence to the EAT-Lancet reference diet [[12], [13], [14], [15], [16], [17], [18], [19], [20]] and to evaluate its effect on incidence of noncommunicable disease risk [12,[14], [15], [16], [17],[21], [22], [23], [24], [25], [26], [27]] and mortality [12,18,21]. However, most of these studies used binary or 3-level scores for food groups, which lead to a narrow range of possible scores, limited discrimination of adherence, and probably low power to detect associations [12,14,[16], [17], [18],[21], [22], [23], [24], [25], [26], [27]]. Some other studies proposed a continuous score with a wider range but categorized palm oil as unsaturated oils [13] or did not give higher scores for consumption of high-quality plant-based foods such as whole grains, legumes, and unsaturated oils [15,19,20], which are aligned with the EAT-Lancet reference diet.

To describe adherence to the EAT-Lancet reference diet more closely, we created a Planetary Health Diet Index (PHDI) that includes 15 food groups and ranges from 0 to 140. In this analysis, we aimed to assess the association between the PHDI and total and cause-specific mortality in 3 large prospective cohorts of adults in the United States. We hypothesized that a higher PHDI score would be associated with lower total and cause-specific mortality.

Methods

Study population

We conducted the analysis in 3 prospective cohort studies: Nurses’ Health Study (NHS1), Nurses’ Health Study II (NHS2), and Health Professionals Follow-up Study (HPFS). The NHS1 included 121,700 female registered nurses aged 30 to 55 y in 1976 [28,29]. The NHS2 recruited 116,429 female registered nurses aged 25 to 42 y in 1989 [28,29]. The HPFS started following 51,529 male health professionals (including veterinarians, dentists, pharmacists, optometrists, osteopathic physicians, and podiatrists) aged 40 to 75 y in 1986 [30]. Participants of the 3 cohorts provided information on their lifestyle factors, current health status, and medical history through mailed questionnaires every 2 to 4 y. All 3 cohorts had an ∼90% response rate in follow-up cycles [28,31].

We excluded participants who had any major chronic diseases (including cancer, diabetes, myocardial infarction, angina, stroke), or who underwent coronary artery bypass graft (CABG) at baseline (1986 for NHS1 and HPFS, and 1991 for NHS2). We also excluded people who had missing food frequency questionnaire (FFQ) data, reported implausible energy intake (<600 or >3500 kcal/d for females and <800 or >4200 kcal/d for males), or died during the first 2 y of follow-up to reduce reverse causation, leaving a total of 159,130 females and 44,275 males for analysis. The flowchart of participants is presented in Supplemental Figure 1. The study protocol was approved by the institutional review boards (IRBs) of the Brigham and Women’s Hospital and Harvard T.H. Chan School of Public Health and those of participating registries as required (further details in the Acknowledgments). The IRBs allowed participants’ completion of questionnaires to be considered as implied consent.

Dietary assessment

Dietary data were collected using self-administered, validated [32,33] semiquantitative FFQs every 4 y from 1986 to 2010 in NHS and HPFS and 1991 to 2015 in NHS2 [28,34]. The FFQs contained >130 food items with specified serving sizes that were commonly used for the United States population. Participants were asked how often they consumed 1 serving size of each food item on average over the past year. We converted the reported frequency of intake into servings/day using our conversion factors (Supplemental Table 1). From estimated consumption in servings/day, we calculated the corresponding consumption in grams/day for each food item using our serving weight database. Nutrient intakes were calculated by multiplying the frequency of each food intake by its nutrient content based on the USDA and Harvard University Food Composition Database [35].

Developing the PHDI

To quantify the adherence to the reference diet included in the EAT-Lancet report, we created scoring criteria for each of 15 food groups presented in Table 1. The reference diet was based on evidence for health outcomes and was subsequently determined to be consistent with staying within planetary boundaries for GHG emissions and other environmental parameters, assuming that the world’s population by 2050 consumed this diet [2]. The intent was to make the reference diet applicable to individuals in the global population aged >2 y, recognizing that the needs of some subgroups, such as pregnant and lactating females, may vary and that the overall energy and food intake of individuals will depend on body size and physical activity [2]. Also, the local availability of foods within broad categories and personal preferences varies within and between populations. For these reasons, and because the optimal intake of 1 food group often depends on intakes of other foods, we aimed to make the overall dietary pattern as flexible as possible. Therefore, in addition to a specific target number for each food group, a range of possible intakes was given that allowed exchanges within a given total energy intake.

TABLE 1.

The Planetary Health Diet Index (PHDI) components and criteria for scoring

No.	Component	EAT-Lancet reference diet (for 2500 kcal/d demand)		Criteria for scoring for PHDI (range 0–140)
		g/d	kcal/d	Min score (0) in g/d	Max score (10) in g/d	Weight in total score
1	Whole grain	232 (0%–60% of TEI)	811	0	≥75 g/d for females ≥90 g/d for males	1
2	Tubers (e.g., potatos, cassava)	50 (0–100)	39	≥200	≤50	1
3	Vegetable (not including potato or other starchy vegetables)	300 (200–600)	78	0	≥300	1
4	Whole fruit	200 (100–300)	126	0	≥200	1
5	Dairy foods (e.g., milk, cheese, yogurt)	250 (0–500)	153	≥1000	≤250	1
6	Red/processed meat (e.g., beef, lamb, pork)	14 (0–28)	30	≥100	≤14	1
7	Chicken and other poultry (e.g., duck, goose, ostrich)	29 (0–58)	62	≥100	≤29	1
8	Eggs	13 (0–25)	19	≥120	≤13	1
9	Fish and shellfish	28 (0–100)	40	0	≥28	1
10	Nuts (e.g., peanuts, tree nuts, such as walnuts, almonds, hazelnuts, pecan, cashews, pistachios)	50 (0–75)	291	0	≥50	1
11	Nonsoy legumes (e.g., dry beans, lentils, peas)	50 (0–100)	172	0	≥100	0.5
12	Soybean/soy foods	25 (0–50)	112	0	≥50	0.5
13	Added fat – unsaturated oils (not including transfat) (e.g., olive soybean, rapeseed, sunflower, peanut oil)	40 (20–80)	354 (14.16% of TEI)	≤3.5% of TEI	≥21% of TEI	1
14	Added fat – saturated oils and transfat (e.g., palm oil, coconut oil, dairy fat [butter], margarine, lard, tallow)	11.8 (0–11.8)	96 (3.8% of TEI)	≥10% of TEI	0% of TEI	1
15	Added sugar and sugar from fruit juice	31 (0–31)	120 (4.8% of TEI)	≥25% of TEI	≤5% of TEI	1

Abbreviations: PHDI, Planetary Health Diet Index; TEI, total energy intake.

The detailed rationale for our score and the list of food items used for each food group in our cohorts are provided in Supplemental Table 2. The scoring criteria were not statistically driven but derived from the reported biological dose–response relationships between each food group and risks of major chronic diseases in meta-analyses, pooled prospective cohort studies, large individual studies, and/or feeding trials. For the ranges of intakes, we also examined the distribution of consumption across countries in the Food and Agriculture Organization of the United Nations’s Food Balances website in 2018 [36]. Because the overall diet was for a specified energy intake, the intakes for food groups incorporate the fundamental concept of substitution, e.g., increasing the intake of 1 food group implies decreasing the intake of others. Where available, results from substitution analyses were considered, particularly in the case of comparing health risks of plant-based and animal-based food groups.

The minimum score for each food group (which was 0) was based on the consumption per day that reflects the least beneficial health effect of that food group (usually 0 g/d for healthy food groups). The maximum score for each food group [which was 10, except for nonsoy legumes (maximum 5) and soy foods (maximum 5)] was based on the consumption per day that reflected the greatest beneficial health effect of that food group (usually 0 g/d for unhealthy food groups). Depending on whether the relationship between each food group and disease risk was positive or inverse, the threshold for minimum and maximum score would change accordingly.

The interval of consumption (grams/day) between minimum and maximum score criteria should cover most of the distribution of consumption of that food group across countries. No deduction or additional score was given for consumption out of that minimum–maximum range. Individual scores for a food group were assigned proportionally for the range between minimum and maximum. The scores for each are then summed to calculate a total score; therefore, the total PHDI score ranged from 0 (nonadherence) to 140 (perfect adherence).

Ascertainment of death

Information on deaths was identified through next of kin, postal authorities, state vital statistics, death certificates, or through the National Death Index. The National Death Index has a sensitivity of 98% and a specificity of 100%, making it a highly sensitive method for identifying deaths in these cohorts [37,38]. Causes of death were further ascertained by reviewing medical records or autopsy reports by physicians with permission from study participants or family members [37,39]. Death causes were classified into nontrauma, cardiovascular diseases, cancer, neurodegenerative disease, respiratory disease, infectious disease, and other causes. The International Classification of Diseases (ICD) codes used to group these causes in each cohort (ICD-8 in the NHS and ICD-9 in the HPFS) are shown in Supplemental Table 3.

Assessment of environmental impact

For respondents to the 2011 NHS2 questionnaire (n = 62,620), we estimated GHG emissions, use of high-quality cropland, reactive nitrogen from fertilizer, and irrigation water from field to farm gate. We calculated these environmental impacts in the same manner as nutrient intakes using an environmental database created for our questionnaires described elsewhere [40].

Statistical analysis

Each participant provided follow-up time from the return date of the baseline questionnaire until the date of death or the end of follow-up (31 December, 2019, for females and 31 January, 2019, for males), whichever came first. The cumulative average of the PHDI scores up to the start of each follow-up interval was used to reduce measurement errors and thus obtain greater statistical power to detect associations compared to a 1-time dietary assessment [41]. We used Cox proportional-hazards models stratified for age months and follow-up cycles to calculate HRs for death by quintiles of the cumulative average PHDI with adjustment for potential confounders including time-fixed variables [i.e., White race, baseline BMI (in kg/m²), family history of diabetes, family history of myocardial infarction or cancer, baseline personal history of hypertension or hypercholesterolemia] and time-varying variables (i.e., marital status, living alone, neighborhood socioeconomic status z-score [42], premenopausal status and hormone use (for females only), smoking, alcohol consumption, physical activity, high blood pressure, high cholesterol, multivitamin use, aspirin use, and total energy intake). We tested the linear trend across quintiles by assigning a median value to each quintile and modeling these as a continuous variable. We also calculated HRs for death for every 20-point increase in the PHDI, which approximates to the difference between 10^th and 90^th percentile of PHDI in these cohorts. An example of this change would be replacing refined grain with whole grain (≥75 g/d for females and ≥90 g/d for males) and replacing added saturated fat (e.g., 1 cube of butter, ∼10 g/d) with added healthy plant oils (e.g., 4 teaspoons of olive oil, ∼20 g/d). In addition, we examined the associations between each food group and risk of total mortality to explore the extent that each food group contributed to the association with the total score.

We conducted subgroup analysis by potential effect modifiers of the association between the PHDI score and disease risk such as age (<65 or ≥65 y), race (White or Black or Asian), neighborhood socioeconomic status z-score (above or below median), smoking status (never, past, or current smoker), physical activity [≥7.5 metabolic equivalent of task (MET)-hours/week or not], alcohol (never, 0–9.9 g/d, or ≥10 g/d), baseline hypercholesterolemia, baseline hypertension, family history of diseases, BMI (obese or overweight or normal/underweight). Race and ethnicity were self-reported, with further details in Supplemental File 1. We evaluated potential interactions using Wald test between the cumulative continuous PHDI and subgroup variables. Lagged analysis for 4, 8, 12, 16, and 20 y were carried out to account for possible reverse causation due to dietary change prior to death [43].

We performed further sensitivity analyses to examine the robustness of our findings. First, to minimize reverse causation due to changes in diet when developing a chronic disease diagnosis, we stopped updating PHDI if participants had cancer, diabetes, myocardial infarction, stroke, or CABG during follow-up. Second, we estimated the association of only a 1-time measured PHDI at baseline with the outcome. Third, to estimate the confounding effect of socioeconomic status, we excluded neighborhood socioeconomic status z-score from the full model. Fourth, we further adjusted for time-varying continuous or quintiles of BMI to assess the potential mediating effect of weight change on the causal pathway. Fifth, we estimated the effect of PHDI independent of other diet patterns such as Alternative Healthy Eating Index (AHEI) [10], Dietary Approaches to Stop Hypertension (DASH) [44], Alternate Mediterranean Diet Score (AMED) [45], Healthy Eating Index 2015 (HEI-2015) [46], Plant-based Diet Index (PDI), Healthy Plant-based Diet Index (hPDI), and Unhealthy Plant-based Diet Index (uPDI) [47]. Sixth, to account for the competing events, we ran cause-specific hazard models [48], treating competing cause-specific deaths as a censoring mechanism, while adjusting for covariates mentioned above.

For each person, we standardized environmental impacts to energy intake of 2500 kcal/d using residuals from linear regression. To estimate associations between PHDI and standardized environmental impacts, we applied general linear models with further adjustment for total energy intake [47]; we also calculated mean values for quintiles of the PHDI.

We combined the results of female and male cohorts using inverse variance-weighted meta-analysis. All analyses were conducted in SAS version 9.4 (SAS Institute, Inc.), and a 2-sided P value of 0.05 was considered statistically significant.

Results

Age-standardized baseline characteristics of participants according to quintiles of the PHDI in each cohort are shown in Table 2. Compared with participants in the lower quintiles, those in the higher PHDI quintiles were older, had higher neighborhood socioeconomic z-score, used more multivitamin supplements, smoked less, consumed alcohol moderately, had a slightly lower BMI, were more physically active, consumed less energy per day.

TABLE 2.

Age-standardized baseline characteristics of participants by quintiles of Planetary Health Diet Index (PHDI) in 3 prospective cohorts

	Quintile 1	Quintile 2	Quintile 3	Quintile 4	Quintile 5
Nurses’ Health Study (NHS1)
Number of participants	13,326	13,338	13,342	13,338	13,348
PHDI in 1986, mean (SD)	60.8 (5.1)	70.1 (1.8)	75.7 (1.5)	81.3 (1.8)	90.7 (5.4)
PHDI in 1986, median (IQR)	62.1 (58.1–64.7)	70.1 (68.6–71.6)	75.7 (74.3–77)	81.2 (79.7–82.8)	89.4 (86.7–93.3)
Age in 1986,1 y	50.7 (7.2)	51.6 (7.2)	52.3 (7.1)	53.1 (7.1)	54.1 (6.8)
White, %	99	98	98	98	97
Married, %	68	70	72	71	72
Lived alone, %	10	10	10	10	10
Premenopausal, %	34	33	33	33	33
Neighborhood socioeconomic z-score, mean (SD)	−0.4 (3.6)	−0.2 (3.7)	0.0 (3.7)	0.2 (3.8)	0.6 (4.0)
Multivitamin use, %	39	41	42	44	48
Aspirin use, %	68	69	68	68	66
Current smoker, %	27	23	20	19	16
Alcohol drinking (g/d), median (IQR)	1.1 (0–6.4)	1.8 (0–7.3)	1.9 (0–7.9)	2 (0–8.6)	2 (0–8.8)
BMI (kg/m2), mean (SD)	25.4 (5.0)	25.4 (4.9)	25.3 (4.6)	25.1 (4.5)	24.5 (4.2)
Total activity MET-h/wk), median (IQR)	4.6 (1.7–13.6)	6.5 (2.4–16.5)	7.7 (2.9–18.3)	8.9 (3.2–20.9)	10.6 (4–24)
Total energy intake (kcal/d), mean (SD)	1937 (552)	1833 (545)	1771 (516)	1688 (483)	1609 (471)
History of high blood pressure, %	30	30	30	30	29
History of hypercholesterolemia, %	11	11	12	14	16
Family history of diabetes, %	28	29	29	29	28
Family history of myocardial infarction, %	25	25	25	25	26
Family history of cancer, %	14	14	15	15	15
Nurses’ Health Study 2 (NHS2)
Number of participants	18,485	18,494	18,488	18,484	18,487
PHDI in 1991, mean (SD)	55.4 (5.0)	64.6 (1.8)	70.1 (1.5)	75.9 (1.9)	86.2 (6.2)
PHDI in 1991, median (IQR)	56.6 (52.7–59.3)	64.6 (63.1–66.1)	70.1 (68.8–71.5)	75.8 (74.3–77.5)	84.5 (81.6–89)
Age in 1991,1 y	35.1 (4.8)	35.8 (4.7)	36.1 (4.6)	36.5 (4.6)	37 (4.5)
White, %	97	97	97	96	95
Married, %	88	89	87	87	84
Lived alone, %	7	7	8	8	10
Premenopausal, %	97	97	97	97	97
Neighborhood socioeconomic z-score, mean (SD)	–0.6 (3.3)	–0.3 (3.5)	–0.1 (3.6)	0.3 (3.8)	0.8 (4.0)
Multivitamin use, %	42	42	43	44	47
Aspirin use, %	12	11	11	11	11
Current smoker, %	15	13	12	11	10
Alcohol drinking (g/d), median (IQR)	0 (0–2.4)	0.9 (0–2.9)	0.9 (0–3.4)	1.1 (0–4)	1.5 (0–4.7)
BMI (kg/m2), mean (SD)	25.2 (5.9)	24.9 (5.5)	24.7 (5.3)	24.4 (5.0)	23.8 (4.6)
Total activity MET-h/wk, median (IQR)	9.1 (3.5–20.9)	10.7 (4.2–22.9)	12.1 (4.9–25.8)	14.5 (6–29)	18.3 (7.9–35.9)
Total energy intake (kcal/d), mean (SD)	1984 (553)	1836 (554)	1774 (550)	1711 (525)	1638 (492)
History of high blood pressure, %	7	6	6	6	6
History of hypercholesterolemia, %	15	15	14	14	14
Family history of diabetes, %	30	29	30	30	29
Family history of myocardial infarction, %	20	20	20	20	20
Family history of cancer, %	12	12	12	12	13
Health Professionals Follow-up Study (HPFS)
Number of participants	9438	9455	9459	9465	9457
PHDI in 1986, mean (SD)	59.9 (5.9)	70.6 (2.1)	77.1 (1.8)	83.7 (2.1)	94.5 (6.0)
PHDI in 1986, median (IQR)	61.3 (56.8–64.4)	70.7 (68.9–72.4)	77.1 (75.6–78.6)	83.6 (81.9–85.5)	93 (89.9–97.5)
Age in 1986,1 y	51.9 (9.6)	53.2 (9.8)	54.0 (9.8)	54.9 (9.7)	55.8 (9.5)
White, %	93	92	90	90	89
Married, %	89	90	91	91	90
Lived alone, %	7	6	6	6	6
Neighborhood socioeconomic z-score, mean (SD)	−0.7 (3.5)	−0.3 (3.6)	0.1 (3.6)	0.3 (3.8)	0.5 (3.8)
Multivitamin use, %	36	40	41	43	50
Aspirin use, %	28	29	29	29	30
Current smoker, %	15	11	9	7	5
Alcohol drinking (g/d), median (IQR)	4.2 (0–13.9)	5.8 (0.9–15.5)	6 (1–15.3)	6.4 (1–15.8)	6 (0.9–14.8)
BMI (kg/m2), mean (SD)	25.7 (3.4)	25.7 (3.3)	25.6 (3.5)	25.4 (3.1)	25 (3.2)
Total activity MET-h/wk, median (IQR)	7.8 (2.4–22)	10.7 (3.5–24.9)	11.9 (4.2–27.4)	14.4 (5.1–30.2)	17.7 (6.8–35.7)
Total energy intake (kcal/d), mean (SD)	2237 (660)	2056 (633)	1941 (591)	1873 (567)	1829 (558)
History of high blood pressure, %	24	26	27	28	29
History of hypercholesterolemia, %	8	11	12	14	18
Family history of diabetes, %	20	20	21	21	22
Family history of myocardial infarction, %	31	32	33	34	36
Family history of cancer, %	34	34	34	35	34

Values are means (SD) or median (IQR) for continuous variables, percentages for categorical variables, and are standardized to the age distribution of the study population.

Abbreviations: MET, metabolic equivalent of task; SD, standard deviation.

¹Nonstandardized variables.

Intakes of each food group over follow-up cycles are presented in Supplemental Table 4. In all 3 cohorts, consumption of high-quality plant-based foods including whole grains, nuts, soy foods, and added unsaturated fats increased over time, whereas consumptions of tubers and animal-based foods including red/processed meats, poultry, and fish/shellfish decreased. Participants in NHS1 and NHS2 gradually reduced their consumption of dairy foods. Also, NHS2 participants increased consumption of vegetables and whole fruits over time. The trend of PHDI scores over survey years in 3 cohorts is shown in Supplemental Figure 2. While the total score increased by 12% in NHS1 (from 75.7 ± 10.7 in 1986 to 84.5 ± 13.3 in 2010) and 22% in NHS2 (from 70.4 ± 11.1 in 1990 to 85.9 ± 14.7 in 2015), it remained stable at ∼78 in males during the follow-up (Supplemental Table 5). The increase of the total PHDI in NHS1 and NHS2 was attributable to increases in the scores for whole grains, unsaturated fats, saturated fats, nuts, poultry, red/processed meat, and tubers (Supplemental Figure 3). PHDI had a moderate to strong positive correlation with AHEI, AMED, HEI-2015, and hPDI in all 3 cohorts (Supplemental Table 6).

In the age-adjusted model, we found a strong inverse association between the PHDI and total mortality [HR_{Q5 vs. Q1}: 0.59; 95% confidence interval (CI): 0.58, 0.61]. In multivariable-adjusted models, whether defined as quintiles (Table 3), deciles (Figure 1), or continuously (Figure 2), the PHDI was also inversely associated with total mortality. When comparing the highest with the lowest quintile of PHDI, the pooled multivariable-adjusted HRs were 0.77 (95% CI: 0.75, 0.80) for all-cause mortality, 0.86 (95% CI: 0.81, 0.91) for cardiovascular mortality, 0.90 (95% CI: 0.85, 0.95) for cancer mortality, 0.53 (95% CI: 0.48, 0.59) for respiratory mortality, 0.72 (95% CI: 0.67, 0.78) for neurodegenerative mortality, and 0.78 (95% CI: 0.63, 0.98) for infectious mortality (Supplemental Table 7). The inverse associations between PHDI and cause-specific mortality were consistent in both the male and female cohorts except that no significant association with infectious disease mortality was seen in males (HR_{Q5 vs. Q1}: 1.16; 95% CI: 0.89, 1.50) (Supplemental Tables 7 and 8). When lagging cumulative average PHDI to reduce possible reverse causation, long-lagged PHDI was more strongly inversely associated with total and cause-specific mortality than short-lagged PHDI (Supplemental Table 9). In further sensitivity analyses, when we stopped updating the PHDI if participants had been diagnosed with common chronic diseases, the inverse associations were slightly less strong for most mortality outcomes (for total mortality, HR_{Q5 vs. Q1}: 0.79; 95% CI: 0.77, 0.81) but a bit stronger for cancer mortality and infectious mortality (Supplemental Table 10). One-time measured PHDI at baseline was more weakly associated with risk of death than cumulative average PHDI scores (the HR_{Q5 vs. Q1} with baseline only was 0.93 (95% CI: 0.90, 0.95)] compared with 0.77 (95% CI: 0.75, 0.80) with cumulative update). Removing neighborhood socioeconomic z-score or adding time-varying BMI to the main model did not materially change the results of total mortality (HR_{Q5 vs. Q1}: 0.76; 95% CI: 0.74, 0.79) as well as other cause-specific mortality. When we further adjusted for other diet scores such as the AHEI, AMED, DASH, HEI, or hPDI, the pooled estimates for total and cause-specific mortality were still consistent with the primary analysis and statistically significant except for cancer mortality and infectious disease mortality (Supplemental Table 10). Results for cause-specific mortality were also consistent when we further accounted for competing risks (Supplemental Table 11).

TABLE 3.

Hazard ratios (95% CIs) for total mortality according to quintiles of PHDI

Cause of death	Planetary Health Dietary Index by quintiles
	Q1	Q2	Q3	Q4	Q5	P-linear trend
Pooled
Case	11,439	11,451	11,115	10,685	9929
Person-years	1,095,803	1,113,863	1,124,168	1,134,181	1,149,126
Age-adjusted HR	1	0.86 (0.84, 0.88)	0.77 (0.75, 0.79)	0.69 (0.67, 0.71)	0.59 (0.58, 0.61)	<0.0001
Multivariable-adjusted HR	1	0.96 (0.94, 0.99)	0.91 (0.88, 0.93)	0.86 (0.84, 0.88)	0.77 (0.75, 0.80)	<0.0001
NHS1 & NHS2
Median score (IQR)	62 (58–65)	70 (67–72)	75 (73–77)	81 (78–83)	88 (86–92)
Case	6775	6686	6339	6041	5572
Person-years	858,070	874,727	882,430	888,515	897,933
Age-adjusted HR	1	0.86 (0.83, 0.89)	0.75 (0.72, 0.78)	0.68 (0.66, 0.70)	0.58 (0.56, 0.60)	<0.0001
Multivariable-adjusted HR	1	0.97 (0.94, 1.00)	0.90 (0.87, 0.93)	0.85 (0.82, 0.88)	0.77 (0.74, 0.80)	<0.0001
HPFS
Median score (IQR)	63 (59–66)	72 (70–74)	78 (77–80)	84 (83–86)	93 (90–97)
Case	4664	4765	4776	4644	4357
Person-years	237,733	239,136	241,738	245,666	251,193
Age-adjusted HR	1	0.86 (0.82, 0.89)	0.79 (0.76, 0.82)	0.71 (0.68, 0.74)	0.61 (0.59, 0.62)	<0.0001
Multivariable-adjusted HR	1	0.96 (0.92, 1.00)	0.93 (0.89, 0.97)	0.88 (0.84, 0.92)	0.78 (0.74, 0.81)	<0.0001

Multivariable-adjusted HRs of cumulative average quintiles of PHDI were stratified jointly by age in months and follow-up cycles and adjusted for race (White or not), marriage status (married or not), living status (alone or not), neighborhood socioeconomic status z-score, menopausal status (pre or postmenopausal [never, past, or current menopausal hormone use], for females only), multivitamin use (yes/no), aspirin use (yes/no), total energy intake (kcal/d), baseline BMI (kg/m²: <23, 23–24.9, 25–29.9, 30–34.9, ≥35), smoking status (never smoker, former smoker, current smoker: 1–14, 15–24, or ≥25 cigarettes/d), alcohol drinking (g/d: 0, 0.1–4.9, 5.0–9.9, 10.0–14.9, 15.0–29.9, or ≥30), physical activity (MET-h/wk: <3, 3-8.9, 9–17.9, 18–26.9, or ≥27), history of hypertension (yes/no), history of hypercholesterolemia (yes/no), family history of myocardial infarction (yes/no), family history of diabetes (yes/no), family history of cancer (yes/no). Fixed effects meta-analysis of the 3 cohorts (NHS1, NHS2, and HPFS) based on Dersimonian–Laird approach with inverse-variance weight was used to pool hazard ratios. Bold texts showed statistically significant HRs (P ≤ 0.05).

Abbreviations: CI, confidence interval; HPFS, Health Professionals Follow-up Study; HR, hazard ratio; NHS1, Nurses’ Health Study; NHS2, Nurses’ Health Study II; PHDI, Planetary Health Diet Index.

FIGURE 1.

Pooled HRs (95% CIs) for deciles of the PHDI in relation to total mortality among 206,404 males and females (54,536 deaths). HRs of cumulative average deciles were stratified jointly by age in months and follow-up cycles and adjusted for race (White or not), marriage status (married or not), living status (alone or not), neighborhood socioeconomic status z-score, menopausal status (pre or postmenopausal [never, past, or current menopausal hormone use], for females only), multivitamin use (yes/no), aspirin use (yes/no), total energy intake (kcal/d), baseline BMI (kg/m²: <23, 23–24.9, 25–29.9, 30–34.9, or ≥35), smoking status (never smoker, former smoker, current smoker: 1–14, 15–24, or ≥25 cigarettes/d), alcohol drinking (g/d: 0, 0.1–4.9, 5.0–9.9, 10.0–14.9, 15.0–29.9, or ≥30), physical activity (MET-h/wk: <3, 3–8.9, 9–17.9, 18–26.9, or ≥27), history of hypertension (yes/no), history of hypercholesterolemia (yes/no), family history of myocardial infarction (yes/no), family history of diabetes (yes/no), family history of cancer (yes/no). Fixed effects meta-analysis of the 3 cohorts (NHS1, NHS2, and HPFS) based on Dersimonian–Laird approach with inverse-variance weight was used to pool HRs. Mean (SD) of each decile of the PHDI for all 3 cohorts are presented. CI, confidence interval; HPFS, Health Professionals Follow-up Study; HR, hazard ratio; MET, metabolic equivalent of task; NHS1, Nurses’ Health Study; NHS2, Nurses’ Health Study II; PHDI, Planetary Health Diet Index; SD, standard deviation.

FIGURE 2.

Pooled HRs (95% CIs) of the PHDI for 20-point increase in total and cause-specific mortality in the 3 prospective cohorts. HRs of the continuous cumulative average PHDI scaled by 20 points (i.e., approximate to the difference between 10th and 90th percentiles of the PHDI) were stratified jointly by age in months and follow-up cycles and adjusted for race (White or not), marriage status (married or not), living status (alone or not), neighborhood socioeconomic status z-score, menopausal status (pre or postmenopausal [never, past, or current menopausal hormone use], for females only), multivitamin use (yes/no), aspirin use (yes/no), total energy intake (kcal/d), baseline BMI (kg/m²: <23, 23–24.9, 25–29.9, 30–34.9, or ≥35), smoking status (never smoker, former smoker, current smoker: 1–14, 15–24, or ≥25 cigarettes/d), alcohol drinking (g/d: 0, 0.1–4.9, 5.0–9.9, 10.0–14.9, 15.0–29.9, or ≥30), physical activity (MET-h/wk: <3, 3–8.9, 9–17.9, 18–26.9, or ≥27), history of hypertension (yes/no), history of hypercholesterolemia (yes/no), family history of myocardial infarction (yes/no), family history of diabetes (yes/no), family history of cancer (yes/no). Fixed effects meta-analysis of the 3 cohorts (NHS1, NHS2, and HPFS) based on Dersimonian–Laird approach with inverse-variance weight was used to pool HRs. P value was estimated from Wald test of the PHDI variable. ∗NHS2 data were not included due to low number of cases. CI, confidence interval; HPFS, Health Professionals Follow-up Study; HR, hazard ratio; MET, metabolic equivalent of task; NHS1, Nurses’ Health Study; NHS2, Nurses’ Health Study II; PHDI, Planetary Health Diet Index.

The PHDI was inversely associated with total mortality in all subgroups of the combined cohorts (Figure 3). A higher PHDI was most strongly associated with mortality in those who were current smokers (HR_{Q5 vs. Q1}: 0.72, 95% CI: 0.64, 0.82), drank >10 g/d of alcohol (HR_{Q5 vs. Q1}: 0.68, 95% CI: 0.64, 0.72), and more weakly associated in those with obesity (BMI ≥30) (HR_{Q5 vs. Q1}: 0.89, 95% CI: 0.83, 0.97) (Supplemental Table 12). The inverse association with total mortality was significant in the Asian American group (HR: 0.70, 95% CI: 0.52, 0.94) (Supplemental Table 12); the CIs for other non-White groups were wide due to the low number of cases. The subgroup analysis for total mortality in male and female cohorts is shown in Supplemental Tables 13 and 14. In addition, the inverse associations between the PHDI and respiratory mortality were consistent in all smoking subgroups including never-smokers (Supplemental Table 15).

FIGURE 3.

Pooled HRs (95% CIs) of PHDI for 20-point increase in total mortality across subgroups in the 3 prospective cohorts. HRs of continuous cumulative average PHDI scaled to 20 points were stratified jointly by age in months and follow-up cycles and adjusted for race (White or not), marriage status (married or not), living status (alone or not), neighborhood socioeconomic status z-score, menopausal status (pre or postmenopausal [never, past, or current menopausal hormone use], for females only), multivitamin use (yes/no), aspirin use (yes/no), total energy intake (kcal/d), baseline BMI (kg/m²: <23, 23–24.9, 25–29.9, 30–34.9, ≥35), smoking status (never smoker, former smoker, current smoker: 1–14, 15–24 or ≥25 cigarettes/d), alcohol drinking (g/d: 0, 0.1–4.9, 5.0–9.9, 10.0–14.9, 15.0–29.9, or ≥30), physical activity (MET-h/wk: <3, 3–8.9, 9–17.9, 18–26.9, or ≥27), history of hypertension (yes/no), history of hypercholesterolemia (yes/no), family history of myocardial infarction (yes/no), family history of diabetes (yes/no), family history of cancer (yes/no), except the corresponding subgroup variables. Time-fixed effect modifiers included race, baseline hypercholesterolemia, baseline hypertension, family history of diseases (diabetes, myocardial infarction, or cancer), baseline BMI; other effect modifiers were time-varying. Fixed effects meta-analysis of the 3 cohorts (NHS1, NHS2, and HPFS) based on Dersimonian–Laird approach with inverse-variance weight was used to pool HRs. P value for interaction was estimated from Wald test of the interaction terms between continuous PHDI variables and the corresponding subgroup variables in the pooled data set. CI, confidence interval; HPFS, Health Professionals Follow-up Study; HR, hazard ratio; MET, metabolic equivalent of task; NHS1, Nurses’ Health Study; NHS2, Nurses’ Health Study II; PHDI, Planetary Health Diet Index.

The food group scores contributing most strongly to the inverse association of the PHDI with mortality included unsaturated fats, whole grains, nuts, and red/processed meats (Figure 4 with further details in Supplemental Table 16). With intakes as gram weights or percent of total energy intake, in all 3 cohorts, higher intakes of whole grains, whole fruits, poultry, nuts, soy foods, and added unsaturated fats were associated with lower risks of total mortality, whereas consuming more tubers, red/processed meats, eggs, added saturated fats, added sugars, and sugar from fruit juices were associated with increased risks of total mortality.

FIGURE 4.

Pooled HRs (95% CIs) of Q5 vs. Q1 of each food group index and intake for total mortality in the 3 prospective cohorts. Hazard ratios (HRs) of each food group intake and index were stratified jointly by age in months and follow-up cycles and adjusted for race (White or not), marriage status (married or not), living status (alone or not), neighborhood socioeconomic status z-score, menopausal status (pre or postmenopausal [never, past, or current menopausal hormone use], for females only), multivitamin use (yes/no), aspirin use (yes/no), total energy intake (kcal/d), baseline BMI (kg/m²: <23, 23–24.9, 25–29.9, 30–34.9, ≥35), smoking status (never smoker, former smoker, current smoker: 1–14, 15–24, or ≥25 cigarettes/d), alcohol drinking (g/d: 0, 0.1–4.9, 5.0–9.9, 10.0–14.9, 15.0–29.9, or ≥30), physical activity (MET-h/wk: <3, 3–8.9, 9–17.9, 18–26.9, or ≥27), history of hypertension (yes/no), history of hypercholesterolemia (yes/no), family history of myocardial infarction (yes/no), family history of diabetes (yes/no), family history of cancer (yes/no). Fixed effects meta-analysis of the 3 cohorts (NHS1, NHS2, and HPFS) based on Dersimonian–Laird approach with inverse-variance weight. ∗The scores for these unhealthy food groups were reverse-coded to show that higher scores correspond to lower consumption. ∗∗Due to low variation of this food consumption, only 2 quantiles were made; HRs of the highest quantile compared with lowest quantile were reported. CI, confidence interval; HPFS, Health Professionals Follow-up Study; HR, hazard ratio; MET, metabolic equivalent of task; NHS1, Nurses’ Health Study; NHS2, Nurses’ Health Study II; Q, quantile.

In our environmental impact analysis, comparing the top with the bottom quintile of PHDI, GHG emissions were 29% lower, fertilizer needs were 21% lower, cropland use was 51% lower, and irrigation water needs were 13% lower (Supplemental Table 17). The environmental impacts of PHDI were similar to those of AHEI, AMED, and DASH. However, the inverse associations of the PHDI with environmental impacts were stronger than those of the HEI, PDI, and hPDI. For example, comparing the top and bottom quintiles of the HEI, GHG emissions were 16% lower than 29% lower for the PHDI. The uPDI scores were not associated with environmental impacts except that they were positively associated with fertilizer and cropland needs.

Discussion

In 3 large United States-based prospective cohorts of males and females with up to 34 y of follow-up, we found strong inverse associations between PHDI and total and cause-specific mortality. Compared with those in the lowest quintile of PHDI, those in the highest quintile had 23% lower risk of total mortality, 14% lower risk of cardiovascular mortality, 10% lower risk of cancer mortality, 47% lower risk of respiratory mortality, and 28% lower risk of neurodegenerative mortality. These associations remained robust after adjusting for lifestyle factors, high blood pressure, and elevated cholesterol levels, which may act as mediators of the association of interest, and other potential dietary confounders and within subgroups. The scores of unsaturated fats, whole grains, nuts, and red/processed meats were the most important components contributing to the inverse association of PHDI with mortality. Overall, our results support reductions in death from a variety of diseases with increasing adherence to a healthy and environmentally sustainable dietary pattern described by the EAT-Lancet Commission.

Compared to a similar analysis of other health indices in these 3 prospective cohorts [49], the PHDI had a modestly stronger inverse association with mortality, which indicates that adherence to the Planetary Health Diet (PHD) does not compromise health compared to other healthy dietary patterns.

Inverse associations between planetary health diets and risk of death [12,18], diabetes [12,22,23,25], cardiovascular diseases [12,24,27] have been reported in large prospective studies on Swedish, Danish, and United Kingdom populations. In the European Prospective Investigation into Cancer and Nutrition (EPIC)-Oxford study, which followed 46,069 participants throughout the United Kingdom ≤23.6 y, the highest adherence to EAT-Lancet diet (12–14 points) was associated with 9% lower risk of all-cause mortality, 28% lower risk of ischemic heart disease, and 59% risk of diabetes than the lowest adherence (4–9 points) [12]. In the Malmö Diet and Cancer cohort with 22,421 participants and a mean of 20 y of follow-up, the highest adherence to the EAT-Lancet diet was associated with 25% lower risk of all-cause mortality, 24% lower risk of cancer mortality, and 32% lower risk of cardiovascular mortality than the lowest adherence (≤13 points) [18].

In contrast, some prospective and cross-sectional studies did not find a significant association between a planetary health diet and total mortality [21], cardiovascular diseases [15,26], cancer [15], type 2 diabetes [17], or cardiometabolic risk markers [16]. The major possible explanation for these nonsignificant findings is the scoring used to measure adherence to the EAT-Lancet diet. Most of these studies only assigned a binary score for each food group, which limits the range for the total score and the power to detect associations [16,17,21,26,50,51]. In an analysis comparing alternative methods for scoring adherence to the EAT-Lancet reference diet, Stubbendorff et al. [50] showed that, compared to the use of a wide range of possible scores for each food group, the use of only binary scores generally showed weaker associations with mortality and GHG emissions [51]. Our scoring approach considered the distance from the reference intake as has been used for the HEI, AHEI, and most other dietary quality scores [52]. Also, some studies combined refined grain with whole grain [12,16,21,26], did not credit whole grain intake [15], or even negatively scored intake of unsaturated fat [17], which does not correctly represent the EAT-Lancet reference diet [2]. In our study, whole grains and unsaturated fats were among the most important components accounting for the associations of the PHDI score with mortality in both the male and female cohorts. Also, other limitations likely reduced the potential to detect associations, including having only a 1-time dietary assessment [16,17,21,26], a low number of cases [15], small sample size [16], or short duration of follow-up [15,17]. The far larger number of deaths than in previous studies and the use of many repeated assessments of diet to reduce measurement error [41] also added to the statistical power of our analysis and precision of our findings. We documented these possible issues in our sensitivity analyses, including the value of multiple dietary assessment and the benefit of long-term intake in the lagged analysis. Regarding the lagged analysis, we found a stronger association with longer-lagged PHDI; this may be due to a long biological latency between dietary factors and death, a reduction in reverse causation, or differences in the population that survived at least a certain number of years in each lagged subpopulation. However, this does suggest that our estimate of reduction mortality is an underestimate.

The EAT-Lancet diet may reduce the risk of chronic diseases and mortality in part by lowering BMI [12,16], waist circumference [53], and cardiometabolic risk factors [12,54]. Consistent with our findings, in the EPIC-Oxford study, adherence to the EAT-Lancet diet was associated with 1.4 lower BMI, 0.5 mmol/L lower plasma non-HDL cholesterol, and 3.5 mm Hg lower systolic blood pressure compared with low adherence in the United Kingdom population [12]. These results were supported by the findings of the Brazilian Longitudinal Study of Adult Health, where inverse associations were found between adherence to the EAT-Lancet diet and cardiometabolic risk profile [54]. Also, in a 2-y randomized trial, a Mediterranean diet, similar to the PHD, effectively reduced long-term body weight by 4.4 ± 6.0 kg [55].

Among cause-specific mortality outcomes, our results for respiratory mortality were the most striking in our study in both males and females. In these 3 cohorts, strong inverse associations have previously been reported between healthy dietary patterns for respiratory mortality [49] and chronic obstructive pulmonary disease (COPD) [[56], [57], [58], [59]]. Adherence to the HEI-2015, AMED, hPDI, and AHEI were associated with a reduction of 56%, 43%, 50%, and 55% risk of respiratory mortality, respectively, for every 25-percentile increment of these dietary scores [49]. Also, in these cohorts, those in the highest compared with the lowest quintiles of the AHEI had a 33% lower risk of developing COPD [57], and a similar inverse association was seen with the hPDI [56]. Further studies on pulmonary oxidant/antioxidant balance, metabolomics, and inflammatory markers are needed to investigate the possible mechanisms of these dietary associations.

Regarding the environmental impacts, our results support the strong inverse association between PHDI and environmental footprints. These results are consistent with the previous findings in NHS2 that showed adherence to healthy diet patterns such as AHEI, hPDI, and PDI was associated with significant reductions in GHG emissions, use of nitrogenous fertilizer, cropland, and irrigation water [47]. In the NHANES (2005–2016) in the United States, adherence to the AHEI was associated with lower use of agricultural land, fertilizer, and irrigation water [60].

Strengths and limitations

Strengths of our study include long-term follow-up with repeated dietary assessment, the large sample size and number of endpoints, and numerous sensitivity analyses that indicated the robustness of our findings. However, the study has some limitations. First, although the PHD was designed to be globally generalizable, the scoring of foods that we used may not represent all the major foods and their relationships with mortality in all countries. Second, some measurement error of dietary intake is inevitable and will tend to bias true associations toward the null [43]. However, the dietary questionnaires used in the 3 cohorts were rigorously validated against repeated diet records and biomarkers [32,61,62]. Third, changes in diet due to development of chronic disease could lead to over- or underestimation of the effect of adherence to the PHDI on mortality. However, this difference was small based on our sensitivity analysis stopping update of the PHDI. Fourth, residual and unmeasured confounding cannot be completely controlled in this observational study. However, the restriction of participants to those with similar education and occupation limits confounding, and the results remained unchanged when controlling for geocoded neighborhood socioeconomic status z-score. Fifth, the external generalizability to the large population can be limited as mostly participants in our study are non-Hispanic White health professionals. This may limit the range of exposures, but in general, most biological risk factors are similar across racial and ethnic groups.

Conclusion

In 3 large United States-based prospective cohorts of males and females with up to 34 y of follow-up, we found that increasing adherence to a dietary pattern developed to promote both human and planetary health was associated with lower risk of total and cause-specific mortality.

Author contributions

The authors’ responsibilities were as follows – LPB, WCW: conceived the study idea, designed the research, and analyzed the data; TTP, FW, MG: provided statistical expertise and verified the analysis; LPB, WCW: drafted the manuscript; LPB: had primary responsibility for final content; and all authors: provided critical revision of the manuscript and approved the final manuscript.

Conflict of interest

The authors report no conflicts of interest.

Funding

This study was supported by the following research grants UM1 CA186107, P01 CA87969, R01 HL034594, U01 CA176726, U01 CA167552, R01 HL035464, R01 DK120870, and R01 DK126698 from National Institutes of Health (NIH). Dr. Marta Guasch-Ferre is supported by Novo Nordisk Foundation grant NNF18CC0034900. Dr. Qi Sun is supported by the NIH grants DK129670 and ES022981. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Data availability

Because of participant confidentiality and privacy concerns, data cannot be shared publicly and requests to access Nurses’ Health Studies/Health Professionals Follow-up Study data must be submitted in writing. Further information including the procedures to obtain and access data from the Nurses’ Health Studies and Health Professionals Follow-up Study is described at https://www.nurseshealthstudy.org/researchers (contact e-mail: nhsaccess@channing.harvard.edu) and https://sites.sph.harvard.edu/hpfs/for-collaborators/.

Appendix A. Supplementary data

The following are the Supplementary data to this article: