Association between Diet-Related Greenhouse Gas Emissions and Mortality among Japanese Adults: The Japan Collaborative Cohort Study

Abstract

Background:

Planetary and human health are highly intertwined; our current food system is associated with high greenhouse gas emissions (GHGE) and burden of disease.

Objective:

The aim of this study was to investigate the associations of diet-related GHGE with all-cause and cause-specific mortality in Japan.

Methods:

This study included 58,031 Japanese adults (35,078 women and 22,953 men) 40–79 y of age who participated in the Japan Collaborative Cohort Study during the period 1988–1990. Diet-related GHGE was calculated from dietary intake estimated by a validated food frequency questionnaire and previously developed GHGE tables of each food and beverage. Participants were classified into quintiles of diet-related GHGE per kg food/d. Hazard ratios (HRs) of all-cause and cause-specific mortality were calculated using the Cox proportional hazard and restricted cubic spline models.

Results:

The average diet-related GHGE was $\mathrm{1,522}{\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{kg}\text{\hspace{0.17em}food}/\mathrm{d}$. Over a period of 19.3 y (955,819 person-years) of median follow-up, 11,508 deaths were documented. After adjusting for lifestyle and medical history, in comparison with the fourth quintiles of diet-related GHGE, the first and fifth quintiles were associated with a higher risk of all-cause mortality: multivariable HR of all-cause mortality was 1.11 [95% confidence interval (CI): 1.05, 1.18] and 1.09 (95% CI: 1.03, 1.17) for the lowest and highest GHGE, respectively; those of cardiovascular disease mortality were 1.23 (95% CI: 1.10, 1.38) and 1.22 (95% CI: 1.08, 1.37), respectively. The diet-related GHGE range with the lowest HR of all-cause mortality was $\mathrm{1,400}–\mathrm{1,600}{\mathrm{g}\text{-}\text{CO}}_2\text{eq}/\mathrm{kg}$ food/d ($p$ for nonlinearity $<0.001$). Replacing one serving of red meat with one serving of pulses was inversely associated with all-cause mortality ($\text{HR}=0.96$; 95% CI: 0.93, 0.99) and GHGE (mean change, $-347{\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{kg}/\mathrm{d}$; 95% CI: $-353$, $-342$).

Discussion:

Diet-related GHGE was associated with all-cause and cardiovascular disease mortality in a U-shaped fashion. This finding could be useful for creating a policy for sustainable shifts in dietary habits that will benefit the population and environmental health. https://doi.org/10.1289/EHP14935

Introduction

Climate change is the biggest global health threat of the 21st century¹ because it adversely affects lifestyle through food shortages, extreme weather conditions, sea-level rise, and reduced productivity.² The Paris Agreement, signed by 196 nations in 2015, legally binds countries to work toward limiting global warming to below $2°\mathrm{C}$ above preindustrial levels³ and requires them to submit climate action plans.⁴ In the Japanese action plan, the original target was a 26% reduction in greenhouse gas emissions (GHGEs) by 2030, although this target has recently been increased to 46% to achieve carbon neutrality by 2050. To meet the goals of the Paris Agreement, a reduction in GHGEs not only from fossil fuels but also from agriculture and food production is required.⁵ Given that the Japanese plan does not include a policy concerning the dietary dimension of environmental change, clarification of diet-related GHGEs is essential to achieve the goals. Globally, when pre- and postproduction processes of the food system are considered, diet-related GHGEs constitute 21%–37% of the total GHGEs.^6,7 Hence, improving dietary choices is imperative to reduce GHGE levels and contribute to planetary health.^8,9

In a meta-analysis of seven cohort studies conducted in Western countries, the highest diet-related GHGE level was associated with a higher all-cause mortality rate than the lowest level.¹⁰ However, because dietary patterns vary among cultures and countries, it is necessary to examine how a reduced GHGE level affects health in non-Western countries. In the context of GHGEs, although previous studies have not specified the foods and beverages that are associated with mortality risk,^11–13 reporting the health and ecological impacts of substituting proteins associated with high GHGEs with those associated with low GHGEs could help address human and planetary health priorities.

Therefore, this study aimed to evaluate the association between diet-related GHGEs and all-cause and cause-specific mortality in Japanese adults. Considering previous reports,^11–13 we hypothesized that there would be a U-shaped relationship between diet-related GHGEs and mortality and that pulse consumption would be beneficial to both planetary and human health.¹⁴

Methods

Study Design and Population

Data from Japanese adults who participated in the Japan Collaborative Cohort (JACC) Study (a large-scale population-based cohort study) were used. Details of the study are described elsewhere.¹⁵ The JACC study is a prospective cohort study in which 110,585 residents 40–79 y of age lived in 45 areas during the recruitment period between 1988 and 1990 (Figure 1). This investigation was a multicenter collaborative study in which 24 institutions voluntarily participated. The recruitment of the study participants fell to each investigator, who had the responsibility to construct a cohort in each area. The majority of participants included in this study (42 of 45 regions) were invited by all residents living in a given target area or those who had undertaken a basic health examination. The follow-up of this study was completed at the end of 2009. Dietary intake, smoking, alcohol consumption, and medical history were obtained with a self-administered questionnaire at baseline.

Figure 1.

Participant flow diagram for the analysis of diet-related greenhouse gas emissions and mortality in Japanese men and women 40–79 y of age in the Japan Collaborative Cohort Study. Note: JACC, Japan Collaborative Cohort Study.

Of these participants, we excluded residents living in areas not surveyed for 40-item food frequency questionnaires (FFQ) ($n=\mathrm{30,220}$), individuals with five or more missing items in the FFQ ($n=\mathrm{13,531}$), those who reported extreme energy intake ($<500\text{\hspace{0.17em}kcal}$ or $>\mathrm{3,500}\text{\hspace{0.17em}kcal}$) ($n=\mathrm{5,859}$), and those with a history of cancer, stroke, or myocardial infarction at baseline ($n=\mathrm{2,944}$). Finally, 58,031 participants (35,078 women and 22,953 men) were included in this study.

This study was carried out following the Declaration of Helsinki.¹⁶ In 36 regions, written informed consent was obtained individually and directly from study participants. In nine areas, individual consent was not obtained but considered to be obtained when they answered the questionnaire, following the approval for the study protocol by the internal Ethics Committee of the JACC Study Group and the community leaders and mayors after explaining the study purpose and data confidentiality. The waiver of individual consent was approved because a) the research involves no more than minimal risk of harm to the participants, b) the waiver will not adversely affect the rights and welfare of the participants, c) all participants were guaranteed the opportunity to refuse their participation in the research, and d) the research could not practically be conducted in the local context without the waiver. The last reason was well discussed with the community leaders and mayors, and it was decided to be appropriate. This study protocol was first evaluated and approved by the Ethics Committee of Nagoya University School of Medicine as an external ethics committee in 2000 when the study protocol review by an ethics committee became a requirement in Japan before starting human research. Recently, because of a change in the affiliation of the principal investigator (A.T.), the overall study protocol was approved by the Ethics Committee of Hokkaido University (approval no. 14–044). Moreover, the use of data by the first author (D.W.) was approved by the Ethics Committee of Waseda University (approval no. 2023–441).

Dietary Assessment and Calculation of Diet-Related Greenhouse Gas Emissions

Dietary intake was evaluated using a validated self-administered food-based FFQ comprising 40 food and beverage items.¹⁷ The participants were asked about the frequency of consumption of each food and beverage item in the previous year. The answer options were the following five categories: a) never or seldom, b) 1–2 times per month, c) 1–2 times per week, d) 3–4 times per week, and e) once per day. These frequencies were assigned 0, 0.05, 0.21, 0.5, and 1.0 points, respectively. The intake levels of each food and beverage item were calculated by multiplying the intake frequency by the portion size. Energy and nutrient intakes were obtained from the food and beverage intake levels using the Standard Tables of Food Composition in Japan (fifth revised and enlarged edition).¹⁸ To calculate the consumption of each food group, food items were categorized as follows: cereals (white rice); potatoes (potatoes); sugar (sugar added to coffee and black tea); pulses (boiled beans and tofu); vegetables (fried vegetables, spinach, or garland chrysanthemum, carrot or pumpkin, tomatoes, cabbage or head lettuce, Chinese cabbage, edible wild plants, pickles); fruits (citrus and noncitrus fruits); mushrooms (fungi); seaweeds (algae); tsukudani (preserved foods using soy sauce); fish (fresh fish, kamaboko, and dried or salted fish); meat [beef, pork (excluding ham or sausage), ham or sausage, chicken, and liver]; eggs (eggs); dairy products (milk, yogurt, and cheese); fat and oil (butter, margarine, and creamer added to coffee and black tea); confectioneries (sweets); alcoholic beverages (Japanese sake, shochu, beer, whiskey, and wine); nonalcoholic beverages (fresh fruit juice, coffee, black tea, Japanese green tea, and oolong tea); seasonings (miso soup); and cooked foods (deep-fried foods or tempura).

Diet-related GHGE (${\mathrm{g}\text{-}\text{CO}}_2\text{-}\text{eq}/\mathrm{d}$) was calculated by summing the products of the intake level and GHGE value for each food and beverage item. The GHGE values were estimated using the production-based Japanese input-output table (IOT)–applied method for each food and beverage item listed in the Japanese Food Composition Tables.¹⁹ This GHGE database includes 354 foods for the production-based method. Our use of this database involved retrieval of data for components of emissions, including those related to food production and processing, but not marketing- and waste processing-related data. In addition, GHGE values of imported products were calculated based on the assumption that their environmental impact is equivalent to that of domestically produced products of the same type. For the food and beverage items comprising multiple foods and beverages in our FFQ, diet-related GHGEs were determined by summing the GHGEs of each item with a weight that was based on the consumption level of the respective item in the 1988–1990 National Survey of Family Income and Expenditure in Japan²⁰ when the FFQ survey was conducted (Supplementary Table 1). Diet-related GHGE was expressed as ${\mathrm{g}\text{-}\text{CO}}_2$ equivalents per kg food weight per day (${\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{kg}/\mathrm{d}$), according to a previous study.¹¹

Diet-related GHGEs per kg food/d were divided into quintiles. The contribution and absolute value of the diet-related GHGEs for each food and beverage item were evaluated. The dietary intake according to the quintile group for diet-related GHGEs was presented as the median intake per 1,000 kcal by the energy-density method²¹ because the planetary health diet by the EAT-Lancet Commission used the energy-density method.²² The association between diet-related GHGEs and dietary intake was confirmed using Spearman’s correlation analysis.

Other covariates.

All covariates were obtained from questionnaire data from a baseline survey. Data were collected on the following basic characteristics: age (years); sex (male or female); height (centimeters); body weight (kilograms); smoking status (“Do you smoke?”: never smoker, past smoker, or current smoker); drinking status (“Do you drink alcohol?”: never drinker, past drinker, or current drinker); sleep duration (“How much time do you sleep on weekdays?”: minutes); watching TV (“How much time do you watch TV on average per day?”: minutes); education attainment (“How old did you go to school?”: years); occupation status (“What is your occupation status?”: regular work, part-time job, self-employed, homemaker, unemployed, or others); marital status (“What is your marital status?”: married, divorced, single, or widowed); sports or exercise status (“How long do you play sports or exercise in a week on average?”: never, $<1$, 1–2, 3–4, or $\ge 5\mathrm{h}/\mathrm{wk}$); walking status (“How long do you spend walking indoors or outdoors on average per day”: never, $<30$, 30–60, or $>60\text{\hspace{0.17em}min}/\mathrm{d}$); diabetes (“Do you have diabetes?”: yes, no); hypertension (“Do you have hypertension?”: yes, no). Body mass index was calculated using self-reported height and weight as body weight divided by the height squared (${\mathrm{kg}/\mathrm{m}}^2$).

Mortality Surveillance

For the follow-up information, the dates and causes of death were confirmed by systematically reviewing death certificates. The underlying causes of death were coded for the National Vital Statistics database according to the International Classification of Diseases, 10th revision (ICD–10). Cause-specific mortality was determined separately for cancer (C00–C97), total cardiovascular disease (CVD) (I01–I99), and respiratory disease (J00–J99). The date of moving out of the study area was verified by referring to the local governmental office. After leaving their original communities, the relevant participants were treated as censored cases. Among the 45 areas included in the survey, the follow-up survey was discontinued at the end of 1999 in four areas, in 2003 in another four areas, in 2008 in two areas, and in 2009 in the remaining 35 areas.¹⁵ We calculated each participant’s person-years of follow-up from the return date of the baseline questionnaire to the date of death, relocation from the study area, or the end of follow-up, whichever occurred first. The rates of all-cause and cause-specific mortality for each quintile group of diet-related GHGEs are shown as the number of events per 1,000 person-years.

Statistical Analysis

For descriptive statistics of participants’ characteristics, continuous variables are presented as means and standard deviations for normally distributed variables and median and interquartile range for nonnormally distributed variables. Categorical variables are presented as numbers and percentages. Missing data regarding body mass index ($n=\mathrm{2,078}$; 3.6%), smoking status ($n=\mathrm{3,192}$; 5.5%), alcohol consumption status ($n=\mathrm{1,371}$; 2.4%), marital status ($n=\mathrm{2,917}$; 5.0%), occupational status ($n=\mathrm{3,157}$; 5.4%), educational attainment ($n=\mathrm{4,753}$; 8.2%), time spent watching television ($n=830$; 1.4%), sleep duration ($n=\mathrm{1,587}$; 2.7%), sports or exercise status ($n=\mathrm{1,829}$; 3.2%), walking status ($n=\mathrm{3,622}$; 6.2%), history of diabetes ($n=\mathrm{3,643}$; 6.3%), and history of hypertension ($n=\mathrm{2,911}$; 5.0%) were handled by creating missing indicators.

A multivariable Cox proportional hazards model was used to calculate the hazard ratios (HRs) and 95% confidence intervals (CIs) for all-cause and cause-specific mortality according to diet-related GHGEs. The reference level for diet-related GHGEs was set to the fourth quintile because previous studies have indicated a J- or U-shaped relationship between diet-related GHGEs and mortality.^11–13 The assumptions of the Cox proportional hazards model were confirmed using the Schoenfeld residual test ($p=0.21$). Furthermore, to evaluate the curvilinearity of the associations, we used a multivariable-adjusted restricted cubic spline model with four knots (fifth, 35th, 65th, and 95th percentiles), setting $\mathrm{1,522}{\mathrm{g}\mathrm{CO}}_2\text{-eq}/\mathrm{kg}/\mathrm{d}$ (population average) diet-related GHGEs as the reference value. The statistical significance of nonlinearity was assessed using the Wald test, comparing the likelihood ratio of the spline model with the linear model, and $p$-values $<0.05$ were regarded as indicating a statistically significant nonlinear relationship between exposure and outcome.²³ Linear trend $p$-values were calculated using the likelihood ratio test with continuous variables of diet-related GHGEs. The risk of mortality and change in GHGE level associated with the replacement of one serving of red meat (beef, pork, or liver) with one serving of an alternative protein source was calculated using the substitution method^24,25 because reducing meat consumption may benefit both a reduction in GHGEs and health outcomes.⁴ The substitution analyses were based on the leave-one-out model, where the substitution models included the substitute (processed meat, chicken, fish, eggs, dairy products, and pulses) as a separate variable and a sum variable, including the substitute and the substituted food (red meat).²⁶ The derived HRs and 95% CI of the substitutes were interpreted as the estimated mortality risk related to increased intake of each other protein source-based substitute and a concurrent decrease in the substituted red meat intake. In this model, the decrease in red meat intake considered in the substitution models was $41\mathrm{g}/\mathrm{d}$ per serving,¹⁷ with corresponding increases in serving sizes for intakes of other protein source-based substitutes. The serving size of each substituted protein source was $22\mathrm{g}$ for processed meat, $43\mathrm{g}$ for chicken, $42\mathrm{g}$ for fish, $37\mathrm{g}$ for eggs, $97\mathrm{g}$ for dairy products, and $45\mathrm{g}$ for pulses.¹⁷

To confirm the stability of our findings, a sensitivity analysis was performed according to the following four methods: a) to examine the possibility of reverse causality, we excluded death events recorded in the first 5 y of follow-up (1,015 men and 656 women); b) we excluded participants with a history of diabetes and hypertension, c) we performed a similar analysis using a dataset in which multiple imputations supplemented missing values for covariates²⁷; and d) we used diet-related GHGEs per $\mathrm{1,000}\text{\hspace{0.17em}kcal}$ to examine whether the assessment method for diet-related GHGEs affects the associations. Multiple imputation analysis pooled the results from 20 datasets created with random numbers using the multiple imputation method to supplement the missing values of covariates. The “mi estimate” command in STATA MP (version 15.0; StataCorp) was used. All missing values were presumed to be missing at random.

Multivariable analysis was conducted by modeling potential confounding factors reported in previous studies.^11,12 Model 1 was adjusted for age (continuous), area, and sex (male or female). Model 2 was additionally adjusted for body mass index ($<18.5$, 18.5–24.9, 25–29.9, $\ge 30{\mathrm{kg}/\mathrm{m}}^2$, or missing), smoking status (never smoker, past smoker, current smoker, or missing), alcohol consumption (never drinker, past drinker, current drinker, or missing), occupational status (regular work, part-time job, self-employed, homemaker, unemployed, others, or missing), educational attainment (school up to an age of $<15$, 15–18, or $\ge 19\mathrm{y}$, or missing), marital status (married, divorced, single, widowed, or missing), energy intake (continuous), time spent watching television ($<2$, 2 to $<3$, 3 to $<4$, 4 to $<5$, $\ge 5\mathrm{h}/\mathrm{d}$, or missing), sleep duration ($<6$, 6 to $<7$, 7 to $<8$, $\ge 8\mathrm{h}/\mathrm{d}$, or missing), sports or exercise status (rarely, 1–2, 3–4, $\ge 5\mathrm{h}/\mathrm{wk}$, or missing), walking status (rarely, $<30$, 30–60, $>60\text{\hspace{0.17em}min}/\mathrm{d}$, or missing), history of diabetes (yes, no, or missing), and history of hypertension (yes, no, or missing).

Statistical significance was set at a two-tailed $p$-value $<0.05$. All statistical analyses were performed using STATA MP (version 15.0).

Results

Table 1 shows the participant characteristics by quintile of diet-related GHGEs in the analyzed cohort. The average diet-related GHGE level was $\mathrm{1,522}{\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{kg}/\mathrm{d}$. Diet-related GHGE levels were positively associated with the proportions of alcohol consumers and married individuals. Individuals with a diet lower in GHGEs were less likely to engage in sports or exercise, were less likely to have a walking habit, and were more likely to have hypertension.

Table 1.

Baseline characteristics of the study participants by quintile of diet-related greenhouse gas emissions in the Japan Collaborative Cohort Study.

	Total ($n=\mathrm{58,031}$)	Quintile of diet-related greenhouse gas emissions					$p$ For trend
		Q1 ($n=\mathrm{11,598}$)	Q2 ($n=\mathrm{11,557}$)	Q3 ($n=\mathrm{11,659}$)	Q4 ($n=\mathrm{11,591}$)	Q5 ($n=\mathrm{11,626}$)
Age (y)a	$56.1\pm 9.9$	$57.8\pm 10.0$	$56.3\pm 9.8$	$55.8\pm 9.7$	$55.4\pm 9.7$	$55.3\pm 10.0$	$<0.001$
Sex [$n$ (%)]b
Women	35,078 (60.4)	6,701 (57.8)	6,443 (55.7)	6,641 (57.0)	7,393 (63.8)	7,900 (68.0)	$<0.001$
Men	22,953 (39.6)	4,897 (42.2)	5,114 (44.3)	5,018 (43.0)	4,198 (36.2)	3,726 (32.0)	—
Body mass index ($\mathrm{kg}/{\mathrm{m}}^2$)a	$22.8\pm 3.3$	$22.9\pm 3.2$	$22.9\pm 2.9$	$22.9\pm 2.9$	$22.8\pm 3.5$	$22.8\pm 3.9$	$<0.001$
Smoking [$n$ (%)]b
Current smoker	13,693 (25.0)	2,909 (26.3)	3,003 (27.3)	2,885 (26.1)	2,567 (23.5)	2,329 (21.5)	$<0.001$
Past smoker	6,023 (11.0)	1,261 (11.4)	1,309 (11.9)	1,318 (11.9)	1,107 (10.1)	1,028 (9.5)	—
Never smoker	35,123 (64.0)	6,870 (62.2)	6,670 (60.7)	6,850 (62.0)	7,237 (66.3)	7,496 (69.1)	—
Missing	3,192	558	575	606	680	773	—
Alcohol drinking [$n$ (%)]b
Current drinker	25,129 (44.4)	4,226 (37.3)	5,266 (46.4)	5,562 (48.7)	5,282 (46.7)	4,793 (42.6)	$<0.001$
Past drinker	1,814 (3.2)	496 (4.4)	332 (2.9)	292 (2.6)	274 (2.4)	420 (3.7)	—
Never drinker	29,717 (52.4)	6,600 (58.3)	5,743 (50.6)	5,568 (48.7)	5,757 (50.9)	6,049 (53.7)	—
Missing	1,371	276	216	237	278	364	—
Marital status [$n$ (%)]b
Married	48,914 (88.8)	9,446 (86.2)	9,756 (89.0)	10,042 (90.2)	9,937 (90.0)	9,733 (88.3)	$<0.001$
Widowed	4,452 (8.1)	1,109 (10.1)	888 (8.1)	807 (7.3)	800 (7.2)	848 (7.7)	—
Divorced	814 (1.5)	203 (1.9)	173 (1.6)	144 (1.3)	171 (1.5)	243 (2.2)	—
Single	934 (1.7)	205 (1.9)	148 (1.3)	134 (1.2)	129 (1.2)	198 (1.8)	—
Missing	2,917	635	592	532	554	604	—
Occupational status [$n$ (%)]b
Regular work	13,975 (25.5)	2,615 (24.2)	2,856 (26.5)	2,979 (27.0)	2,819 (25.5)	2,706 (24.2)	$<0.001$
Part-time job	4,140 (7.5)	709 (6.6)	755 (7.0)	847 (7.7)	918 (8.3)	911 (8.1)	—
Self-employed	14,013 (25.5)	2,913 (26.9)	2,952 (27.3)	2,852 (25.9)	2,724 (24.6)	2,572 (23.0)	—
Homemaker	10,294 (18.8)	1,585 (14.7)	1,711 (15.9)	2,074 (18.8)	2,437 (22.0)	2,487 (22.2)	—
Unemployed	9,271 (16.9)	2,279 (21.1)	1,825 (16.9)	1,677 (15.2)	1,597 (14.4)	1,893 (16.9)	—
Others	3,181 (5.8)	709 (6.6)	695 (6.4)	596 (5.4)	564 (5.1)	617 (5.5)	—
Missing	3,157	788	763	634	532	440	—
Educational attainment [$n$ (%)]b
$<15\mathrm{y}$	19,431 (36.5)	5,003 (47.9)	4,112 (39.2)	3,575 (33.3)	3,262 (30.3)	3,479 (32.1)	$<0.001$
15–18 y	26,501 (49.7)	4,439 (42.5)	5,106 (48.7)	5,592 (52.1)	5,743 (53.4)	5,621 (51.8)	—
$\ge 19\mathrm{y}$	7,346 (13.8)	1,006 (9.6)	1,271 (12.1)	1,571 (14.6)	1,749 (16.3)	1,749 (16.1)	—
Missing	4,753	1,150	1,068	921	837	777	—
Watching TV (h/d)a	$2.7\pm 1.5$	$2.8\pm 1.5$	$2.7\pm 1.5$	$2.7\pm 1.5$	$2.7\pm 1.5$	$2.8\pm 1.6$	$<0.001$
Sleep duration (h/d)a	$7.2\pm 1.1$	$7.3\pm 1.1$	$7.3\pm 1.1$	$7.2\pm 1.0$	$7.1\pm 1.0$	$7.1\pm 1.1$	$<0.001$
Energy intake (kcal/d)a	$\mathrm{1,561}\pm 449$	$\mathrm{1,506}\pm 457$	$\mathrm{1,602}\pm 463$	$\mathrm{1,609}\pm 451$	$\mathrm{1,587}\pm 437$	$\mathrm{1,501}\pm 435$	$<0.001$
Sports or exercise status [$n$ (%)]b
Rarely	41,376 (73.6)	8,593 (77.0)	8,305 (74.4)	8,176 (72.3)	7,970 (70.7)	8,332 (73.8)	$<0.001$
1–2 h/wk	8,406 (15.0)	1,393 (12.5)	1,637 (14.7)	1,761 (15.6)	1,941 (17.2)	1,674 (14.8)	—
3–4 h/wk	3,375 (6.0)	572 (5.1)	630 (5.6)	743 (6.6)	729 (6.5)	701 (6.2)	—
$\ge 5\mathrm{h}/\mathrm{wk}$	3,045 (5.4)	607 (5.4)	598 (5.4)	627 (5.5)	633 (5.6)	580 (5.1)	—
Missing	1,829	433	387	352	318	339	—
Walking status [$n$ (%)]b
Rarely	6,035 (11.1)	1,236 (11.6)	1,173 (11.0)	1,171 (10.7)	1,129 (10.3)	1,326 (11.9)	0.209
$<30\text{\hspace{0.17em}min}/\mathrm{d}$	9,668 (17.8)	1,825 (17.1)	1,810 (16.9)	1,999 (18.3)	1,980 (18.0)	2,054 (18.5)	—
30–60 min/d	11,059 (20.3)	2,096 (19.6)	2,093 (19.5)	2,288 (20.9)	2,328 (21.2)	2,254 (20.3)	—
$>60\text{\hspace{0.17em}min}/\mathrm{d}$	27,647 (50.8)	5,514 (51.7)	5,631 (52.6)	5,489 (50.1)	5,537 (50.5)	5,476 (49.3)	—
Missing	3,622	927	850	712	617	516	—
Diabetes [$n$ (%)]b
Yes	2,427 (4.5)	519 (4.7)	481 (4.4)	455 (4.2)	476 (4.4)	496 (4.6)	$<0.001$
No	51,961 (95.5)	10,465 (95.3)	10,424 (95.6)	10,497 (95.8)	10,374 (95.6)	10,201 (95.4)	—
Missing	3,643	614	652	707	741	929	—
Hypertension [$n$ (%)]b
Yes	10,963 (19.9)	2,597 (23.3)	2,291 (20.7)	2,153 (19.5)	1,992 (18.1)	1,930 (17.7)	$<0.001$
No	44,157 (80.1)	8,526 (76.7)	8,750 (79.3)	8,915 (80.5)	9,012 (81.9)	8,954 (82.3)	—
Missing	2,911	475	516	591	587	742	—
DGHGE (${\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{d}$)a	$\mathrm{2,082}\pm 526$	$\mathrm{1,589}\pm 420$	$\mathrm{1,906}\pm 458$	$\mathrm{2,065}\pm 493$	$\mathrm{2,209}\pm 532$	$\mathrm{2,636}\pm 686$	$<0.001$
(${\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{kg}\cdot \text{FW}/\mathrm{d}$)a	$\mathrm{1,522}\pm 133$	$\mathrm{1,191}\pm 81$	$\mathrm{1,353}\pm 33$	$\mathrm{1,465}\pm 33$	$\mathrm{1,603}\pm 49$	$\mathrm{1,997}\pm 278$	$<0.001$
CDGHGE (${\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{d}$)a,c	$\mathrm{3,106}\pm 785$	$\mathrm{2,371}\pm 627$	$\mathrm{2,844}\pm 684$	$\mathrm{3,080}\pm 735$	$\mathrm{3,297}\pm 793$	$\mathrm{3,934}\pm \mathrm{1,024}$	$<0.001$

Note: The ranges in Q1, Q2, Q3, Q4, and Q5 are as follows: $\le \mathrm{1,292}$; 1,293–1,408; 1,409–1,525; 1,526–1,700; and $\ge \mathrm{1,701}$ ${\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{kg}$ food weight/d for diet-related greenhouse gas emissions, respectively. Body mass index was calculated as body weight (kg) divided by height squared (${\mathrm{m}}^2$). —, no data; CDGHGE, calibrated diet-related greenhouse gas emissions; DGHGE, diet-related greenhouse gas emissions; FW, food weight; TV, television.

^a Continuous variables are expressed as mean and standard deviation, and groups were compared using the analysis of variance.

^b Categorical variables are expressed as numbers and percentages (excluding missing observations), and groups were compared using the chi-square test.

^c We calibrated the estimated diet-related GHGE from the food frequency questionnaire by multiplying by 1.49 because the energy intake estimated using the food frequency questionnaire was 33% lower than the energy intake estimated using 12-d weighed dietary records in a subcohort population of this study according to the previous study.¹⁷

In our population, cereals contributed to 19.4% of diet-related GHGEs; fish to 11.6%; and meat to 10.5% (Supplementary Table 2). Diet-related GHGEs were moderately correlated with protein ($r=0.41$), fat ($r=0.50$), vitamin D ($r=0.42$), meat ($r=0.46$), and fish ($r=0.40$) intakes but inversely correlated with carbohydrate ($r=-0.46$), manganese ($r=-0.47$), and cereal ($r=-0.49$) intakes (Table 2; Supplementary Table 3).

Table 2.

Median consumption levels of food and beverage and those correlated with diet-related greenhouse gas emissions among Japanese adults 40–79 y of age.

		Quintile of diet-related greenhouse gas emissions
Type	Total ($n=\mathrm{58,031}$)	Q1 ($n=\mathrm{11,598}$)	Q2 ($n=\mathrm{11,557}$)	Q3 ($n=\mathrm{11,659}$)	Q4 ($n=\mathrm{11,591}$)	Q5 ($n=\mathrm{11,626}$)	$r$ a
Cereals ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	297.6	362.8	316.0	289.9	265.8	256.9	$-0.49$
Potatoes ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	10.1	8.8	9.7	10.7	11.7	10.5	0.06
Sugar ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	0.0	0.0	0.0	0.0	0.0	0.0	0.06
Pulses ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	43.4	39.2	42.4	44.2	46.0	45.0	0.09
Vegetables ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	96.0	83.6	92.3	96.3	103.0	102.8	0.13
Fruits ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	52.8	47.5	57.2	56.5	54.2	48.7	0.15
Mushrooms ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	2.2	1.1	2.1	2.3	2.5	2.6	0.19
Seaweeds ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	3.2	2.4	2.9	3.4	3.8	3.6	0.16
Fish ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	24.5	14.2	21.7	26.9	32.3	31.3	0.40
Meat ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	16.7	9.7	14.1	17.2	21.3	24.7	0.46
Beef ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	1.3	0.0	1.0	1.2	3.1	4.6	0.39
Pork ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	5.1	2.3	4.7	5.3	5.8	6.1	0.23
Processed meat ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	1.0	0.6	0.8	1.5	2.5	2.5	0.27
Chicken ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	5.2	2.2	4.7	5.3	5.9	6.1	0.22
Liver ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	0.0	0.0	0.0	0.9	1.3	1.3	0.19
Eggs ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	14.0	9.9	13.3	14.5	15.8	15.7	0.15
Dairy products ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	56.6	29.6	53.6	60.7	67.1	65.8	0.13
Milk ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	61.9	42.1	58.5	65.5	70.5	70.4	0.11
Yogurt ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	0.0	0.0	0.0	0.0	0.0	0.0	0.21
Cheese ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	0.0	0.0	0.0	0.0	0.4	0.0	0.21
Fat and oil ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	0.5	0.2	0.3	0.5	0.9	1.0	0.28
Confectioneries ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	5.4	2.6	4.8	5.6	6.6	7.2	0.22
Alcoholic beverages ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	0.0	0.0	0.0	0.0	0.0	0.0	0.02
Nonalcoholic beverages ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	152.8	185.3	157.9	144.6	138.2	142.3	$-0.26$
Seasonings ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	8.3	8.5	8.5	8.4	8.3	8.1	$-0.06$
Cooked foods ($\mathrm{g}/\mathrm{1,000}\text{\hspace{0.17em}kcal}/\mathrm{d}$)	15.1	11.3	14.3	15.4	16.4	16.9	0.20

Note: All values are shown as medians or correlation coefficients. Food and beverage intake was adjusted for energy intake via the nutrient density method, using energy intake. Values are shown as medians in each group. Q1 through Q5 included diet-related greenhouse gas emissions of $\le \mathrm{1,292}$; 1,293–1,408; 1,409–1,525; 1,526–1,700; and $\ge \mathrm{1,701}$ ${\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{kg}$ food weight/d.

^a Spearman’s correlation analysis was used to evaluate the relationship between nutrient intake and adherence score.

The median follow-up period for all the participants was 19.3 y (interquartile range: 11.4–20.8). In total, 11,508 individuals (19.8%) died during the follow-up period (955,819 person-years). Figure 2 shows the relationship between diet-related GHGEs and all-cause and cause-specific mortality. After adjusting for confounders, in comparison with the fourth quintile of diet-related GHGEs, the lowest and highest quintiles were associated with a higher HR for all-cause mortality and CVD-related mortality: the multivariable HR for all-cause mortality was 1.11 (95% CI: 1.05, 1.18) for the lowest and 1.09 (95% CI: 1.03, 1.17) for the highest diet-related GHGE levels. The corresponding HR for CVD-related mortality was 1.23 (95% CI: 1.10, 1.38) and 1.22 (95% CI: 1.08, 1.37). These relationships became slightly weaker after excluding participants with an event in the first 5 y of follow-up but remained significant (Supplementary Table 4). Sensitivity analyses using multiple imputation or after excluding participants with a history of diabetes and hypertension did not fundamentally change the main results (Supplementary Tables 5 and 6). In addition, using diet-related GHGEs per $\mathrm{1,000}\text{\hspace{0.17em}kcal}$ instead of that per food weight did not change the primary results (Supplementary Table 7).

Figure 2.

Association between diet-related greenhouse gas emissions per food weight and all-cause and cause-specific mortality calculated using the multivariable Cox proportional hazards model among Japanese men and women 40–79 y of age in the JACC study. ($n=\mathrm{955,819}$ person-years of follow-up). The ranges in Q1, Q2, Q3, Q4, and Q5 are as follows: $\le \mathrm{1,292}$; 1,293–1,408; 1,409–1,525; 1,526–1,700; and $\ge \mathrm{1,701}$ ${\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{kg}$ food weight/day for diet-related greenhouse gas emissions, respectively. Model 1 adjusted for age, area, and sex. Model 2 was further adjusted for body mass index, smoking status, alcohol drinking, marital status, occupation status, educational attainment, time spent watching television, sleep duration, energy intake, sports or exercise status, walking status, history of diabetes, and history of hypertension. The x-axis of the plot is on the log scale. Note: CI, confidence interval; CVD, cardiovascular disease; DGHGE, diet-related greenhouse gas emissions; JACC, Japan Collaborative Cohort Study; PY, person-years; Q, quintiles; RD, respiratory disease; Ref, reference.

Figure 3 shows the dose–response relationship between diet-related GHGEs and all-cause and cause-specific mortality using a restricted cubic spline model. The diet-related GHGE range with the lowest HR for all-cause mortality was $\mathrm{1,400}–\mathrm{1,600}{\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{kg}$ food per day (U-shaped relationship; $p$ for nonlinearity $<0.001$). This spline analysis model fits the data well in comparison with the linear regression analysis (Akaike information criterion: 229,105 vs. 229,127). Similar results were observed for absolute diet-related GHGEs and in the sex-stratified analyses (Supplementary Figures 1–3). Replacing one serving of red meat with one serving of pulses was inversely associated with all-cause mortality (HR, 0.96; 95% CI: 0.93, 0.99) and GHGE levels (mean change, $-347{\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{kg}/\mathrm{d}$; 95% CI: $-353$, $-342$) (Table 3).

Figure 3.

Dose–response relationship between diet-related greenhouse gas emissions per food weight and all-cause and cause-specific mortality using a restricted cubic spline model among Japanese men and women 40–79 y of age in the JACC study. ($n=\mathrm{955,819}$ person-years of follow-up). (A) all-cause mortality, (B) cancer mortality, (C) CVD mortality, and (D) RD mortality. Solid lines represent hazard ratios, and dashed lines represent 95% CIs. As a reference, we calculated the hazard ratio using the $\mathrm{1,522}{\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{kg}$ food weight (FW)/d of diet-related greenhouse gas emissions. We estimated a $p\ge 0.05$ when the 95% CI of the hazard ratio did not exceed 1.00 and a $p<0.05$ when the 95% CI of the hazard ratio exceeded 1.00. Adjustment factors included the participants’ age, areas, sex, body mass index, smoking status, alcohol drinking, marital status, occupation status, educational attainment, time spent watching television, sleep duration, energy intake, sports or exercise status, walking status, history of diabetes, and history of hypertension. Note: CI, confidence interval; CVD, cardiovascular disease; JACC, Japan Collaborative Cohort Study; RD, respiratory disease.

Table 3.

Hazard ratios for all-cause mortality and mean change in diet-related greenhouse gas emissions for replacement of one serving of various protein sources among Japanese men and women 40–79 y of age in the JACC study.

	Population averagea	Increment of one serving of various protein sources	Substitution of one serving of red meat for various protein sources
All-cause mortality
Person years	955,819
No. of deaths	11,508
Rate/1,000 PY (95% CI)	12.0 (11.8, 12.3)

Food groups	Mean (SV)	Median (SV)	HR	(95% CI)	$p$ For trend	HR	(95% CI)	$p$ For trend
Red meat	0.4	0.3	1.00	(0.94, 1.06)	0.970	—	—	—
Processed meat	0.2	0.1	0.98	(0.89, 1.08)	0.749	1.01	(0.91, 1.11)	0.880
Chicken	0.2	0.2	0.97	(0.88, 1.07)	0.545	0.99	(0.89, 1.09)	0.807
Fish	0.9	0.8	0.97	(0.93, 1.00)	0.056	0.98	(0.94, 1.02)	0.230
Eggs	0.6	0.5	1.01	(0.96, 1.06)	0.768	1.04	(0.98, 1.09)	0.198
Dairy products	0.7	0.7	0.97	(0.94, 1.00)	0.061	0.97	(0.94, 1.00)	0.105
Pulses	1.5	1.5	0.95	(0.92, 0.98)	0.003	0.96	(0.93, 0.99)	0.035
DGHGE (${\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{kg}/\mathrm{d}$)

Food groups	Mean	Median	Mean change	(95% CI)	$p$ For trend	Mean change	(95% CI)	$p$ For trend
Red meat	176	165	428	(418, 438)	$<0.001$	—	—	—
Processed meat	43	13	210	(195, 226)	$<0.001$	$-218$	($-233$, $-202$)	$<0.001$
Chicken	31	32	131	(114, 148)	$<0.001$	$-297$	($-314$, $-280$)	$<0.001$
Fish	254	233	277	(271, 283)	$<0.001$	$-151$	($-157$, $-145$)	$<0.001$
Eggs	41	34	61	(52, 70)	$<0.001$	$-367$	($-376$, $-358$)	$<0.001$
Dairy products	133	122	173	(168, 179)	$<0.001$	$-255$	($-260$, $-249$)	$<0.001$
Pulses	122	121	81	(75, 86)	$<0.001$	$-347$	($-353$, $-342$)	$<0.001$

Note: HRs for mortality and mean change in DGHGE were calculated using the Cox proportional hazards models and generalized linear model, respectively. The serving size of each protein source was $41\mathrm{g}$ for red meat, $22\mathrm{g}$ for processed meat, $43\mathrm{g}$ for chicken, $42\mathrm{g}$ for fish, $37\mathrm{g}$ for eggs, $97\mathrm{g}$ for dairy products, and $45\mathrm{g}$ for pulses. Adjustment factors included age, areas, sex, body mass index, smoking status, alcohol drinking, marital status, occupation status, educational attainment, time spent watching television, sleep duration, energy intake, sports or exercise status, walking status, history of diabetes, and history of hypertension. Linear trend $p$-values were calculated using the likelihood ratio test and a continuous variable of diet-related exposure variable. —, no data; CI, confidence interval; DGHGE, diet-related greenhouse gas emissions; ${\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{kg}/\text{day}$, grams-carbon dioxide-equivalent/kilograms food weight/d; HR, hazard ratio; PY, person-years; SD, standard deviation; SV, serving.

^a Variables are expressed as the average and standard deviation of consumed serving size and diet-related greenhouse gas emissions for each protein source in this study.

Discussion

This large-scale Japanese cohort study revealed a U-shaped relationship between diet-related GHGE levels and all-cause and CVD-related mortality rates. Our findings suggest that replacing one serving of red meat with one serving of pulses is associated with a 4% lower risk of all-cause mortality and a GHGE reduction of $\sim 81\%$. To the best of our knowledge, this is the first study to evaluate the relationships between diet-related GHGE and mortality and the impact of replacing red meat with other protein sources on mortality risk and GHGEs in a non-Western country.

The average level of diet-related GHGEs was shown to be $\mathrm{1,522}{\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{kg}$ food per day and $\mathrm{3,106}{\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{d}$ for calibration. This value is similar to that reported in previous studies conducted in Western countries that examined the relationship between diet-related GHGEs, estimated using FFQs, and mortality risk.¹² A previous study indicated that the average diet-related GHGE level estimated from 4-d dietary records was $\mathrm{4,031}{\mathrm{g}\text{-}\text{CO}}_2\text{-eq}/\mathrm{d}$ in healthy Japanese adults 20–69 y of age in 2013.¹⁹ Because the per capita consumption of meat, which has a high contribution to diet-related GHGEs, was $\sim 1.5$-fold higher in 2019²⁸ than in 1990²⁹ (increasing from $71\mathrm{g}$ to $103\mathrm{g}$), the more recent diet-related GHGE levels in the Japanese population could be higher than those found in our study. However, comparing the diet-related GHGE levels found in our study with those of a previous Japanese study¹⁹ is challenging because of differences in survey years (1988–1990 vs. 2013) as well as dietary assessment methods (FFQ vs. dietary records). In addition, it may not be feasible to directly compare GHGE values between previous studies because the method used to calculate GHGE, such as the literature-based method, including a literature review of life cycle assessment studies of Japanese foods or the production- and consumption-based IOT-applied method, affects the estimated diet-related GHGE levels, regardless of the type of dietary survey used, including dietary record and FFQ.¹⁹

Several prospective cohort studies have reported U-shaped associations between diet-related GHGE levels and all-cause^11–13 and coronary heart disease-related^11,13 mortality rates. Our findings are in line with those of previous studies conducted in Western countries. Two mechanisms could explain the U-shaped association between diet-related GHGE and mortality risk. Diet-related GHGE is less likely to be derived from plant-origin foods like vegetables, fruits, and beans but more likely to be derived from animal-origin foods.¹⁹ Therefore, diets with lower GHGE could lean toward a plant-based diet, and diets with higher GHGE could lean toward to an animal-based diet. Generally, a plant-based diet is considered to be a healthy human diet,³⁰ but some plant-origin foods such as refined grain and sugars are known to be unhealthy for humans.³¹ Furthermore, diets with lower GHGE may be more likely to cause a protein deficiency without close attention to protein intake because more than half of our protein intake is from animal sources.³² Both the reduction in total protein intake and excessive animal protein intake could increase the mortality risk.³³ Furthermore, the insufficiency of some micronutrients may occur from a plant-only diet.⁶ The increase in unhealthy plant-origin food intake and the insufficiency of protein and micronutrients could be the reason for the increase in mortality risk among individuals with lower diet-related GHGE. In contrast, a diet with high GHGE is due to a higher intake of animal-origin foods. Among those animal-origin foods, red and processed meats have a high contribution to GHGE in comparison with other types of animal-origin foods such as chicken.¹⁹ As well known, excessive intake of red and processed meats, leading to higher diet-related GHGE, could increase the mortality risk.^25,34 Those mechanisms mentioned above suggest that simply lowering an individual’s diet-related GHGEs may not always benefit human health.

Our results indicate that daily replacing one serving of red meat with one serving of pulses is inversely associated with all-cause mortality and GHGE levels. Given that the diet must meet the body’s energy and protein requirements, human health cannot be achieved simply by reducing foods with high GHGE levels. Both human and planetary health should be considered by replacing unhealthy foods with healthy foods with lower GHGE levels.²⁴ A previous study conducted among US adults showed that daily replacing one serving of red meat with one serving of pulses is associated with a 6% lower risk of all-cause mortality²⁵—our findings are similar to those of the aforementioned study²⁵ and another study conducted in Japan.¹⁴ The GHGEs related to beef consumption are reported to be 18–25 times higher than those related to soybean and grain consumption.³⁵ Therefore, for the same amount of protein intake, soybeans and grains have lower GHGEs than beef has. Two modeling studies have shown that a reduction in red meat consumption accompanied by an increase in fruit, vegetable, and legume consumption will reduce GHGEs and contribute to life-years gained.^4,36 This finding corroborates our results. Therefore, the replacement of red meat with pulses as a protein source may be beneficial for both human and planetary health. However, future studies should reevaluate our results for the purpose of generalization, given that pulses may be associated with higher GHGEs than meat due to transportation requirements in certain countries or areas, and substituting pulses for meat may be difficult because of cultural and social backgrounds.³⁷

The strengths of this study are its prospective design, long follow-up period, and large sample size. However, it also has some methodological limitations. First, dietary intake estimated using a self-administered questionnaire may have been affected by systematic error related to individual characteristics,³⁸ which hinders the accurate evaluation of food and beverage consumption. Moreover, our results may be an underestimation of diet-related GHGEs because of the use of a limited 40-item FFQ. Indeed, the average energy intake in our study was $\mathrm{1,561}\text{\hspace{0.17em}kcal}/\mathrm{d}$, possibly because of the lack of bread and noodles in our FFQ. In participants of the Japan Public Health Center-Based Prospective Study, which began in the period 1990–1993, the mean bread and noodle intakes estimated by 28- or 14-d dietary records were reported as 20.7 and $62.6\mathrm{g}/\mathrm{d}$, respectively.³⁹ Caution should be exercised when interpreting the absolute value of diet-related GHGE. Moreover, we could not consider foods associated with a planetary health diet, such as nuts and whole grains.²² Although these foods were rarely consumed among Japanese people in 1988–1990,²⁹ the accuracy of the diet-related GHGE levels calculated using FFQs needs to be confirmed against that calculated using dietary records. Second, we may have underestimated the diet-related GHGE levels because, in most cases, the reference values used for GHGEs per weight of each food item did not include the GHGEs from international transportation. If the production processes of imported products and domestic products are significantly different, the actual situation of imported products may deviate from the estimated environmental burden. This deviation occurs because the database assumes equivalence to that of domestically produced products of the same type. Thus, these GHGE reference values depend on the classification of food items and the quality of the price data; misclassification of food items or inaccuracies in unit prices may lead to random errors in estimating diet-related GHGEs.¹⁹ Third, diet-related GHGEs were assessed only once at baseline during the period 1988–1990 ($\sim 35\mathrm{y}$ ago). Participants’ diet-related GHGEs may have changed during the follow-up period, leading to the misclassification of diet-related GHGEs, and this kind of misclassification in the exposure assessment may have weakened the relationship between diet-related GHGEs and mortality. However, this study confirmed a U-shaped relationship between diet-related GHGEs and mortality in both the main and sensitivity analyses. Finally, although the present study included an adjustment for various confounders, there may have been residual confounding factors in the association between diet-related GHGEs and mortality.

In summary, this study revealed a U-shaped relationship between diet-related GHGEs and all-cause and CVD-related mortality. Replacing red meat with pulse consumption may be beneficial in terms of both a reduction in GHGEs and health outcomes. These findings suggest that a small improvement in an individual’s current diet could help reduce GHGEs and the risk of diet-related death. This knowledge is useful for creating dietary guidelines aimed at sustainable dietary shifts that benefit both the health of the population and the environment.

Conclusions and opinions are those of the individual authors and do not necessarily reflect the policies or views of EHP Publishing or the National Institute of Environmental Health Sciences.