Adherence to a Planetary Health Diet, Environmental Impacts, and Mortality in Chinese Adults

Study Population

Data used in this study were from the Singapore Chinese Health Study (SCHS), which was a population-based prospective cohort study launched between 1993 and 1998. A total of 63 257 Chinese adults aged 45 to 74 years were recruited at baseline. The participants were permanent residents or citizens of Singapore who belonged to the Hokkien or Cantonese dialect group that originated either from the Fujian or Guangdong province in the southern part of China, respectively.¹⁶

For the current study, we excluded participants who had implausible energy intake (under 600 or above 3000 kcal/d for women and under 700 or above 3700 kcal/d for men; 1060 participants), and those who had self-reported diagnosis of cardiovascular diseases (3401 participants) or cancer at baseline (1718 participants), leaving 57078 participants for analysis (eFigure 1 in Supplement 1). The study was approved by the institutional review board of the National University of Singapore, and all participants gave written informed consent at enrollment. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Calculation of the PHD Score (PHD-S)

Scoring for the PHD-S was constructed using a validated food frequency questionnaire based on the EAT-Lancet report (eMethods 1 and eTable 1 in Supplement 1).⁶ All 14 dietary components were classified into 3 categories based on their health effects: adequacy, optimum, and moderation.¹⁷ In the current study, vegetables, fruits, nuts, legumes, unsaturated fats, and fish were defined as adequacy components. Potatoes, dairy, poultry, and eggs were classified as optimum components. Total grains, red meat, saturated fats, and added sugar were considered as moderation components (eTable 2 in Supplement 1). The score for each dietary component ranged from 0 to 10 and the total PHD-S ranged from 0 to 140 (highest adherence) (eMethods 2 and eTable 2 in Supplement 1). For the purpose of computing the PHD-S, we standardized the energy intake of individuals to 2500 kcal/d.

Estimation of Environmental Impacts

GHG emissions, total water footprint (TWF), and land use were estimated from the daily dietary intake of the participants using a conversion database from a study that investigated the environmental impacts of diet in the China Health and Nutrition Survey.¹⁸ Briefly, the environmental impact of the daily diet for individuals was calculated by multiplying the average environmental impact per gram of food by the amount of food consumed. GHG emissions were estimated from field to farm gate (impact of food production until the food was ready for consumption). TWF for nonaquatic foods was calculated using the database of the Water Footprint Network, and that for aquatic foods was calculated following the method in a previous study.¹⁹ Land use was estimated using data from the Food and Agriculture Organization Statistics database. GHG emissions were expressed as kilogram CO₂ equivalents, TWF as meters cubed, and land use as meters squared.

Assessment of Covariates and Mortality

At baseline, face-to-face interviews were conducted by trained interviewers using a structured questionnaire to collect information. More detailed information on covariates is available in eMethods 3 in Supplement 1. Deaths from CVD, cancer, and respiratory disease were coded according to the International Classification of Diseases, Ninth Revision (ICD-9) up to December 31, 2011, or ICD-10 from 2012 to 2020, and ascertained via linkage with the Singapore Registry of Births and Deaths through December 31, 2020 (eMethods 4 in Supplement 1).

Statistical Analysis

Linear regression models adjusted for age at baseline (years), sex (men or women), and total energy intake (kcal/d) were used to estimate the associations between PHD-S and environmental impacts. Person-years were calculated from the date of recruitment to either the date of death, loss to follow-up, or December 31, 2020, whichever came first. Cox proportional hazards regression models were used to estimate the hazard ratios (HRs) with corresponding 95% CIs for the associations of PHD-S with the risk of mortality using the lowest quintile of PHD-S as the reference group. The Schoenfeld residuals method was used to test the proportionality assumption of the Cox models and no violation was observed. In model 1, we adjusted for age, sex, and total energy intake. In model 2, we additionally adjusted for dialect group (Cantonese or Hokkien), educational level (no formal education, primary school, or secondary school or higher), BMI (continuous; calculated as weight in kilograms divided by height in meters squared), smoking status (never, former, or current), alcohol frequency (never, monthly, weekly, or daily), moderate or vigorous physical activity level (under 0.5, 0.5 to 3.9, or 4.0 or more h/wk), sleep duration (under 6, 6 to 8, or 8 h/d), and self-reported history of physician-diagnosed hypertension and diabetes. Linear trends were tested by using the median PHD-S within each quintile.

We repeated analyses stratified by age (younger than 55 years or 55 years and older), sex (using sex-specific quintiles for PHD-S in men and women), BMI (under 23 or 23 and above),²⁰ and smoking status (never or ever) to test potential variation in associations among different subpopulations. Likelihood ratio test was used to examine the potential interaction. In sensitivity analysis, we first excluded participants with history of hypertension or diabetes at baseline. Second, we excluded participants who died within 5 years from recruitment to minimize the potential reverse causality. Third, we repeated the analyses with PHD-S calculated using the method proposed by Knuppel.¹²

The statistical analyses were conducted using Stata/MP version 17.0 (StataCorp LLC). A 2-sided P < .05 was considered statistically significant. Data analysis was performed from September 2022 to April 2023.

Baseline Characteristics

Of the 57 078 participants in the SCHS, 31 958 (56.0%) were women, and the median (IQR) age was 55.0 (49.0-62.0) years. Participants with higher adherence to PHD were younger (median [IQR] age: quintile 5, 54.0 [48.0-60.0] years vs quintile 1, 57.0 [51.0-64.0] years), more likely to be women (7514 of 11 415 [65.8%] vs 5352 of 11 416 [46.9%]), more highly educated (secondary school or higher: 4178 [36.6%] vs 2281 [19.9%]), never smokers (9212 [80.7%] vs 6409 [56.2%]), nondrinkers (9533 [83.5%] vs 8785 [77.0%]), and physically active (less than 0.5 h/wk: 6956 [60.9%] vs 8234 [72.1%]) (Table 1).

This Singapore Chinese population consumed more fruits, fish, grains (mostly refined), red meat, and saturated fats and less vegetables, nuts, legumes, unsaturated fats, sugar, potatoes, dairy, and poultry than what was recommended by the PHD (eTable 3; eFigure 2 in Supplement 1). The PHD-S ranged between 13 and 95 points, with a median of 55 points (eFigure 3 in Supplement 1). Adherence to PHD was therefore considered low in this study population, and more than 80% of respondents had good compliance (more than 6 points for each dietary component) in only 3 out of 14 components, including fruits, unsaturated fats, and fish (eFigure 4 in Supplement 1).

PHD-S and Environmental Impacts

Among all the participants, the estimated median (IQR) GHG emissions, TWF, and land use from daily food consumption were 2.7 (2.5-2.9) kg CO₂ equivalents, 2.5 (2.4-2.7) m³, and 3.1 (2.8-3.5) m², respectively. Compared with the lowest quintile of PHD-S, the highest quintile had reduced GHG emissions by 7.1% (median [IQR]: Q5, 2.6 [2.4-2.8] kg CO₂ equivalent vs Q1, 2.8 [2.6-3.0] kg CO₂ equivalent), and increased TWF by 8.3% (2.6 [2.4-2.7] m³ vs 2.4 [2.3-2.6] m³) and land use by 10.3% (3.2 [2.9-3.7] m² vs 2.9 [2.6-3.3] m²) (Table 1). Higher quintiles of PHD-S were associated in a stepwise manner with lower GHG emissions (β = −0.13 kg CO₂ equivalent; 95% CI, −0.14 to −0.12 kg CO₂ equivalent), higher TWF (β = 0.12 m³; 95% CI, 0.11 to 0.13 m³), and more land use (β = 0.29 m²; 95% CI, 0.28 to 0.31 m²) (all P for trend < .001) (Table 2).

The dietary components that contributed the most to GHG emissions included total grains (54.61%), fish (10.95%), and red meat (9.22%) (Figure 1). Plant-based food that contributed the most to TWF and land use was total grains (37.25% of TWF and 34.39% of land use), followed by fruits, which was responsible for 8.61% of TWF and 10.28% of land use. Red meat was also resource-intensive and was responsible for 10.07% of TWF and 10.86% of land use. Aside from red meat, animal-based food with the highest land use included dairy (10.35%), poultry (8.38%), and fish (5.87%) (Figure 1). Higher PHD-S was associated with overall lower GHG emissions, and this was largely explained by decreased contribution of GHG emissions from reduced consumption of total grains (quintile 5, 1.32 kg CO₂ equivalent vs quintile 1, 1.65 kg CO₂ equivalent) and red meat (quintile 5, 0.21 kg CO₂ equivalent vs quintile 1, 0.27 kg CO₂ equivalent) in higher PHD-S quintiles (Figure 2; eTable 4 in Supplement 1). In contrast, higher PHD-S was associated with overall higher TWF and land use, and this was mainly due to increased contribution of TWF and land use from greater consumption of fruits, dairy, vegetables, and legumes in higher PHD-S quintiles (eFigures 5 and 6 and eTables 5 and 6 in Supplement 1).

Association of PHD-S With Mortality

During 1 211 192 person-years of follow-up (median [IQR] follow-up, 23.4 [18.7-26.2] years), a total of 22 599 deaths were documented (39.6%), including 6948 CVD deaths (3830 from ischemic heart disease, 1866 from stroke), 7338 cancer deaths, and 4891 respiratory disease deaths [3944 from pneumonia, 807 from chronic obstructive pulmonary disease (COPD)] (Table 3; and eTable 7 in Supplement 1).

In model 1, higher PHD-S was significantly associated with lower risk of all-cause, CVD, cancer, and respiratory disease mortality (all Ps for trend <.001) (Table 3). The associations were attenuated but remained significant after additional adjustment for other potential confounders. In model 2, participants in the highest quintile of PHD score had lower risk of all-cause mortality (HR, 0.85; 95% CI, 0.81-0.89; P < .001), CVD mortality (HR, 0.79; (95% CI, 0.73-0.85; P < .001), cancer mortality (HR, 0.93; 95% CI, 0.86-1.00; P = .04), and respiratory disease mortality (HR, 0.81; 95% CI, 0.74-0.89; P < .001) (Table 3).

We also estimated the associations between PHD-S and subtypes of mortality from CVD and respiratory disease. Participants in the highest quintile, compared with those in the lowest quintile, had an 11% to 35% lower risk of ischemic heart disease, stroke, pneumonia, and COPD mortality (eTable 7 in Supplement 1).

The stepwise reduction in mortality risk with increasing quintile of PHD-S was greater for individuals who had ever smoked than in those who had never smoked (P for interaction = .01); the HR comparing extreme quintiles was 0.80 (95% CI, 0.75-0.86) in ever-smokers vs 0.86 (95% CI, 0.81-0.91) in never-smokers. The stepwise reduction in mortality risk with increasing sex-specific quintile of PHD-S was higher in women than in men (P for interaction = .04); HR between extreme quintiles was 0.86 (95% CI, 0.81-0.91) in men vs 0.83 (95% CI, 0.78-0.88) in women (eTable 8 in Supplement 1). In sensitivity analyses, the results remained materially unchanged (eTables 9 and 10 in Supplement 1).

Limitations

This study had several limitations. First, the dietary data was only collected at baseline and subsequent changes of dietary intake during follow-up were not measured. However, given the prospective design, these changes may likely lead to nondifferential misclassification and underestimation of associations. Second, the methods used to evaluate the environmental impacts in the current study were derived from a previous study conducted in China,¹⁸ which might not be generalizable to Singapore. Third, given that several of the covariates (age, sex, education, and smoking) had associations with PHD-S, the association between PHD-S and mortality was susceptible to confounding and, as expected, the effect size of the associations was attenuated after further adjustment for potential confounders. Hence, the results of this study and causal inferences should be interpreted with caution as this study is observational in design and residual confounders cannot be completely ruled out. Fourth, as the study participants were Singapore Chinese, it might limit the generalizability of the findings to other populations.