Healthy diets can create environmental trade-offs, depending on how diet quality is measured

The global imperative to improve diet quality while simultaneously reducing environmental burdens is one of society’s most pressing challenges today. Suboptimal diet quality is now the leading behavioral risk factor for premature death [1], accounting for over 11 million deaths worldwide [2]. Diets high in sodium, and low in whole grains, fruits, vegetables, and nuts and seeds account for a substantial share of this burden due to the detrimental effects on vascular integrity, metabolic regulation, and gastrointestinal health [2]. These conditions play a major role in the etiology of coronary heart disease, stroke, diabetes, and neoplasms, which cumulatively represent the predominant share of premature death and disability worldwide [2]. According to recent estimates, improved diet quality could avert nearly 25% of premature deaths globally [3]. Meanwhile, massive amounts of agricultural resources are mobilized throughout the world to support current diet patterns, including non-renewable energy, fertilizer nutrients, pesticides, agricultural land, and water; all of which are linked with substantial environmental burdens, including greenhouse gas emissions, eutrophication, acidification, biodiversity loss, soil erosion, and water scarcity [4, 5].

In recognition of these dual health and environmental threats, the United Nations Sustainable Development Goals set ambitious targets for 2030. These include a reduction of premature mortality from non-communicable diseases (target 3.4) and sustainable management of natural resources (target 12.2) [6]. Clearly, there is an urgent need to assess the linkages between diet patterns, health outcomes, and environmental sustainability, and research has expanded greatly in this domain [7], with several high-profile studies published recently [5, 8–12].

However, there are several notable gaps in the literature that preclude a more complete understanding of the linkage between diet patterns, health outcomes, and environmental sustainability. In most cases, studies have assessed theoretical diet patterns that reflected perfect adherence to dietary recommendations [9–11, 13–16] or varying amounts of animal protein [4, 10, 15, 17, 18], rather than self-selected diets [19–22] that vary by validated measures of diet quality [23, 24]. The literature is also replete with assessments of greenhouse gas emissions at the expense of other environmental sustainability indicators [7, 25, 26]. To the best of our knowledge, no studies have evaluated the relationship between incremental differences in diet quality and multiple environmental burdens, while also accounting for the separate contributions of retail losses, inedible portions, and consumer waste.

To fill these important research gaps, the present study assesses the relationship between observed diet quality among a nationally-representative sample of over 50 thousand Americans and the amount of agricultural land, fertilizer nutrients, pesticides, and irrigation water used to produce food. We focus on the United States (US) because suboptimal diet quality is the leading risk factor for premature death and a predominant risk factor for morbidity [27], and the majority of consumed food is produced domestically [28, 29]. Thus, shifts in diet quality among Americans would have meaningful implications for environmental sustainability within US borders and beyond.

This study included 50,014 individuals ≥2 y who provided complete and reliable dietary data as determined by a trained NCHS interviewer (Supplemental Table 2). The mean overall HEI-2015 score was 58.4 (95% CI 57.7–59.0) out of 100 (Supplemental Table 3), and the mean overall AHEI-2010 score was 41.8 (41.4–42.2) out of 100 (Supplemental Table 4). Per capita Total Food Demand represented 1673 g (95% CI 1647–1699 g), and 7% (111 g, 110–112 g) was lost at the retail level (Fig. 1). Purchased food represented 1563 g (1537–1588 g), 16% (245 g, 234–256 g) of which was inedible. Of the remaining 1317 g (1300–1335 g) of edible food, 31% (410 g, 400–420 g) was wasted, and 907 g (897–917 g) was consumed.

Higher diet quality (HEI-2015 and AHEI-2010) was associated with greater per capita Total Food Demand, as well as greater retail loss, inedible portions, consumer waste, and consumed food (P < 0.001 for all comparisons; Fig. 2). By food group, higher HEI-2015 scores were associated with greater Total Food Demand for dairy, soup, nuts and seeds, fruits and vegetables, table oils and salad dressing, and salty snacks (P < 0.001 for all comparisons; Supplemental Table 5). Higher AHEI-2010 scores were associated with greater Total Food Demand for soup, grains, nuts and seeds, fruits and vegetables, table oils and salad dressing, and salty snacks (P < 0.01 for all comparisons; Supplemental Table 6).

The equivalent of 185.9 million hectares (95% CI 182.1–189.4 million hectares) of agricultural land, 7068 million kg (6923–7203 million kg) of fertilizer nutrients (N + P₂O₅ + K₂O), 243 million kg (238–247 million kg) of pesticides, and 65.2 billion cubic meters (63.9–66.5 billion cubic meters) of irrigation water were used to produce Total Food Demand on an annual basis (Fig. 3). Consumed food accounted for 57–74% of each agricultural resource category, retail loss accounted for 4–5%, inedible portions accounted for 2–15%, and consumer waste accounted for 20–23%.

Relationships between diet quality and agricultural resource use were similar between HEI-2015 and AHEI-2010 for all land use categories except for total land (fertilizer nutrients, pesticides, and irrigation water), and feed grains and oilseeds (agricultural land, fertilizer nutrients, pesticides, and irrigation water; Table 1). Greater HEI-2015 scores (i.e. greater diet quality) were associated with less agricultural land use (P = 0.042), and greater use of pesticides (P < 0.001) and irrigation water (P = 0.007; Fig. 4). Across HEI-2015 quintiles (quintile 1 = lowest diet quality, quintile 5 = greatest diet quality), permanent pasture (P = 0.037) accounted for most of the decreased trend in agricultural land use (difference of 10.8 million hectares between quintiles 1 and 5; Supplemental Table 7). No difference (P = 0.091) across quintiles was observed for fertilizer nutrients (Supplemental Table 8), largely because the positive relationship between diet quality and use of fertilizer nutrients on land used to produce fruits and vegetables (increase of 121 million kg and 129 million kg, respectively, between quintiles 1 and 5) was compensated by a negative relationship between diet quality and fertilizer nutrients on land used to produce hay (decrease of 271 million kg between quintiles 1 and 5). Fruits accounted for most of the increased trend in pesticide use (P = 0.001; increase of 35 million kg between quintiles 1 and 5; Supplemental Table 9) and irrigation water use (P = 0.001; increase of 5452 million cubic meters between quintiles 1 and 5; Supplemental Table 10).

Greater AHEI-2010 scores (i.e., greater diet quality) were associated with less agricultural land use (P = 0.006) and less use of fertilizer nutrients (P = 0.021; Fig. 5). Across AHEI-2010 quintiles (quintile 1 = lowest diet quality, quintile 5 = greatest diet quality), permanent pasture (P = 0.006) accounted for most of the decreased trend in agricultural land use (decrease of 31 million hectares between quintiles 1 and 5; Supplemental Table 11), and hay (P = 0.006) and feed grains and oilseeds (P = 0.004) accounted for most of the decreased trend in fertilizer nutrients (decrease of 731 million kg and 140 million kg, respectively, between quintiles 1 and 5; Supplemental Table 12). No difference (P = 0.862) across quintiles was observed for pesticides (Supplemental Table 13), largely because the positive relationship between diet quality and use of pesticides on land used to produce fruits and vegetables (increase between quintiles 1 and 5 of 8 million kg and 5 million kg, respectively) was compensated by a negative relationship between diet quality and use of pesticides on land used to produce hay and feed grains and oilseeds (decrease of 15 million kg and 3 million kg, respectively, between quintiles 1 and 5). No difference (P = 0.066) across quintiles was observed for irrigation water (Supplemental Table 14), largely because the positive relationship between diet quality and use of irrigation water on land used to produce fruits and vegetables (increase of 1243 million cubic meters and 1629 million cubic meters, respectively, between quintiles 1 and 5) was compensated by a negative relationship between diet quality and use of irrigation water on land used to produce hay (decrease of 4829 million cubic meters between quintiles 1 and 5).

Sensitivity analyses demonstrated that using FCID (2005–2010) to disaggregate NHANES foods (2005–2016) likely underestimated daily per capita Total Food Demand (Supplemental Table 15). When mean differences between the two time points were calculated as proportions of total food demand, there were no meaningful differences across diet quality quintiles. Separate sensitivity analyses demonstrated that the modified AHEI-2010 score was 0.390 points higher than the original score among individuals ≥18 y (P < 0.001), and no difference (P = 0.946) was observed among individuals < 18 y (Supplemental Table 16).

In this nationally-representative study of over 50 thousand Americans, we used an interdisciplinary modeling framework that integrated methods from nutritional epidemiology with food system science to evaluate the linkage between diet quality and environmental sustainability. Higher diet quality was associated with greater Total Food Demand, including greater retail loss, inedible portions, consumer waste, and consumed food. Higher diet quality was associated with lower use of agricultural land, but the relationship to fertilizer nutrients, pesticides, and irrigation water was dependent on the tool used to measure diet quality (HEI-2015 vs. AHEI-2010). Over one-quarter of the agricultural inputs used to produce Total Food Demand were attributable to edible food that was not consumed, including retail loss and consumer waste.

This study demonstrates that the relationship between diet quality and agricultural resource use depends on how diet quality is measured. Using AHEI-2010, we show that higher diet quality is associated with similar or decreased use of agricultural resources; whereas, when using HEI-2015, we show that higher diet quality is associated with tradeoffs: decreased land use, no difference in use of fertilizer nutrients, and greater use of pesticides and irrigation water. These mixed relationships reflect different approaches to measuring healthy diets, even though validation studies have shown that both indices adequately predict chronic disease risk in prospective cohorts [38, 40, 58–60]. HEI-2015 [37, 38] measures broad adherence to the Healthy Eating Patterns of the 2015–2020 Dietary Guidelines for Americans, which were derived using a food pattern modeling approach that optimized nutrient intake through consumption of a variety of foods within each food group [61]. In contrast, AHEI-2010 was constructed based on epidemiologic evidence that links specific foods and nutrients with chronic disease risk [40]. As a result, these indices use difference scoring standards, especially for fruit, meat, and dairy [38, 40], which largely explains their different associations with environmental sustainability in this study.

Compared to HEI-2015, AHEI-2010 uses more stringent scoring standards for fruit (fruit juice is scored as sugar sweetened beverage rather than a fruit), does not score dairy consumption (whereas HEI-2015 rewards greater consumption of dairy), and rewards lesser consumption of meat (whereas HEI-2015 does not score meat explicitly). Although HEI-2015 does reward greater intake of protein foods, which can indirectly reward greater meat and dairy intake among those that report consuming these foods, it also rewards lesser consumption saturated fat, which indirectly penalizes greater intake of animal-based foods (beef and dairy account of 15% and 18% of daily per capita saturated fat intake, respectively) [62]. As a result, the difference in consumption amounts between the lowest and highest AHEI-2010 quintiles in this study is + 322 g for fruit (+ 550 g using HEI-2015), − 121 g for meat (− 5 g using HEI-2015), and − 2 g for dairy (+ 106 g using HEI-2015). Consumer demand for these foods drives the amount of agricultural resources used to produce fruit, feed grains and oilseeds, and hay, which represent the predominant share of agricultural resources used to produce all food in this study. Although direction of the associations between diet quality and agricultural resource use were similar between HEI-2015 and AHEI-2010 within each land use category, the magnitude of these relationships were different (especially for fruit, meat, and dairy); which is apparent when the values for each land use category are summed to estimate results for total land area.

In a recent review, Reinhardt et al. (2020) [7] reported that a shift to the DGA recommended Healthy US-Style diet would result in similar or decreased agricultural land use, and similar or greater use of blue water, which is consistent with our findings. Few studies have evaluated the environmental impacts associated with incremental improvements in diet quality in the US using data collected directly from individuals, rather than theoretical diet patterns that reflect perfect adherence to national (or global) dietary recommendations. Others have observed that higher HEI-2015 scores were associated with less agricultural land use [63], greater food waste, and greater waste of pesticides and irrigation water [35]. Rose et al. demonstrated that individuals consuming diets with greater greenhouse gas emissions had lower HEI-2010 scores [23], although a modeling approach used by Hitaj et al. recently found that a shift from current consumption patterns to the DGA recommended diet pattern had a minimal effect (0.4% reduction) on greenhouse gas emissions without decreasing the intake of animal protein [22]. To the best of our knowledge, the present study is the first to evaluate the relationship between multiple indicators of environmental sustainability and incremental improvements in diet quality, using multiple measures of diet quality, and using dietary data collected directly from individuals.

The present study demonstrates that the relationship between diet quality and environmental sustainability is more nuanced than previously understood, and therefore challenges the notion that healthy diets are inherently more environmentally sustainable [64]. Conrad et al. demonstrated that healthier diets were associated with greater consumer food waste and associated agricultural resources [35], and Rose et al. demonstrated that diets responsible for greater greenhouse gas emissions had greater content of some micronutrients (vitamin A, vitamin D, choline, iron, calcium, and potassium) [23]. The present study also supports the main findings of several reviews that reported that healthier diets are not necessarily more environmentally sustainable [7, 65].

Fruits, vegetables, and nuts, which are key components of a healthy diet pattern, typically require substantially greater inputs per unit land area than most other foods [45]; and sugar-sweetened beverages and refined grains, which have been linked with detrimental health effects [66, 67], have relatively modest environmental impacts [5]. And although greater consumption of red and processed meat has been linked with meaningful health (and environmental) risks [7, 66, 67], the 2015–2020 Dietary Guidelines for Americans and corresponding diet quality index (HEI-2015) do not explicitly recommend limiting red and processed meat intake as part of a healthy diet. Diet quality indices vary in their measurement of these foods, which influences their associations with measures of environmental sustainability.

However, caution is warranted when ascribing a measure of environmental sustainability to any individual food product or group; what matters most is the balance of foods across an eating pattern. Thus, greater efforts are needed to assess the environmental impacts of individual, self-selected diet patterns rather than individual foods. Recently, the EAT-Lancet Commission on Healthy Diets from Sustainable Food Systems published a global recommended diet pattern to promote human health within environmental boundaries [8], and others have developed methods to convert these recommendations to a US population [68]. A logical next step for future research is to evaluate the environmental burden associated with the adoption of the EAT-Lancet diet pattern in the US.

Our findings have implications for the development of sustainable dietary guidelines in the US and beyond. First, the positive association between food loss/waste and diet quality suggests that policy aimed at improving the healthfulness of diets may need to be coupled with efforts to reduce food loss/waste to avoid unintended consequences of pursuing these aims independently. Precedent exists in the Dietary Guidelines for Americans (DGA) for providing consumer guidance beyond food choices; the DGA also include recommendations for physical activity and safe food handling [39]. Second, our finding that greater compliance with the DGA is associated with tradeoffs for agricultural resource use (lower land use but greater use of pesticides and irrigation water) illustrates that dietary recommendations may be moderated by additional considerations of environmental sustainability. In other words, the nutritionally-optimal intake of specific foods may not be optimal for the environment, or may only yield environmentally-positive impacts under certain circumstances. Human diets are significant contributors to ecological crises like climate change that require near-term, large-scale mitigation approaches [8]. As such, developing the institutional processes, political will, and interdisciplinary knowledge required to create sustainable dietary guidelines is a complex endeavor that requires balancing nutritional needs with environmental impact, but should nonetheless be high priority [69].

The strengths of this study include the assessment of individual, self-selected diet patterns differentiated by incremental differences in diet quality, rather than theoretical diets that reflect perfect adherence to dietary recommendations, which enhances the practical application of our findings [7]. The robustness of diet quality assessment was enhanced by using multiple, validated tools that consider all foods reportedly consumed, rather than utilizing a single measure of diet quality or focusing on individual food groups [70]. Diet quality assessment was also adjusted for energy intake to reduce the observed confounding effect of energy intake on food intake and sustainability outcomes [26, 70], and to adapt AHEI-2010 to a population that includes individuals < 18 y; sensitivity analyses demonstrated that this modification did not introduce any meaningful bias into the analyses. Finally, this study fills a gap in the literature by assessing multiple resource use indicators of environmental sustainability, which provides a comprehensive evaluation of the environmental impacts of diet patterns [25, 26].

The limitations of this study should also be considered. The modeling approach we used represents the US as a closed food system, such that all food demanded by consumers was produced domestically. We utilized this approach because US agencies provide high quality, publicly available data on the amount of agricultural resources used for individual crops, rather than for broad crop categories aggregated across diverse global regions. Additionally, the food loss and waste rates used in this study were sourced from LAFA, which may underestimate food loss and waste [71], and the limitations of this dataset have been noted elsewhere [72]. We recognize that food loss and waste may vary in ways not captured by our source data, but these are the most comprehensive, disaggregated, and contemporary data available at the national level. FCID was used as a crosswalk between LAFA and NHANES, but FCID has not been updated since 2010, and sensitivity analyses demonstrated that this approach underestimated daily per capita Total Food Demand by 181 g overall. No meaningful differences were observed across diet quality quintiles when these differences were expressed as a proportion of Total Food Demand mass, suggesting low likelihood of differential bias. FCID is uniquely able to disaggregate NHANES foods into sufficient resolution to be linked with individual LAFA commodities, which allows for an accounting of loss and waste for each NHANES food. Since food loss and waste represents 46% of Total Food Demand (766 g out of 1673 g), not accounting for this food would have substantially underestimated agricultural resource use. Nonetheless, efforts are needed to update FCID to link with contemporary NHANES surveys to reduce systematic bias in these analyses.

This study fills a research gap by focusing on the use of agricultural resources (notably fertilizer nutrients and pesticide use) rather than environmental impacts such as climate change [7]; yet the use of these resources is influenced by numerous factors that could not be incorporated into this analysis due to limited data availability, including the projected availability and intensity of agricultural inputs as a result of climate change [73]. The US Foodprint Model is a generalized instrument that utilizes national average data on biophysical processes, and therefore it does not explicitly incorporate characteristics of the US food system that vary spatially or by type of production system. For example, data on the application rates of agricultural amendments in certified organic operations is not part of the explicit model computations, but these data are nonetheless included in the national averages that the model uses. Additionally, the model does not estimate water scarcity, which can capture regional differences in water stress that are relevant to regional production of certain crops with high consumptive water footprints (e.g., fruits grown in the arid west). More work is needed by the federal government to create linkages between publicly available nutritional and agricultural databases, which can allow researchers to merge data on food intake with different production systems and spatial characteristics. Fortunately, efforts are underway at USDA to coordinate this transfer of nutritional and agricultural data through FoodData Central, which provides a foundation for data linkage and transfer [74], yet greater efforts are needed. Finally, measurement error cannot be ruled out when analyzing self-reported dietary data, since respondents may alter their reported food intake based on the perceived healthfulness of their food choices [75, 76]. However, self-reported dietary data provide a rich source of highly detailed information on the consumption of individual foods, and these data are necessary for estimating diet quality and comparing dietary patterns between groups [77].

By integrating methods from nutritional epidemiology with food system science into an interdisciplinary modeling framework, this study reveals that the link between diet quality and environmental sustainability is more nuanced than previously understood. Higher diet quality was linked with greater Total Food Demand, retail loss, inedible portions, consumer waste, and consumed food. Higher diet quality was associated with lower use of agricultural land, but the relationship to fertilizer nutrients, pesticides, and irrigation water was dependent on the tool used to measure diet quality; this points to the influence that diet quality indices can have on the results of diet sustainability analyses and the need for standardized metrics. Over one-quarter of agricultural resources were used to produce edible food that was not consumed (retail loss and consumer waste). Urgent policy efforts are needed to achieve national and international goals for sustainable development and waste reduction, which include strong and unified leadership, greater investment in research and programming, and facilitated coordination across federal agencies [78]. In the meantime, consumers can make meaningful progress with practical tools they already have [79]. Our findings have important implications for the development of sustainable dietary guidelines, which requires balancing population-level nutritional needs with the environmental impacts of food choices.

Not applicable.

ZC and NTB conceived of this study. All authors curated the data. ZC conducted the analysis and wrote the original draft. ZC and NTB wrote subsequent drafts, and all authors reviewed and edited the final manuscript. All authors read and approved the final manuscript.

Not applicable.

The datasets used in this study are publicly available: https://wwwn.cdc.gov/nchs/nhanes/Default.aspx;↗http://fcid.foodrisk.org/#;↗https://www.ers.usda.gov/data-products/food-availability-per-capita-data-system/↗

This study was deemed exempt from ethics review by the Institutional Review Board at William & Mary.

Not applicable.

The authors declare that they have no competing interests.

Supplementary information accompanies this paper at 10.1186/s12937-020-00629-6.