Achieving Healthy and Sustainable Diets: A Review of the Results of Recent Mathematical Optimization Studies

ABSTRACT

Climate protection and other environmental concerns render it critical that diets and agriculture systems become more sustainable. Mathematical optimization techniques can assist in identifying dietary patterns that both improve nutrition and reduce environmental impacts. Here we review 12 recent studies in which such optimization was used to achieve nutrition and environmental sustainability aims. These studies used data from China, India, and Tunisia, and from 7 high-income countries (France, Finland, Italy, the Netherlands, Sweden, the United Kingdom, and the United States). Most studies aimed to reduce greenhouse gas emissions (10 of 12) and half aimed also to reduce ≥1 other environmental impact, e.g., water use, fossil energy use, land use, marine eutrophication, atmospheric acidification, and nitrogen release. The main findings were that in all 12 studies, the diets optimized for sustainability and nutrition were more plant based with reductions in meat, particularly ruminant meats such as beef and lamb (albeit with 6 of 12 of studies involving increased fish in diets). The amount of dairy products also tended to decrease in most (7 of 12) of the studies with more optimized diets. Other foods that tended to be reduced included: sweet foods (biscuits, cakes, and desserts), savory snacks, white bread, and beverages (alcoholic and soda drinks). These findings were broadly compatible with the findings of 7 out of 8 recent review articles on the sustainability of diets. The literature suggests that healthy and sustainable diets may typically be cost neutral or cost saving, but this is still not clear overall. There remains scope for improvement in such areas as expanding research where there are no competing interests; improving sustainability metrics for food production and consumption; consideration of infectious disease risks from livestock agriculture and meat; and researching optimized diets in settings where major policy changes have occurred (e.g., Mexico's tax on unhealthy food).

Introduction

There is a critical need to address climate change, and changes to food systems are important because these contribute between 19% and 29% of global greenhouse gas emissions (GHGEs) (1). Mitigation in this domain is likely to increase the feasibility of achieving targets for limiting global temperatures, and is likely to reduce the overall cost of mitigation (2). But ideally such changes should also consider the following issues:

• The need to address other environmental issues associated with food production. These include impacts on water availability (e.g., use via irrigation) and water quality (e.g., from animal waste, fertilizer run-off, and soil erosion). Indeed, the food system accounts for ∼70% of global freshwater use (3). There is also the issue of land use and biodiversity (e.g., if pasture has replaced forests).

• The need to also improve human nutrition, particularly in terms of preventing noncommunicable diseases (i.e., ensuring sufficient intake of indispensable nutrients, such as vitamins and minerals, indispensable fatty acids and protein, and avoiding excessive levels of sodium, free sugars, and saturated fat, etc.).

• The need to enhance food security (i.e., reliable access to a sufficient quantity of affordable, nutritious food), which is a particular challenge for some low-income countries, but also has impacts for deprived groups in all countries. The increasing world population and harm to agricultural production from climate change may intensify such problems in particular areas.

One approach for identifying nutritious diets that are more sustainable is to use optimization modeling. One form of this is “linear programming” where multiple variables and constraints can be used to determine mathematically “optimal solutions.” It is a technique that has been used to solve dietary problems since the 1940s (e.g., obtaining all nutrients for the lowest cost) (4). However, the first study to use such mathematical optimization to address diet and sustainability issues together appears to have been published in 2012 (5). In the following year another such study also aimed at reducing GHGEs, meeting nutritional constraints, and reducing food costs (6). Subsequently there has been further expansion of this literature—but as yet no specific reviews on the results of this body of work have been published.

Given this background, in this review we aimed to examine the results of the most recent studies that have used optimization methods to address diet and environmental sustainability issues. We also aimed to compare the results of these studies with the findings of recent reviews of a wider range of studies on diet and sustainability. Our focus was on articles published since 1 January 2015 (with searches conducted in PubMed on 27 June 2018 and additional articles identified from bibliographies of identified studies). Key search terms included “diet” and “optimization”/“optimisation” and environmental terms (i.e., sustainable/carbon/greenhouse/methane/water/land).

We specifically excluded studies on very defined populations (e.g., the diet of students in their final year of high school in 1 city (7)); specific foods as opposed to diets (8); where hypothetical dietary shifts did not involve any optimization process (9); and where only an abstract was published, e.g., a study in India (10).

Recent Optimization Studies and What They Have Found

We identified 12 studies published since January 2015 where diets were optimized for ≥1 metric of sustainability (Table 1, which includes 2 publications from 1 study). These studies used data from China, a country in rapid economic transition, India (a low-income country), and Tunisia (a middle-income country), and from 7 high-income countries [France (n = 3 studies), Finland, Italy, the Netherlands (n = 3), Sweden, the United Kingdom (n = 3), and the United States].

TABLE 1.

Key features of the identified studies published since January 2015 that used mathematical optimization relating to human diets and ≥1 aspect of environmental sustainability¹

Study (ordered alphabetically by first author surname)	Key variables to reduce/increase or which imposed constraints	Main findings	Comments
Barré et al. 2018 (11)	Reducing each of diet-related environmental impacts (GHGEs, marine eutrophication, atmospheric acidification); agricultural coproduction; nutritional adequacy; nutrient bioavailability; acceptability; affordability	“Decreasing meat content was strictly needed to achieve more sustainable diets for French adults,” but the reduction was reduced by 6% when nutrient bioavailability and 25% when coproduction links were taken into account (e.g., coproduction included both milk and meat from dairy cattle) Food costs were lower in the more sustainable diets	A study strength in terms of realism was modeling around “departing the least from the mean observed French diet”; other strengths were the unique focus on both “nutrient bioavailability” and “coproduction”; and considering multiple environmental impacts and affordability
Gephart et al. 2016 (12)	Reducing each of: GHGEs (carbon footprint), nitrogen release (nitrogen footprint), water use (blue and green water footprint), and land use (land footprint); adequate nutrients	“We find that diets for the minimized footprints tend to be similar for the four footprints, suggesting there are generally synergies, rather than trade-offs, among low footprint diets. Plant-based food and seafood (fish and other aquatic foods) commonly appear in minimized diets” “Livestock products rarely appear in minimized diets, suggesting these foods tend to be less efficient from an environmental perspective, even when nutrient content is considered” “The results' emphasis on seafood is complicated by the environmental impacts of aquaculture versus capture fisheries, increasing in aquaculture, and shifting compositions of aquaculture feeds”	A major strength of this US study was the use of 4 environmental indicators (although they were given equal weighting which may be a problematic simplification). The issues around seafood use (aquaculture vs. capture fisheries) are also well discussed. But only 19 nutrients were considered in this study. Also for some analyses only requirements for calories, protein, carbohydrate, and fiber were involved. The authors acknowledged that “there are other aspects of environmental change not covered with these footprints including antibiotic and pesticide use, animal welfare (when applicable), biodiversity, GMO's, industrial pollution, disease risk, etc.” Also “the negative impacts of overfishing and bycatch in capture fisheries are not considered here”
Green et al. 2015 (13), with the health aspects covered further in reference (14)	Reducing GHGE; nutritional adequacy; reducing the loss of consumer welfare (reflected by the deviation between the current and the optimized diets, and the use of own-price elasticities)	If “average diets among UK adults conformed to WHO recommendations, their associated GHG emissions would be reduced by 17%. Further GHG emission reductions of around 40% could be achieved by making realistic modifications to diets so that they contain fewer animal products and processed snacks and more fruit, vegetables and cereals. However, our models show that reducing emissions beyond 40% through dietary changes alone will be unlikely without radically changing current consumption patterns and potentially reducing the nutritional quality of diets” “Our dietary optimisations show that emissions reductions can be achieved by reducing consumption of animal products, switching to meats and dairy products with lower associated emissions (e.g., pork, chicken and milk), reducing consumption of savoury snacks, switching to fruits and vegetables with lower emissions”	A novel strength was how consumer welfare was considered. Also the use of a complete life-cycle analysis of emissions specific to the UK (where possible). It noted the problem of the “fact that consumption of unhealthy foods is more likely to be under-reported than consumption of healthy foods means that we are likely to have underestimated the unhealthiness of the UK diet” In the associated publication the health aspects were quantified: “Our model suggests that it would save almost 7 million years of life lost prematurely in the UK over the next 30 years and increase average life expectancy by over 8 months” (14)
Horgan et al. 2016 (15)	Reducing GHGEs; achieving dietary recommendations	“The healthy diets and sustainable diets produced a 15 and 27% reduction in greenhouse gas emissions respectively” “Sodium proved the most difficult nutrient recommendation to meet” “While the majority had to make more substantive dietary changes to the amount of any food currently eaten and addition of new foods to their diet, for only a very small proportion of the sample foods needed to be removed from their diet.” The optimized diets involved “reductions in sweet foods (biscuits, cakes and desserts), processed meats, alcohol and white bread”. Also “total meat consumption, especially beef and lamb, decreased more to achieve sustainable diet than just a healthy diet”	Key strengths were the consideration of individual diets (n = 1491 in a UK national diet survey) and then minimizing the changes they need to make. The authors acknowledged the following limitations: GHGEs were not for the full life cycle (e.g., in processing composite foods and waste disposal); the sustainability of fishing was not considered; and the monetary cost of the dietary changes was also not included.
Kramer et al. 2017 (16)	Reducing an integrated impact covering: GHGEs, fossil energy use and land occupation; nutrient composition; a metric for popularity (weight of food consumed)	“Reducing meat is the most effective option for lowering the environmental impact of diets in all age–gender groups. Reducing alcoholic and non-alcoholic beverages is another option. Leaving out fish and dairy products are not” “[T]he preferred environmental savings come almost exclusively from consuming less meat, especially beef” “Our results indicate that alcoholic and non-alcoholic beverages are responsible for 14–16% of the total environmental impact of the current diet in the Netherlands”	A key strength of this study was the “cradle-to-grave life cycle analyses” for a large number of products (n = 207). The authors also minimized the distance relative to the current diet. However, a limitation was that they ignored “important global impacts such as overfishing and water stress.” Also diet affordability was not considered. They argued that “fish cannot be replaced, primarily because it is the exclusive dietary source of EPA and DHA”
Milner et al. 2017 (17)	Reducing GHGEs and water per person for irrigation (blue water footprint); achieving nutritional guidelines; acceptability (minimized deviation from existing patterns)	For this study of India: “The optimised diets had up to 30% lower blue water footprints and generally contained lower amounts of wheat, dairy, and poultry, and increased amounts of legumes. In the 2050 scenario, adoption of these diets would on average result in 6800 life-years gained per 100 000 total population (95% CI 1600–13 100) over 40 years. The dietary changes were accompanied by reductions in greenhouse gas emissions. The magnitude of the health and environmental effects varied between dietary patterns” GHGE changes ranged from reductions of 2% to 36%	This appears to be the first such study to particularly focus on the problem of limited freshwater availability for agriculture. It is also a relatively uncommon study in that it covers a low-income country. The health metrics are relatively advanced (life-years gained based on 9 diet-related diseases) and the Monte Carlo simulations improve the considerations around uncertainty Study limitations included the following: “Our scenarios were based on projections of population growth only and did not account for other potential drivers with more complex and uncertain effects, such as climate change and aquifer depletion. We also did not account for non-food crop production, current dietary trends, temporal variation in water availability, or the effects of socioeconomic differences between dietary patterns”
Perignon et al. 2016 (18)	Reducing GHGEs; achieving nutritional adequacy; acceptability	“30% GHGE reduction could be achieved in a nutritionally adequate diet by increasing fruits and vegetables while maintaining intake of meat/fish/eggs at approximately 100 g/d, mainly by substituting ruminant and deli meats by fish products” “Higher GHGE reductions either impaired nutritional quality, even when macronutrient recommendations were imposed, or required non-trivial dietary shifts compromising acceptability to reach nutritional adequacy” “Imposing the nutritional constraints of the ADEQ [all nutrient recommendations] scenario slightly increased the cost of the diet. High GHGE reductions (≥50%) decreased diet cost” “The maximal GHGE reduction achievable from the observed level, while respecting all the nutritional recommendations, was 69.7% for women and 74.0% for men” “High GHGE reductions resulted in the elimination of some food groups, namely Dairy and MFE [meat/fish/eggs]”	Strengths of this study were that it minimized the departure from observed diets in the French diet and included a relatively large number of foods (n = 402). It also had food costs as an outcome measure (which increased slightly for the healthier diet but decreased with higher GHGE reductions) Limitations the authors acknowledged included not covering water footprints and the sustainability concerns around fisheries
Song et al. 2017 (19)	Reducing the carbon footprint; meeting DRIs for adults; acceptability (avoiding marked deviations from current diets)	“The theoretical optimal diet reduced daily footprints by 46%, but this diet was unrealistic due to limited food diversity.” Constrained by acceptability, the optimal diet reduced the daily carbon footprints by 7–28% for men and by 5–26% for women. Seven of 8 scenarios showed that reductions in meat consumption resulted in greater reductions in GHGE. “However, dramatic reductions in meat consumption may produce smaller reductions in emissions, as the consumption of other ingredients increases to compensate for the nutrients in meat. A trade-off between poultry and other meats (beef, pork, and lamb) is usually observed”	Strengths of this work were that it was for a country outside of the high-income group (i.e., China). There was also good consideration of uncertainty (around 443 variables). The lack of China-specific carbon footprint data was a limitation. Nevertheless, the data were sourced from an LCA database derived from 1237 reviewed LCA studies of food carbon footprints along with food supply chains including crop cultivation, breeding, industrial processes, transportation, and storage
Tyszler et al. 2016 (20)	Reducing GHGEs, fossil energy use, land occupation (all combined in an integrated score); nutrient requirements; popularity of food products	“We show, by using linear programming, that it is possible to reach 30% reduction in the environmental impact with a diet which is relatively similar to the current one and could be more likely to be accepted” “Removing meat and fish from the diet reduces the environmental impact by about 21%. A healthy vegan diet reaches 30% environmental impact reduction” “The optimal solution still contains 30% of the amount of meat quantity in the Closest healthy diet, whereas amounts of dairy (liquid and cheese), fish, and egg almost remain constant. The reduced meat consumption is responsible for 60% of the reduction in environmental impact” “Reductions in meat (beef, chicken) and beverages (beer, wine) are making the most important contributions”	As above for Kramer et al, which utilized some of the same data, a key strength of this study was the “cradle-to-grave life cycle analyses” for a large number (n = 207) of food products. Multiple other environmental indicators were also used An acknowledged limitation was that the results were only for 1 population group (Dutch women aged 31–50 y).
van Dooren et al. 2015 (21)	Reducing GHGEs; achieving dietary requirements; maximizing acceptability; reducing costs	“A diet of 63 popular and low priced basic products was found to deliver all required nutrients at an adequate level for both male and female adults. This plant-based, carbohydrate and fiber-rich diet consists mainly of wholegrain bread, potatoes, muesli, open-field vegetables and fruits. The climate impact of this diet is very low (1.59 kg CO2eq/day) compared to the average Dutch diet. By constraining costs, a low carbon diet of €2.59/day is possible. Conclusions: A two-person diet consisting of 63 products and costing €37 per week can simultaneously be healthy and yet have half the average climate impact”	A strength was including food costs in the optimization, as this informs feasibility for low-income groups. Also the large number of nutrients and indicators used (n = 33) and including 206 food products. Acceptability was maximized by “maximizing the most consumed food products based on weight and minimizing absolute change in portions”. The study also showed outcomes by fossil energy use and land use (but these were not used as constraints)
Verger et al. 2018 (22)	Reducing water use and land use effects (including biodiversity); adherence to the Mediterranean diet pyramid	Preliminary results for Tunisia: “Using the first model based on the current dietary habits and practices (individual level), we found that the main dietary changes needed to satisfy all the nutrient recommendations were the increases of fruits and dairy products, and decreases of meat and starchy foods. Nevertheless, these changes increased the environmental impacts of the diets. In a scenario where environmental indicators were limited to their observed levels, the dietary changes needed were still the decreases of meat and starchy foods but also lower increases of fruits and dairy products in favor of vegetables”	This publication describes a detailed framework for considering sustainable food systems for nutrition and health in the Mediterranean region. The results presented for Tunisia (a middle-income country) are preliminary but they add to the data outside of the high-income country grouping. Some of the authors have a long track record for considering many aspects of healthy diets and sustainability
Vieux et al. 2018 (23)	Reducing GHGE; achieving nutritional adequacy	“Diet sustainability can be improved by substituting food items from the sugar/fat/alcohol food group with fruit, vegetables, and starches, and country-specific changes in consumption of animal-based products” “Although an increase in plant-based products was needed in every country, the shifts in animal-based products were not homogeneous across countries or gender,” e.g., meat consumption increased for French men—albeit poultry and pork. Also dairy consumption increased in some populations “A maximal GHGE decrease of 62–78% was theoretically achievable while still ensuring nutritionally adequacy, but at a strong risk of compromising the cultural acceptability of the diets”	A strength was data from 5 countries (France, the UK, Italy, Finland, and Sweden) along with minimizing deviation from observed diets for men and women The authors noted the limitation of just considering GHGEs and not “eutrophication, water footprint, land use or biodiversity indicators”

¹GHG, greenhouse gas; GHGE, greenhouse gas emission; LCA, life-cycle assessment.

Most studies aimed to reduce GHGEs (10 of 12) and half (6 of 12) aimed to also reduce ≥1 other environmental impact, e.g., water use, fossil energy use, land use, marine eutrophication, atmospheric acidification, and nitrogen release. All the studies included nutrient constraints in the optimization, and most (9 of 12) also aimed to maximize acceptability of dietary changes. This was usually in terms of minimal deviation from current dietary patterns but also via an estimate of the loss of consumer welfare. One study also considered the bioavailability for a few nutrients (11), and 2 also aimed to minimize or constrain food costs (11, 21), (although 1 other detailed food costs as an outcome (18)).

Half the studies (n = 6) claimed that they had no competing interests, one did report these [dairy industry funding (16)], and 5 studies had no statements about competing interests. However, our analysis (Table 2) showed that most studies (7 of 12) did have the potential for competing interests given that some authors had either: 1) coauthored work at some point that had received industry funding (meat or dairy industries); 2) had received funding from a foundation established by a food industry leader; 3) worked directly for the food industry; or 4) worked for a consultancy firm that had staff named on research outputs that had at least some food industry funding.

TABLE 2.

Additional details on the identified studies published since January 2015 that use mathematical optimization relating to human diets and ≥1 aspect of environmental sustainability¹

Study (ordered alphabetically by first author surname)	Study population, setting	Summarized aim of the optimization (key objective function/constraints, specifically on the diet component, see Table 1 for the environmental components)	Dietary data used	Issues relating to the revealed or potential competing interests (e.g., work for the food industry)
Barré et al. 2018 (11)	Adults, France	To minimize the departure from the mean observed diet at the food-item level (402 foods) and at food-group level (8 groups), combined with reducing each of the diet-related environmental impacts	Representative cross-sectional diet survey conducted between December 2005 and May 2007 by the former French Food Safety Agency. Also the French Information Center on Food Quality food composition database	Potential: The authors declared “that no competing interests exist.” Nevertheless, the fourth author (Vieux) works for MS-Nutrition, which is “a consulting firm” (http://ms-nutrition.com/en/). In this role he has authored work with meat industry funding (24). Another coauthor (Darmon) has also coauthored work with meat industry funding (24). It was stated that “The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.” Funding was from a university, from the French Research and Technology Agency, but also the consultancy MS-Nutrition (whose staff coauthor works with the meat industry—as above). Also 1 author (Perignon) “received financial support from the Daniel & Nina Carasso Foundation.” Daniel Carasso was a business leader in the food industry (Groupe Danone). However, the website of the Foundation states that: “The Foundation is a family organization, totally independent from the Danone group” (https://fondationcarasso.org/en).
Gephart et al. 2016 (12)	Population-weighted averages of the age and gender recommendations for diets, USA	To obtain adequate amounts of 19 representative micro- and macronutrients, combined with minimizing environmental footprints	Food products were selected and grouped based on the USDA Dietary Guidelines (2010) and the more detailed Harvard University Healthy Eating Plate food groups. Food composition data were from the USDA National Nutrient Database	Nil apparent: There was no statement about competing interests. All authors were university researchers. The work was funded by the NSF Graduate Research Fellowship Program.
Green et al. 2015 (13), with the health aspects covered further in reference (14)	Average adults, UK	To design nutritionally optimized diets for men and women in the UK that conformed to WHO nutritional recommendations but with minimal deviation from the current diet in terms of food groups (with constraints relating to the environmental impacts)	The UK NDNS, a rolling program of cross-sectional surveys that use a 4-d food diary. Also the UK food composition tables	Nil apparent: There was no statement about competing interests. All authors were from either universities or the “Basque Centre for Climate Change.” The work was supported by the European Commission. “The funder had no role in the design, execution or writing up of the study”
Horgan et al. 2016 (15)	Adults, UK	To minimize the changes to current intake of food groups combined with achieving dietary recommendations; and then separately to achieve dietary recommendations and a GHG reduction target. Various steps including changing food quantities, adding new foods and removing some foods.	Adult data in the UK NDNS; nutrient composition data were also from the NDNS	Nil apparent: The authors declared “that they have no competing interests.” The authors were university researchers and researchers with a government funded research institute. The work was funded by the Scottish Government.
Kramer et al. 2017 (16)	Men and women aged 9–69 y, Netherlands	To minimize changes to the current diet such that nutrient requirements (30 micro- and macronutrients) were reached along with satisfying the relevant environmental constraints. In addition, the weight of food (grams) consumed in the survey data was used as a proxy for popularity with proportional penalty weightings applied in the modeling with reduced consumption	The Dutch National Food Consumption Survey 2007–2010. The macro- and micronutrient composition of foods was obtained from the Dutch Food Composition Database	Potential: The following statement was provided: “Conflict of interest: G.F.H.K. and H.B. are employed by Blonk Consultants, a private company doing projects for non-governmental organizations, private companies in the animal feed industry, the food and beverage industry, and public institutions.” More specifically, the lead author (Kramer), second author (Tyszler), and the fourth and final author (Blonk), were working for this consultancy. In particular, the work was funded by a dairy company, albeit with a statement that the company “had no role in the design, analysis or writing of this article”
Milner et al. 2017 (17)	Adult population, India	To achieve WHO nutritional guidelines for carbohydrates, fats, free sugars, protein, sodium, fruits, and vegetables with no change in total dietary energy, combined with reducing blue water use in agriculture (all with minimizing deviation from existing dietary patterns)	Dietary patterns were based on the Indian Migration Study, a large and diverse cross-sectional survey of adult factory-employed urban migrants in Bangalore, Hyderabad, Lucknow, and Nagpur and their rural siblings from 2005 to 2007	Nil apparent: The authors declared that they had no competing interests. The authors were all affiliated with universities or other institutes. Funding was from the Wellcome Trust and the Leverhulme Centre for Integrative Research on Agriculture and Health. It was stated that “the funder of the study had no role in study design, data collection, data analysis, and data interpretation, or writing of the report”
Perignon et al. 2016 (18)	Adult population, France	To achieve diets associated with stepwise GHGE reductions, with minimized departure from observed diet and 3 scenarios of nutritional constraints: none, on 7 macronutrients and for all 33 micro- and macronutrient recommendations	Dietary intakes were derived from the 7-d food records of a nationally representative stratified random sample of French adults (Second French Individual and National Study on Food Consumption cross-sectional dietary survey). The French Information Center on Food Quality database associated with the survey provided nutrient composition data	Potential: The authors reported: “Conflict of interest: None.” Also that: “The funders had no role in the design, analysis or writing of this article.” Funding involved several parties: a national research agency, a graduate school, and the Daniel & Nina Carasso Foundation (for the lead author Perignon and Masset). As detailed for Barré et al. above in this table, Daniel Carasso was a food industry business leader. Also 2 authors (Vieux and Maillot) were with a “consulting firm,” MS-Nutrition with Vieux coauthoring publications with meat industry funding (see Barré et al. above in this table). One of the coauthors (Masset) was described as an employee with “Nestlé Research Centre” in an earlier (2015) publication (25). Another coauthor (Darmon) has also been involved as a coauthor in studies with meat industry funding (as detailed for Barré et al. above in this table).
Song et al. 2017 (19)	Men and women aged 18–50 y, China	To minimize the carbon footprint by changing the amounts of 28 food groups combined with meeting the nutritional requirements (recommended nutrient intakes) (with constraints to avoid radical dietary deviation from current patterns)	Approximately 1.31 million records of food consumption from the China Health and Nutrition Survey. Food composition data (n = 1950 foods) were from the China Food Composition tables	Nil apparent: There was no statement provided on competing interests. All authors were from universities or a nongovernmental organization (World Wide Fund for Nature). Funding was from “Fundamental Research Funds for the Central Universities of China”
Tyszler et al. 2016 (20)	Women (31–50 y), Netherlands	To find solutions for reduced environmental impact that are as close as possible to the current diet, first without any food group constraints and later by imposing constraints on meat, fish, dairy, and eggs	The diet was modeled with 207 selected products of the Dutch Consumption Panel (survey data). Food composition data were obtained from the Dutch Food Composition Table.	Potential: There was no statement provided on competing interests. All 3 authors (Tyszler, Kramer, and Blonk) were consultants (for Blonk Consultants). All these 3 authors were funded by a dairy company in the work by Kramer et al. above in this table, where it was stated that this consultancy does work for “the food and beverage industry” and the “animal feed industry”. No funding sources were stated
van Dooren et al. 2015 (21)	Male and female adults (31–50 y), Netherlands	To maximize the most consumed food products (n = 206) combined with achieving low food costs and low climate impact, yet fulfilling all nutritional requirements	Dutch National Food Consumption Survey 2007–2010. The nutritional data were from the Dutch Nutrient Database	Potential: The authors declared “no conflict of interest.” The second and third authors (Tyszler and Kramer) were from Blonk Consultants (which does work for “the food and beverage industry”, the “animal feed industry,” etc.; see Kramer et al. above in this table)
Verger et al. 2018 (22)	Adult population, Tunisia	To achieve dietary changes at different scales, in order to nutritionally optimize food group consumption-production without increasing the environmental impact	Food consumption datasets for Tunisia were used (Epidemiological Transition and Health Impact in North Africa). Also a specific survey on the women's individual food intake and nutrition was conducted in a rural area (Sidi Bouzid governorate)	Potential: There was no statement about competing interests. The authors were from universities and a national research institute. The work was supported by the Agence Nationale de la Recherche (France). One coauthor (Darmon) has coauthored studies funded by the meat industry (see Barré et al. above in this table). Another coauthor (Perignon) has received support from a foundation whose founder was a food industry leader (see Barré et al. above in this table).
Vieux et al. 2018 (23)	Adults (aged 18–64 y), 5 countries (France, UK, Italy, Finland, and Sweden)	To achieve nutritionally adequate diets (for each country by sex, 22 micro- and macronutrients), combined with minimizing the deviation from the current diets in terms of food groups (and applying stepwise 10% GHGE reductions).	The national FINDIET 2012 Survey in Finland; the Riksmaten 2010 study in Sweden; the INRAN-SCAI-2005 study in Italy; the NDNS rolling program for 2008–2012 in the UK; the INCA2 study 2006–2007 in France. Nutrient composition of the 151 food items studied was estimated for each country	Potential: The authors declared “that they have no conflict of interest.” The first author (Vieux) and third author (Gazan) were with the “consulting firm,” MS-Nutrition, and as part of this consultancy Vieux has coauthored articles with meat industry funding (see Barré et al. above in this table). Two other coauthors (Darmon and Perignon) are discussed in the row immediately above (on Verger et al). Funding was from the French Environment & Energy Management Agency

¹GHG, greenhouse gas; GHGE, greenhouse gas emission; NDNS, National Diet and Nutrition Survey (UK).

The main findings were that in all 12 studies the optimized diets were more plant-based with reductions in meat—particularly ruminant meats such as beef and lamb. This reduction was counterbalanced by an increase of fish in the diets in half (6 of 12) of the studies. There were 3 studies where the results for fish were mixed (11, 18, 13); and 1 where fish decreased (12). In some cases the increase in fish was driven by nutritional constraints for particular minimal intakes of DHA and EPA in national and international recommendations. However, these studies did not include the option of obtaining these fatty acids via supplements made from microalgae (26, 27). Also these studies acknowledged the limitations around not considering the sustainability of various capture fisheries and the environmental impacts of aquaculture [e.g., from the composition of aquaculture feeds (12)].

Dairy products also tended to decrease in most of the studies (7 of 12) with more optimized diets [exceptions were little change in 1 study (11); mixed results among the various dairy products (i.e., cheese, milk, yogurt) in 2 studies (18, 19); mixed results by sexes and country in the 5-country study (23); and an increase in the study for Tunisia (22)]. Other foods that tended to be reduced in the more sustainable and healthy diets included sweet foods (biscuits, cakes, and desserts), savory snacks, white bread, and beverages (alcoholic and soda drinks). In particular, soda drinks were a reduced part of the diet in 4 of the studies (21, 16, 13, 15) compared with increased in 1 study (18), and mixed results in 1 other (11) (out of the 6 studies for which this information was available). The high weight/volume of beverages is likely to be a factor in transport-related environmental impacts, but the type of packaging may also be important [as per alcoholic beverages (28)].

In terms of costs of the healthier and more sustainable diets, 1 Dutch study (which aimed to also reduce food prices) reported that a low-carbon diet costing €2.59/d was possible (around 40% of the average daily Dutch diet) (21). Another study that included food costs in the French diet as a constraint also found lower food costs with the more sustainable diets (11). Another French study found that imposing the nutritional constraints of the healthy diet (meeting all nutrient recommendations) slightly increased the cost of the diet, whereas reducing GHGEs by >50% decreased diet cost (18).

As noted by van Dooren et al. (21), there is wide country-specific variation in food production impacts (e.g., variation in importation levels and use of heated greenhouses) and also cultural patterns around food intakes. Nevertheless, it seems likely that these studies have some broad generalizability to most other countries, namely the benefits of shifting diets towards being more plant based if they are to be more sustainable. But for some studies, the food-specific findings [e.g., the considerations around such foods as “blood sausage” in a French study (11)] may have limited relevance to other countries.

How Does this Optimization Work Fit with Findings from Recent Reviews?

We identified 8 recent review articles that considered the sustainability of diets that were published since January 2015. These are discussed below in terms of health impacts, sustainability impacts, and food cost impacts.

Health

The review article literature appears to be consistent with the optimization literature that sustainable diets can also be healthy. For example, a systematic review reported the general pattern of diets associated with lower GHGEs being healthier (29). “All studies showed positive health effects, ranging from <1% reduction in estimated mortality rates for vegetarian diets, to 19% for vegan diets, though some of these were not statistically significant.”

Another systematic review (30) also reported on the likely disease burden reduction from sustainable diets based on 6 modeling studies. “The health benefits of sustainable diets may derive from increases in fruit and vegetable consumption and reductions in red and processed meat, as well as lower overall calorie intake for those individuals at risk of over-nutrition. However, health and environmental priorities may not always converge, for example, as sugar may have low environmental impacts per calorie relative to other foods, and some fruit or vegetables may have higher GHG emissions per calorie than dairy and non-ruminant meats.”

This latter point was also identified in a review by Perignon et al. (31) and in 2 other recent reviews (32, 33). One of these reviews (33) specifically recommended that dietary recommendations for reduced GHGEs “must also address sugar consumption and micronutrient intake.”

A well-studied specific dietary pattern is the Mediterranean diet. In a recent review (34) the authors reported that the traditional Mediterranean diet “offers considerable nutrition and health benefits, especially in the prevention of major chronic diseases. It also has lower environmental impacts (ecological, carbon and water footprints) than northern Europe and American diets.” However, this statement does not recognize the issues with fish consumption where the environmental impacts are typically not well measured.

Characteristics of Sustainable Diets

Similar to the optimization literature, the review article literature also indicates that reduced meat in diets is a key part of making them more sustainable. In particular, there appears to be a fairly consistent pattern of vegetarian or diets low in animal products being more sustainable according to a systematic review that included 210 scenarios from 63 studies (29). In this review, the 5 most sustainable dietary patterns in terms of GHGEs were: vegan (−45% reduction in emissions); then “ruminants replaced by monogastric sources [e.g., poultry, pork] + no dairy” (−33%); then vegetarian (−31%) and “meat + dairy partially replaced by plant-based food” (−31%); and then pescatarian (−27%). These lower animal product diets were also associated with reductions in land use and water use. Nevertheless, this review also reported that the environmental footprints for some plant-based foods were similar to, or higher than for, various meats (e.g., water use or GHGEs per calorie of nuts, fruits, and vegetables was higher than for several animal-based foods in 5 studies (5 of 63 or 8% of all the studies). However, such “per calorie” comparisons can be criticized as being misleading, because there is limited meaningful exchangeability between fruits/vegetables and meat on an energy basis (i.e., lacto-ovo vegetarians tend to replace meat with eggs, dairy products, legumes, and nuts, etc). Indeed, a far more meaningful comparison is with total dietary patterns that achieve overall nutrient requirements.

Another recent review by Perignon et al. (31) considered 10 studies that used data from self-selected diets and concluded that: “Reductions in meat consumption and energy intake were identified as primary factors for reducing diet-related greenhouse gas emissions. The choice of foods to replace meat, however, was crucial, with some isocaloric substitutions possibly increasing total diet greenhouse gas emissions.” Again, however, considering only isocaloric substitutions is misleading (as argued above). Nevertheless, the main findings around meat and energy intake reductions in the Perignon et al. review are compatible with an earlier (2014) study which was particularly comprehensive in that it considered 120 publications of life-cycle analyses (35). It reported that the most favorable dietary pattern for lowest GHGEs, for lowest cropland requirements (for year 2050 projections), and for lowest all-cause mortality, was vegetarian first, pescatarian second, Mediterranean diet third, and the “typical omnivorous diet” last. The authors concluded that: “Alternative diets that offer substantial health benefits could, if widely adopted, reduce global agricultural greenhouse gas emissions, reduce land clearing and resultant species extinctions, and help prevent such diet-related chronic non-communicable diseases.”

There was just 1 review identified (published in 2015) which questioned whether or not dietary change could contribute to sustainability: “To date, it is difficult to assimilate all of the disparate approaches, and more concerted efforts for multidisciplinary studies are needed” (36). However, it should be noted that the lead author of the review was supported by a “dairy research institute” and the other author was a consultant who worked for food and beverage companies. Also other researchers have published a critique of this review's emphasis on the uncertainties by stating that: “The current body of research, despite methodological differences, generally shows that reducing intake of animal-based products—particularly ruminant meat—proportionally decreases dietary GHG emissions” (37).

Costs of More Sustainable Diets

Perignon et al. (31) noted that based on 5 studies, “diets with good nutrition and low environmental impact could be achieved at no extra cost for the consumers.” Additional studies reported in another review by Jones et al. (38) also reported this pattern, i.e., no extra cost for a diet that reduced energy intake and increased consumption of plant-based foods in diets and which reduced GHGE by 20% (39). But neither of these reviews included optimization studies indicating that low-GHGE diets are associated with lower food costs (21, 18) [or 1 published in 2018 (11)]. Also the Perignon et al. review did not include a number of other relevant studies [e.g., 1 showing cost savings (40) and 3 showing increased costs with more sustainable diets (41–43)].

Furthermore, the Jones et al. review did not include a study by Springmann et al. (44) involving a broader economic perspective. This study used a region-specific global health model based on dietary- and weight-related risk factors with emissions accounting and economic valuation. It was also notable for incorporating both “cost-of-illness” and “value of statistical life” approaches. Overall, these authors estimated the economic benefits of improving diets to be US$1–31 trillion (equivalent to 0.4–13% of global gross domestic product in 2050). The benefits of reduced health costs were highest for vegan and then lacto-ovo vegetarian diets (compared with diets meeting global dietary guidelines on healthy eating). This pattern of the greatest economic benefit being from a vegan diet was the same when considering the social cost of carbon.

Possible Priorities for Future Research on Sustainable Diets

The currently available evidence supporting a shift to healthier and more sustainable diets should probably be enough for policy-makers to actively explore relevant policy measures, e.g., national dietary guidelines that include sustainability (45), healthier and more sustainable school meals (46) and public hospital meals, and food taxes/subsidies [e.g., taxes on unhealthy food (47)]. Nevertheless, additional research will help to further improve the evidence base and allow for the improved prioritization of the different policy options. Here we outline some of the possibilities for such research.

Expanding research with no links to commercial interests

This review identified some absence of reporting of whether or not there were competing interests, and in addition many studies had some potential for competing interests at various levels. This suggests the need for better such reporting in published studies and ideally for research funding to come entirely from bodies that have no commercial interests. It is also probably ideal if researchers doing such work have no track record of doing research funded by such commercial interests.

Improved sustainability metrics for food production and consumption

It seems desirable to keep expanding future optimization modeling to include multiple environmental footprints of different diets (e.g., covering carbon, nitrogen, water use, and land use). Nevertheless, as per the work by Gephart et al. (12) (Table 1), these all may have fairly similar impacts: “We find that diets for the minimized footprints tend to be similar for the four footprints, suggesting there are generally synergies, rather than trade-offs, among low footprint diets.” However, if land use is becoming critical for addressing climate change (i.e., for tree planting for carbon sequestration), then perhaps land use could be the next most critical footprint to consider after carbon. This is starting to happen, with 1 review (38) finding that 49 of the identified studies which “applied a Life Cycle Assessment approach”, 65% (32/49) included land use. However, another review (32) reported that in “the case of GHG emissions, changes in land use and soil carbon stocks were seldom considered.” Although it is true that some livestock grazing on hill country is a way of extracting food from land not suitable for cropping, such terrain may often be better utilized by being reforested (to provide for water quality, erosion control, flood protection, biodiversity benefits, timber, or carbon sequestration). Indeed, the benefit of forested hill country for flood protection may be accelerating as climate change increases the risk of flooding in some settings. Hence improved metrics around land use would help to capture these impacts.

Other reviewers (29) found that water use in food production is also not as well studied as either land use or GHGEs (at 34, 52, and 124 scenarios, respectively, from a total of 210 scenarios in 63 included studies). This review also reported only 2 studies that have estimated the reduction in nitrogen and phosphorus water contamination from sustainable eating patterns. Similarly, work on the “chemical footprint” from pesticide use (48) could be expanded (e.g., consideration of organic food production).

On the food side, there is also a need to pay further attention to the extent of food waste—which relates to both sustainability and food security (49); and which contributes to GHGEs (50). Indeed, an estimated US$2.6 trillion in food is wasted annually, roughly equivalent to the GDP of France (51). Another nutritional issue is to what extent certain nutrients can be cost-effectively obtained via dietary supplements (e.g., the DHA and EPA in commercial supplements from microalgae) as opposed to consuming fish. However, this could potentially generate a problematic shift away from diets and food groups to a reductionist focus on nutrition via supplements.

Consideration of infectious disease risks from livestock agriculture and meat

A complete societal-level analysis of food systems needs to consider the costs of livestock production and meat consumption in terms of zoonotic disease risk. Livestock production carries risks to humans from diseases such as campylobacteriosis, giardiasis, cryptosporidiosis, and infection with enterohemorrhagic Shiga toxin-producing Escherichia coli(STEC) (52). Meat consumers are also at risk of campylobacteriosis from poultry meat (53), salmonellosis from pork and beef (54), listeriosis from meat and poultry products (55), and E. coli infections from undercooked/raw meat (56).

Virtually the entire human population is also at long-term risk from pandemic influenza, which can arise from influenza viruses circulating in pigs, as per the pandemic in 2009 (57). Influenza infection in pigs may also have played a role in the emergence of the 1918 pandemic virus (58) which killed tens of millions of people [equivalent to 51–81 million deaths in today's world (59, 60)]. Poultry may also spread influenza viruses to humans; at present, almost all human infections from avian influenza viruses are transmitted from poultry (61).

Antimicrobial resistance (AMR) is also an important international problem and AMR bacteria or AMR-encoding genes may transfer from animals to humans through the environment, food chain, or by direct contact (62). For this reason the WHO has recommended the “complete restriction of use of all classes of medically important antimicrobials in food producing animals for growth promotion” (63).

Ideally, all of the above issues need some form of quantified risk assessment that allows for inclusion in optimization processes and cost-benefit analyses. If this is not feasible, however, then such information could at least be considered in multicriteria decision analysis so as to inform decision making around imposing tighter regulations on livestock production or appropriate taxes on livestock-related products.

Research on optimized diets in the context of new food policy settings

There is potential scope for future optimization work in settings where average dietary patterns have been shifted by changes in food policies. For example, food prices and consumption patterns have changed in Mexico as a result of a tax on unhealthy food (47), and in settings where soda taxes have been applied (64). Ecolabeling of food for health and sustainability (65) may also be changing consumption patterns in some settings. There is also research interest in studying the application of taxes to ruminant meats and dairy products, e.g., building on previous modeling work indicating health benefits and GHGE reduction benefits of such taxes (60, 66, 67). In one of these studies the carbon footprint from food for an average Danish household was decreased at a cost of 0.15–1.73 DKK per kg CO₂ equivalent (∼US$25–290/tonne) (60). If such taxes were applied in practice, then there would be changes to the cost of diets optimized for health and sustainability (i.e., probably making them even cheaper relative to typical diets). Such interventions could also generate substantial tax revenue (66), which could then fund other health and sustainability interventions.

Research relevant to low- and middle-income countries

In this review we identified only 3 optimization studies (for China, India, and Tunisia) outside the high-income grouping of countries. More such studies are needed, although fortunately a recent study has covered food sustainability through the use of data from 119 countries (with data from ∼38,700 commercially viable farms and with 40 foods representing ∼90% of global protein and calorie consumption) (28). This massive study found major scope for improving the sustainability of food production and affirmed the high environmental impact of all animal products relative to plant foods. Nevertheless, this work did not address the potential health aspects of the more sustainable dietary patterns.

Research on regional and global aspects of food sustainability

Some environmental impacts of agriculture are predominantly local (land use, etc.), but some impacts can cross borders into neighboring countries (e.g., air pollution from ammonia and particulates; nitrogen and enteric pathogens in waterways). Furthermore, the GHGEs from agriculture (carbon dioxide, methane, and nitrogen oxides) have global impacts. There is also the issue that food products are widely traded: imposing GHGE taxes on exported ruminant meat and dairy products will increase food costs for food-importing countries. It may also result in some GHGE “leakage” if food production is partly shifted to settings where there are fewer environmental safeguards and no GHGE charges. These regional and global issues suggest the need for general equilibrium macroeconomic modeling to determine how best to align national environmental regulations and national GHGE taxes with international efforts. Such research could inform the design of future international agreements (e.g., extensions of the Paris Climate Agreement). It could also inform how future trade liberalization might affect food prices and sustainability (68). Existing trade agreements may also need to be adjusted to reduce the policy incoherence around trade and achieve sustainable development goals (69).

Conclusions

This review identified 12 recent studies in which optimization was used to address nutrition and sustainability issues together. Most studies (10 of 12) aimed to reduce GHGEs, and half aimed to also reduce ≥1 other environmental impact, e.g., water use, fossil energy use, land use, marine eutrophication, atmospheric acidification, and nitrogen release. The main findings were that in all 12 studies the optimized diets were more plant based with reductions in meat—particularly ruminant meats such as beef and lamb (albeit with 6 of 12 of the studies showing diets with increased fish). The amount of dairy products also tended to decrease in most (7 of 12) of the studies with more optimized diets. These findings were broadly compatible with the findings of 7 out of 8 recent review articles on the sustainability of diets. The literature suggests that healthy and sustainable diets may typically be cost neutral or cost saving, but this is still not clear overall. Nevertheless, at the societal level there is some work to suggest net economic gains from more sustainable/healthy diets in terms of reduced health care costs or if the social cost of carbon is considered. There remains scope for improvement in such areas as expanding research where there are no competing interests; improving sustainability metrics for food production and consumption; consideration of infectious disease risks from livestock agriculture and meat; and researching optimized diets in settings where major policy changes have occurred (e.g., Mexico's tax on unhealthy food and soda taxes in various jurisdictions). There is also a need for more studies in low- and middle-income countries where some populations may face particular challenges around water scarcity and food/nutrition insecurity.

Notes

Published in a supplement to Advances in Nutrition. This supplement was sponsored by the Harding-Buller Foundation of Ohio. The contents are solely the responsibility of the authors and do not necessarily represent the official views of the sponsors. Publication costs for this supplement were defrayed in part by the payment of page charges. The opinions expressed in this publication are those of the authors and are not attributable to the sponsors or the publisher, Editor, or Editorial Board of Advances in Nutrition.

The authors are supported by a grant from the New Zealand Health Research Council (16/443) for the BODE³ Programme that encompasses work on modeling dietary changes. The first author had travel and accommodation paid by Loma Linda University to present on this work at the 7th International Congress on Vegetarian Nutrition (Loma Linda University, California, 26–28 February 2018).

Author disclosures: The authors report no conflicts of interest.

Abbreviations used: AMR, antimicrobial resistance; GHGE, greenhouse gas emission.