Supporting information

Analysis and valuation of the health and climate change co-

benefits of dietary change

Marco Springmanna,1, H Charles J Godfrayb, Mike Raynera, Peter Scarborougha

a
Oxford Martin Programme on the Future of Food, British Heart Foundation Centre on
Population Approaches for Non-Communicable Disease Prevention, Nuffield Department of
Population Health, University of Oxford. Old Road Campus, Headington, Oxford, OX3 7LF,
UK.

b
Oxford Martin Programme on the Future of Food, Department of Zoology, University of
Oxford. Address: South Parks Road, Oxford OX1 3PS, UK.

1
To whom correspondence should be addressed: Email: [email protected];
Tel: +44(0)7460202512.

Table of Contents

SI.1 Supplementary description of dietary scenarios..................................................................... 2

SI.2 Supplementary health methods ................................................................................................ 6
SI.3 Supplementary environmental methods................................................................................ 13
SI.4 Supplementary economic methods ........................................................................................ 15
SI.5 Supplementary food system results ....................................................................................... 18
SI.6 Supplementary health results ................................................................................................. 20
SI.7 Supplementary environmental results ................................................................................... 25
SI.8 Supplementary valuation results............................................................................................ 29
SI.9 Emissions feedback from reductions in mortality ................................................................ 31
SI.10 Supplementary discussion of environmental results ........................................................ 32
SI.11 Supplementary discussion of valuation results ................................................................. 33

1
Supplementary methods

SI.1 Supplementary description of dietary scenarios

We constructed four dietary scenarios to analyse the environmental and health benefits of
different food consumption patterns (Table S1). The REF scenario is based on the food and
agriculture projections of the Food and Agriculture Organization of the United Nations (FAO)
(26) which cover 110 world regions and 32 food commodities for the years 2005/07, 2030,
and 2050. We aggregated the FAO data to 16 commodities (Table S2) and 107 world regions
(Table S3) to match the detail of data used in the environmental and health analysis; and we
converted the FAO food availability data into estimates of food consumption by accounting
for food waste at the consumption level (food waste at other stages of the production chain is
accounted for in the FAO estimates used (26)) and food processing that separates the edible
parts from non-edible ones (27).

The HGD scenario is based on global guidelines on healthy eating issued by WHO/FAO
Expert Consultations on diet, nutrition (16) and human energy requirements (29) and on
recommendations of the World Cancer Research Fund (28). In particular, it is recommended
to consume at least five portions (400 g) of fruit and vegetables per day (16), less than 10%
(~50 g) of total energy intake from free sugars (16), less than 30% of total energy intake from
fats (16), less than 300 g of red meat a week (very little if any of which is processed) (29), and
to not exceed recommended total energy intake. Global recommendations exist, e.g., also for
other food groups, such as whole grains, pulses, and salt (16, 28), but those are either not
quantified or cannot be implemented in light of current data availability at the global level.
We also want to note that newer epidemiological evidence suggests that chronic disease
incidence is associated more with the composition of fats in the diet rather than with total
dietary fat intake per se (35). We therefore omitted the global recommendation on fat intake
and concentrated on those regarding fruit and vegetables, red meat, sugar, and total energy
intake.

The dietary recommendations have different health implications depending on the food group.
The recommendation on sugar consumption influences the composition of diets (e.g. the
proportion of staple foods), that on energy intake influences weight levels which, in turn, are
associated with chronic disease mortality (18, 19, 29), and the recommendations on fruit and
vegetable consumption and red meat consumption are direct, independent risk factors for
chronic disease mortality (16, 41–46).

The dietary recommendations were implemented as follows. Consumption in regions which

consumed less than the minimum amounts was adjusted upwards to meet the minimum
requirement, whilst consumption that exceeded the maximum recommended intake was
adjusted downward to meet that ceiling. We implemented the dietary recommendations as
constraints on per-capita consumption in terms of grams per day (g/d) (except the total energy
constraint), and calculated the calorie equivalents (kcal/d) by using conversion factors from

2
the FAO (26). Total energy intake was adjusted to recommended energy requirements (29) by
varying the region-specific consumption of staple foods (grains, roots, and pulses), which
preserved the regional character of food consumption.

The energy requirements of adults depend on physical activity levels and body weight (29).
Country-specific energy requirements for maintaining a healthy body weight (as measured by
having a body mass index (BMI) in the normal range (18.5 to 24.9 kg/m2) were calculated
based on WHO/FAO and WCRF recommendations for maintaining moderate physical
activity levels and age-specific energy requirements for the height composition of US
Americans (see, e.g., Dietary Guidelines for Americans 2010 published by the US
Department of Agriculture and the US Department of Health and Human Services) applied to
the age composition of other regions (as an upper bound). The country-specific energy
requirements (averaged across men and women) ranged from 2200 to 2300 kcal per capita per
day.

The VGT and VGN scenarios are based on vegetarian and vegan dietary patterns. According
to dietary food intake surveys in the EPIC-Oxford study (31) and in the Continuing Survey of
Food Intake by Individuals (CSFII) (30), vegetarians eat about six to seven servings of fruits
and vegetables per day (one to two more than typical meat eaters), and vegans consume about
one more serving of fruits and vegetables than vegetarians. Vegetarians have a more than 4-
times higher intake of legumes (94 g/d compared to 21 g/d), something that is in line with
dietary advice given to vegetarians; and the consumption of sugars was within the limits set
out by the global dietary recommendations considered above (see e.g. the Position of the
Academy of Nutrition and Dietetics on Vegetarian Diets)

Based on those observations, we implemented the VGT and VGN scenarios as follows. In the
VGT scenario, food consumption includes at least six portions of fruits and vegetables (6 x 80
g) per day, at least one serving (80 g) of legumes/pulses per day, no red meat or poultry (or
fish), and otherwise meets the global dietary recommendations for sugar and total energy
intake. In the VGN scenario, food consumption includes at least seven portions of fruits and
vegetables per day, at least one serving (80 g) of legumes/pulses per day, no red meat,
poultry, dairy, or eggs (or fish), and otherwise meets the global dietary recommendations for
sugar and total energy intake.

It should be noted that vegetarian and vegan diets can differ from omnivorous diets on several
other aspects. We did adjust the composition of staple foods depending on regional
preferences, but we were not able to resolve nut consumption for example, an aspect with
potential health implications (4) that was found to differ between vegetarians and non-
vegetarians (30). It has also been found that vegetarian and vegans have, on average, lower
energy intake and body weight than non-vegetarians (30, 31), an aspect which would impart
further health benefits compared to non-vegetarians (36). However, because the HGD
scenario based on global dietary guidelines already contains recommendations on energy
intake for maintaining a healthy body weight (29), we do not adopt separate values of total

3
energy intake in the VGT and VGN scenarios, but instead use the same recommendations as
the HGD scenario.

Table S1: Overview of diet scenarios

Scenarios Features Comments

Global energy intake per capita increases by 13% between

2005/07 and 2050; per-capita consumption (in grams)
increases by 15% for fruits and vegetables, 15% for sugar, Based on projections by the Food and
Reference 35% for oils, 13% for red meat, 63% for poultry, 19-21% Agriculture Organization of the United
(REF) for eggs and dairy, 14% for roots and pulses, 20% for Nations (FAO) (26), adjusted for food waste
maize, and 32% for other grains, except wheat and rice and food conversion into edible parts (27).
whose consumption stays constant (wheat) or decreases by
by 5% (rice).

Based on global dietary recommendations by

Global Min five portions per day of fruits&veg, max 300 grams
the World Cancer Research Fund (28) and
dietary per week of red meat, less than 50 grams per day of sugar,
WHO/FAO Expert Consultations on diet and
guidelines total energy intake as recommended for moderately active
nutrition (16) and human energy
(HGD) population (2200-2300 kcal per day).
requirements (29).
Min six portions per day of fruits&veg, one portion per
Vegetarian
day of legumes, no red meat or poultry (or fish), sugar and Based on observed dietary patterns (30-31)
(VGT)
total energy intake as recommended in HGD scenario.
Min seven portions per day of fruits&veg, one portion per
day of legumes, no red meat, poultry, dairy, or eggs (or
Vegan
fish), sugar and total energy intake as recommended in Based on observed dietary patterns (30-31).
(VGN)
HGD scenario.

Notes: fish consumption is omitted as a category in FAO projections, and therefore not included in the dietary scenarios;
however, kcal/d from fish consumption is included in an aggregated 'other' commodity.

Table S2: Food commodities included in the analysis

Food commodities
Vegetables Beef Wheat Roots
Fruits Pork Rice Pulses
Sugar Poultry Maize Other commodities
Oils Eggs Other grains
Dairy
Notes: The 'Beef' category includes beef, veal and buffalo meat;
and mutton, lamb and goat meat; the 'Other commodities' category
includes spices, stimulants, fish and seafood; plantains are
categorized as roots due to its nutritional similarity with roots
(WCRF/AICR, 2007).

4
Table S3: Regional aggregation
Developing countries (DPG)
Latin America and the Caribbean (LAC)
Argentina Dominican Rep. Honduras Peru
Bolivia Ecuador Jamaica Suriname
Brazil El Salvador Mexico Trinidad and Tobago
Chile Guatemala Nicaragua Uruguay
Colombia Guyana Panama Venezuela
Costa Rica Haiti Paraguay
Cuba
Sub-Saharan Africa (SSA)
Angola Dem. Rep. of the Congo Madagascar Senegal
Benin Eritrea Malawi Sierra Leone
Botswana Ethiopia Mali Somalia
Burkina Faso Gabon Mauritania Sudan
Burundi Gambia Mauritius Swaziland
Cameroon Ghana Mozambique Togo
Central African Rep. Guinea Namibia Uganda
Chad Kenya Niger United Rep. of Tanzania
Congo Liberia Nigeria Zambia
Cote d'Ivoire Rwanda Zimbabwe
Eastern Mediterranean (EMR)
Afghanistan Iraq Morocco Turkey
Algeria Jordan Saudi Arabia Yemen
Egypt Lebanon Syrian Arab Rep.
Iran, Islamic Rep. Libya Tunisia
South Asia (SAS)
Bangladesh Nepal Sri Lanka
India Pakistan
East Asia (EAS)
Cambodia Indonesia Mongolia Rep. of Korea
China Lao People's Dem. Rep. Myanmar Thailand
Indonesia Malaysia Philippines Viet Nam
Developed countries (DPD)
Australia Japan South Africa Other Western Europe
Canada New Zealand United States of America Other Eastern Europe
Israel Russian Federation European Union (EU-27) Central Asian Republics

5
SI.2 Supplementary health methods

We analysed the health impacts associated with changes in food consumption by using a
comparative risk assessment framework with three disease states and four diet and weight-
related risk factors. The disease states included coronary heart disease (CHD), stroke, type 2
diabetes (T2DM), and cancer which is an aggregate of site-specific cancers. The four specific
disease states accounted for about 60% of deaths from non-communicable diseases and for
about 40% of deaths globally in 2010 (6). The weight-related risk factors included the
prevalence of overweight (25<body mass index (BMI)<30) and obesity (BMI>30), and the
diet-related risk factors included fruit and vegetable consumption and red meat consumption
which, together, accounted for more than half of all deaths that were attributable to diet-
related risks in 2010 (4).

For the weight-related risk assessment, we used the scenario estimates of total energy intake
to estimate changes in the prevalence of overweight and obesity based on historical
relationships between those weight categories and caloric availability. For that purpose, we
paired FAO food availability data for the years 1980-2009 (smoothed by a three-year moving
average) with WHO data on the prevalence of overweight and obesity, and estimated two
linear regression models:

𝑝𝑜𝑣𝑒𝑟𝑤𝑒𝑖𝑔ℎ𝑡 (𝑟) = 0.02462 ∙ 𝑘𝑐𝑎𝑙(𝑟) − 29.67965; 𝑅 2 = 0.49; 𝑝 < 0.001

𝑝𝑜𝑏𝑒𝑠𝑖𝑡𝑦 (𝑟) = 0.01000 ∙ 𝑘𝑐𝑎𝑙(𝑟) − 14.98936; 𝑅 2 = 0.38; 𝑝 < 0.001

We calculated the prevalence for overweight and obesity in 2050 by first calculating
percentage changes between the prevalence of overweight and obesity in 2005/07 and 2050,
as estimated via the regression equation and then applying those percentage changes to FAO
baseline values for 2005/07.

We estimated the mortality and disease burden attributable to dietary and weight-related risk
factors by calculating population attributable fractions (PAFs). PAFs represent the proportions
of disease cases that would be avoided when the risk exposure was changed from a baseline
situation to a counterfactual situation. For calculating PAFs, we used the general formula (4,
17):

∫ 𝑅𝑅(𝑥)𝑃(𝑥)𝑑𝑥 − ∫ 𝑅𝑅(𝑥)𝑃′ (𝑥)𝑑𝑥 (1)

𝑃𝐴𝐹 =
∫ 𝑅𝑅(𝑥)𝑃(𝑥)𝑑𝑥

where 𝑅𝑅(𝑥) is the relative risk of disease for risk factor level 𝑥, 𝑃(𝑥) is the number of
people in the population with risk factor level 𝑥 in the baseline scenario, and 𝑃′ (𝑥) is the
number of people in the population with risk factor level 𝑥 in the counterfactual scenario. We
assumed that changes in relative risks follow a dose-response relationship (4), and that PAFs
combine multiplicatively (4, 40), i.e. 𝑃𝐴𝐹𝑇𝑂𝑇 = 1 − ∏𝑖(1 − 𝑃𝐴𝐹𝑖 ) where the i’s denote
independent risk factors.

6
In addition to changes in mortality, we also calculated the years of life lost (YLS) due to a
change in dietary and weight-related risk factors. For that purpose, we multiplied each age-
specific death by the life expectancy expected at that age using the Global Burden of Disease
standard abridged life table (40).

We used publically available data sources to parameterize the comparative risk analysis.
Population and mortality data by region and 5-year age group for the year 2050 were adopted
from IIASA and the United Nations Population Division, respectively. All-cause mortality
rates for 2050 were decomposed into cause-specific ones for CHD, stroke, T2DM, an
aggregate of cancers, and an aggregate of all other causes by using burden of disease
estimates for WHO member states in 2008, projected forward to 2050 for the dietary and
weight-related risk factors focussed on here. Given that dietary and weight-related risk factors
are predominantly associated with chronic disease mortality, we focused on the health
implications of changes in those risk factors for adults (aged 20 and older).

We restricted the selection of relative risk parameters to meta-analyses and pooled

prospective cohort studies. The diet and weight-related relative risk parameters were adopted
from pooled analyses of prospective cohort studies (18, 19), and from meta-analysis of
prospective cohort and case-control studies (41–44, 28, 45, 46). The cancer associations have
been judged as probable or convincing by the World Cancer Research Fund, and in each case
a dose-response relationship was apparent and consistent evidence suggests plausible
mechanisms (28). The weight-related relative risk parameters were aggregated to the BMI
categories used in this study and normalized to a risk-neutral normal weight category
consistent with the epidemiological evidence (18, 19). Table S4 lists the relative risk
parameters adopted in this study. The selection procedure is detailed below.

We accounted for the uncertainty related to the relative risk parameters in our mortality
estimates. We approximated the error distribution of relative risks by a normal distribution,
which is justified for the magnitude of errors dealt with here (<50%) (see, e.g., IPCC
Uncertainty Guidelines). We then used standard methods of error propagation to calculate the
uncertainty intervals associated with diet and weight-related changes in mortality.

Table S4: Relative risk parameters

Relative risk per cause of death
Risk factor
CHD Stroke Cancer T2DM
Fruit and vegetable consumption 0.96 (0.93-0.99) 0.95 (0.92-0.97) 0.93 (0.84-0.99)* 0.96 (0.92-0.99)
Red meat consumption 1.25 (1.21-1.29) 1.10 (1.05-1.15) 1.01 (1.00-1.05)* 1.15 (1.07-1.24)
Overweight 1.31 (1.24-1.39) 1.07 (0.73-1.59) 1.10 (1.04-1.17) 1.54 (1.42-1.68)
Obese 1.78 (1.64-1.92) 1.55 (1.14-2.11) 1.40 (1.30-1.50) 7.37 (5.16-10.47)
* global average, actual relative risk is region-specific.
Sources: Dauchet et al (2005, 2006), Micha et al (2010), Chen et al (2013), WCRI/AIC (2007), Li et
al (2014), Feskens et al (2013), Prospective Studies Collaboration (2009), Berrington de Gonzales et
al (2010).

7
Relative risk parameters

Weight-related risk parameters

We inferred the parameters describing relative mortality risk due to weight categories from
two large, pooled analyses of prospective cohort studies (18, 19). We concentrated on four
causes of death: ischaemic/coronary heart disease (CHD), stroke, cancers, and type-2 diabetes
mellitus (T2DM). We adopted the relative risks for CHD, stroke, and T2DM from the
Prospective Studies Collaboration (19), which analysed the association between BMI and
mortality among 900,000 persons in 57 prospective studies that were primarily designed to
evaluate risk factors for cardiovascular disease; and we adopted the relative risks for cancer
from Berrington de Gonzalez and colleagues (18) who examined the relationship between
BMI and mortality in a pooled analysis of 19 prospective studies which included 1.46 million
adults and which were predominantly designed to study cancer.

From each study, we adopted the relative risk rates for lifelong non-smokers to minimize
confounding and reverse causality, and, to increase comparability, we normalized the relative-
risk schedule to the lowest risk which, in each case corresponded to a body-mass index (BMI)
of 22.5-25. We then used the number of cause-specific deaths to aggregate the BMI intervals
of 2.5 that have been used in the studies to the WHO classification of BMI ranges.

Dietary risk parameters

Dietary risks have been the leading risk factors for death globally in 2010 (4). The Global
Burden of Disease Study included 14 different components as dietary risks, such as not eating
enough fruit, nuts and seeds, vegetables, whole grains, and omega-3s and eating too much salt
and processed meat (4). In this study, we focused on changes in the consumption of fruits,
vegetables, and meat. Those categories constituted about two thirds of the total dietary risk in
2010 (excluding potential double counting, e.g. of fibre found in vegetables and sodium found
in processed meat) (4), and country-level trends and data are available for most countries
worldwide. We adopted relative-risk parameters for developing specific diseases from recent
meta-analyses of existing studies.

Meat consumption and cardiovascular disease

The relative risks of coronary heart diseases due to meat consumption were adopted from
Micha and colleagues (41). Their comprehensive systematic review and meta-analysis of the
relationship between meat consumption (processed, red, and total meat) and cardiovascular
diseases (coronary heart disease (CHD), diabetes mellitus, and stroke) included 20 studies (17
prospective cohorts and 3 case-control studies) with 1,218,380 individuals from 10 countries.
However, analyses of specific subcategories, e.g. total meat consumption and stroke, included
significantly less studies. The results show positive associations between consumption of
processed and total meat and the incidence of CHD, diabetes mellitus, and stroke. Since the
publication of Micha et al (41), updated reviews of the association between meat consumption

8
and stroke have become available. We therefore only adopted the estimates for the association
between meat consumption and coronary heart disease from the Micha et al study. We
adopted the findings for total meat indicating that consumption of 100 g per day increases
CHD risk by 25% (RR=1.25; 95% CI, 1.21 to 1.29). The estimate is based on data from 4
prospective cohort studies; one extremely positive finding from a case-control study was
excluded in the estimate (41).

The relative risk of stroke due to meat consumption was adopted from Chen et al (42) which,
for stroke, provided an updated meta-analysis of Micha et al (41) containing five large
independent cohort studies (compared to two in Micha et al (41)). Chen et al (42) found that
consumption of red and/or processed meat increases the risk of stroke, in particular, ischemic
stroke. Their dose–response analysis of the primary studies showed that the risk of stroke
increased significantly by 10% for each 100 g per day increment in total meat consumption
(RR=1.10; 95% CI, 1.05–1.15), by 13% for each 100 g per day increment in red meat
consumption (RR=1.13; 95% CI, 1.03–1.23) and by 11% for each 50 g per day increment in
processed meat consumption (RR=1.11; 95% CI, 1.02–1.20), with low study heterogeneity.
We adopted the estimate for total meat consumption.

Meat consumption and diabetes

The relative risk of T2DM due to meat consumption was adopted Feskens et al (46) who
updates the meta-analysis of Micha et al (41) for T2DM. For total meat consumption, Feskens
et al (46) included findings from 14 separate cohorts result, resulting in a pooled RR of 1.15
per 100 g (RR=1.15; 95% CI, 1.07-1.24), indicating that for each 100 g of total meat
consumed, the risk of T2DM increases by 15 %. For red meat, the overall RR based on 14
individual studies was 1.13 per 100 g (95% CI 1.03–1.23), and for processed meat, the
summary estimate of 21 separate cohorts was 1.32 per 50 g (95 % CI 1.19–1.48). We adopt
the total meat estimate which includes red and processed red meat.

Meat consumption and cancer

The association between meat consumption and cancer was reviewed in the Second Expert
Report "Food, Nutrition, Physical Activity, and the Prevention of Cancer: a Global
Perspective" published in 2007 by the World Cancer Research Fund together with the
American Institute for Cancer Research (28). The report is based on reviews and meta-
analysis of over 7,000 scientific studies published on cancer prevention. It was the outcome of
a 5-year project which involved a panel of 21 leading scientists and 9 research centres around
the world.

With respect to meat, the report concluded that (28): red and processed meats are convincing
causes of colorectal cancer; there is substantial amount of evidence, with a dose-response
relationship apparent from case-control studies (red meat) and cohort studies (processed
meat); there is evidence (red meat) and strong evidence (processed meat) for plausible
mechanisms operating in humans. The report also noted that there is limited evidence

9
suggesting that red meat is a cause of cancers of the oesophagus, lung, pancreas and
endometrium; and that processed meat is a cause of cancers of the oesophagus, lung, stomach
and prostate.

We followed the conclusions of the expert report and its updates that highlighted a convincing
causal link between meat consumption and colorectal cancer and adopted the following
estimate: consumption of 100 g/day of red and processed meat increases the risk of colorectal
cancer by 16% (RR=1.16, CI=1.04-1.30). We aggregated the estimate to region-specific
relative risks for all cancers by weighing it by the ratio of regional deaths due to colorectal
cancer to all cancer deaths in that region. Globally, this resulted in a relative risk of cancer of
RR=1.01.

Fruit and vegetable consumption and cardiovascular disease

The relative risks of stroke and CHD due to fruit and vegetable consumption were adopted
from Dauchet and colleagues (43, 44). Dauchet et al (44) conducted a meta-analysis for CHD
and its association with fruit and vegetable consumption. The analysis included nine
prospective cohort studies that consisted of 91,379 men, 129,701 women, and 5,007 CHD
events. Pooled relative risks showed that CHD was decreased by 4% [RR (95% CI): 0.96
(0.93–0.99)] for each additional portion of 106 g per day of fruit and vegetable intake and by
7% [0.93 (0.89–0.96)] for fruit intake. We adopted the estimate for the aggregate of fruit and
vegetable consumption, i.e. RR=0.96 (0.93-0.99).

Dauchet et al (43) undertook a similar meta-analysis for stroke. The analysis included seven
cohort studies with 90,513 men, 141,536 women, and 2,955 strokes. Pooled relative risks
showed that the risk of stroke was decreased by 11% (RR 95% CI: 0.89 [0.85 to 0.93]) for
each additional portion of 106 g per day of fruit, by 5% (RR: 0.95 [0.92 to 0.97]) for fruit and
vegetables, and by 3% (RR: 0.97 [0.92 to 1.02]) for vegetables. The study found that the
association between fruit or fruit and vegetables and stroke was linear, suggesting a dose-
response relationship. We adopted the estimate for the aggregate of fruit and vegetable
consumption, i.e. RR=0.95 (0.92-0.97).

Fruit and vegetable consumption and diabetes

The relative risk of diabetes due to fruit and vegetable consumption was adopted from Li et al
(45). In their meta-analysis, Li et al (45) included 10 prospective cohort studies. Eleven
comparisons from nine studies reported an association between fruit intake and risk of T2DM,
with 22 995 T2DM outcomes and 424 677 participants. Overall, fruit intake was inversely
associated with risk (relative risk 0.93, 95% CI 0.88 to 0.99). Dose–response analysis
indicated that a 1 serving/day (106 g/d) increment of fruit intake was associated with a 6%
lower risk of T2DM (relative risk 0.94, 95% CI 0.89 to 1.00). Seven comparisons from six
studies reported an association between green leafy vegetables (GLV) intake and risk of
T2DM, with 19 139 T2DM outcomes and 251 235 participants. Overall, GLV intake was
inversely associated with risk (relative risk 0.87, 95% CI 0.81 to 0.93). Dose–response

10
analysis indicated that a 0.2 serving/day (0.2 x 106 g/d) increment of GLV intake was
associated with a 13% lower risk of T2D (relative risk 0.87, 95% CI 0.76 to 0.99). We
weighted the GLV-related RR per 0.2 servings with a risk neutral RR per 0.8 servings of other
vegetables, and with the fruit-related RR. This yielded a RR of 0.96 (0.92-0.99).

Fruit and vegetable consumption and cancer

The relative risks of various cancers due to fruit and vegetable consumption were adopted
from the 2007 expert report of the World Cancer Research Fund and the American Institute
for Cancer Research (28). The export report concluded that:

“[N]on-starchy vegetables and also fruits probably protect against cancers of the mouth,
larynx, pharynx, oesophagus, and stomach, and that fruits also probably protect against lung
cancer. The case that vegetables, fruits, and pulses (legumes) may be protective against
cancers of some sites is supported by evidence on foods containing micronutrients found in
these and other plant foods. Foods containing carotenoids probably protect against cancers of
the mouth, pharynx, larynx, and lung; foods containing beta-carotene and also vitamin C
probably protect against oesophageal cancer; foods containing selenium and also lycopene
probably protect against prostate cancer; and foods containing folate probably protect against
pancreatic cancer. [..] [It was also found that] foods containing dietary fibre, found in plant
foods (particularly when in whole or relatively unprocessed forms), probably protect against
colorectal cancer.”

We adopted the following relative risk parameters for which the expert report indicated a
substantial amount of consistent evidence for plausible mechanisms and a dose-response
relationship:
Cancers of the mouth, pharynx, and larynx:
- Consumption of non-starchy vegetables reduces the risk of cancers of the mouth,
pharynx, and larynx by 28% per 50-g serving per day (RR=0.72; 95% CI: 0.63-0.82).
- Consumption of fruits reduces the risk of cancers of the mouth, pharynx, and larynx
by 28% per 100-g serving per day (95% CI: 0.59-0.87).
- We adopted the simple average of vegetable and fruit consumption for the relative risk
of cancers of the mouth, pharynx, and larynx, i.e. RR=0.72 (0.61-0.85).
Oesophageal cancer:
- Consumption of raw vegetables reduces the risk of oesophageal cancer by 31% per 50-
g serving per day (RR=0.69; 95% CI: 0.58-0.83).
- Consumption of fruits reduces the risk of oesophageal cancer by 44% per 100-g
serving per day (RR=0.56; 95% CI: 0.42-0.74).
- We adopted the simple average of vegetable and fruit consumption for the relative risk
of oesophageal cancer, i.e. RR=0.63 (0.50-0.79).
Stomach cancer:
- Consumption of non-starchy vegetables reduces the risk of stomach cancer by 30%
per 100-g serving per day (RR=0.70; 95% CI: 0.62-0.79).

11
 [Estimates of green-yellow vegetables yield RR=0.59 (0.46-0.75) per 100 g/d; green,
leafy vegetables yield RR=0.43 (0.24-0.77) per 50 g/d; and raw vegetables yield
RR=0.5 (0.38-0.65) per 100 g/d]
- Consumption of fruits reduces the risk of stomach cancer by 33% per 100-g serving
per day (RR=0.67; 95% CI: 0.59-0.76).
- We adopted the simple average of vegetable and fruit consumption for the relative risk
of stomach cancer, i.e. RR=0.69.
Lung cancer:
- Consumption of fruits reduces the risk of lung cancer by 6% per 80-g serving per day
(RR=0.94; 95% CI: 0.90-0.97).
- We adopted the simple average of vegetable and fruit consumption for the relative risk
of lung cancer (assuming no effect of vegetable consumption, RR=1), i.e. RR=0.97.
Overall relative risk:
- We aggregated the cause-specific relative-risk estimates to region-specific all-cancer
estimates by weighing each risk by the ratio of regional deaths due to the specific
cancer divided by all cancer deaths in that region. Globally, this yielded an aggregate
all-cancer risk of RR=0.93.

12
SI.3 Supplementary environmental methods

We calculated the environmental impacts associated with the different dietary scenarios by
using commodity-specific GHG emissions factors. The emissions factors are adopted from a
recent meta-analysis of life-cycle analyses (LCAs) which estimated the ‘cradle to farm gate’
emissions of different food items (12) and adjusted to account for productivity improvements
that would reduce GHG intensity over time based on output (3).

The meta-analysis included 120 publications with 555 LCAs of GHG emissions from 82
different food items (12). The system boundaries of those LCAs included emissions from pre-
farm activities, such as fertilizer and feed production, as well as infrastructure construction,
but excluded emissions from land-use change and post-farm-gate activities, such as
processing, packaging, and transportation to the household. We adopted the emissions factors
for 15 food commodities corresponding to the commodity detail chosen in this study (Table
S5). We did not account for GHG emissions related to the consumption of fish and seafood,
because those food groups are not resolved in the projections of food demand used in this
study (26).

To obtain estimates of the total GHG emissions associated with each diet scenario, we
multiplied commodity-specific gross consumption estimates (i.e. without deducting food
waste) (measured in kcal/d/cap) by the corresponding emissions factors (measured in g/kcal)
and population numbers, and converted the result to GHG emissions per region per year. We
used the standard deviation reported in the meta-analysis (12) to calculate error bounds for
our estimates.

Significant improvements in the GHG intensity of cereals and livestock products occurred
over the period 1961-2010 due to a move from less to more intensive production systems (3).
To account for future productivity improvements, we used the historical changes in GHG
emissions intensity per change in output between 1961-2010 for beef, pork, poultry, eggs,
dairy, and rice (3) (excluding small absolute increases in the emissions intensity for wheat and
maize) and applied those changes to the changes in production between the base year of
2005/07 and 2050. The resulting reductions in GHG emissions intensity were 14.9% for beef,
5.8% for pork, 9.8% for poultry, 10.0% for eggs, 22.5% for dairy, and 6.2% for rice (Table
S6). It is possible that part of those efficiency improvements could be offset by increased
emissions from land-use change that may be associated with increased production. However,
our system boundary, given by the baseline emissions intensities (12), does not include
emissions from land-use change. We therefore did not attempt to quantify such offsetting
effects, but we discuss potential emissions feedbacks in the discussion section of the main
text.

For contextualization in terms of climate change, we calculated the ratio of food-related GHG
emissions to GHG emissions from all sources. We used historical data for the base year
2005/07, and we adopted emissions estimates from the International Panel on Climate Change
(IPCC) for an emissions pathway consistent with a likely (>66%) chance of limiting global

13
temperature increase to below 2°C (32), an aim that was agreed to by the international
community in 2010 as part of the Cancun Agreements to the United Nations Framework
Convention on Climate Change.

Table S5: GHG emissions factors by commodity (adopted from Tilman and Clark, 2014).
Carbon-eq emissions (mean and standard deviation (std))
Food item
g/kcal g/kcal_std g/serving g/serving_std
Sugar 0.02 0.00004 0.9 0.002
Roots 0.03 0.02 0.84 0.04
Pulses 0.02 0.002 1.9 0.22
Maize 0.03 0.004 3 0.36
Wheat 0.06 0.009 5.2 0.86
Other grains 0.05 0.005 5.4 0.46
Fruits* 0.12 0.03 7.75 1.85
Rice 0.14 0.02 14 2.1
Vegetables 0.68 0.25 14 3.5
Oils 0.16 0.04 20 5.4
Eggs 0.59 0.03 24 1
Poultry 1.3 0.05 52 2.1
Pork 1.6 0.1 61 3.6
Dairy 0.52 0.04 74 2.5
Beef 5.6 0.41 330 18
Other commodities** 1.2 0.325 27 7.25
* The GHG coefficient for fruits is taken as average between temperate and tropical
fruits (0.10 g/kcal and 0.14 g/kcal, respectively).

** includes stimulants, spices, and seafood; we don't account for changes in that
category, and therefore also omit this item in the main scenarios' GHG accounting.

Table S6: Improvements in GHG emissions intensity by commodity. Future changes in

emissions intensity were calculated based on historical changes in emissions intensity per
change in output. Past changes in emissions intensity and production were adopted from
Tubiello et al. (2014).
Change in production (%) Change in emissions intensity (%)
Food item
1961-2010 2005/07-2050 1961-2010 2005/07-2050
Beef 132 73 -27 -15
Pork 335 43 -45 -6
Poultry 1,011 123 -80 -10
Eggs 359 63 -57 -10
Dairy 110 65 -38 -22
Rice 230 29 -49 -6

14
SI.4 Supplementary economic methods

We estimated the economic benefits of changes in diets by monetizing the changes in GHG
emissions and health. For monetizing the GHG emissions, we used estimates of the social cost
of carbon (SCC) which represents the monetized damages associated with an incremental
increase in carbon emissions. The values adopted are based on a comprehensive integrated-
assessment modelling exercise facilitated by technical experts from several US agencies (20).
The interagency group selected four SCC values for use in regulatory analyses. Three values
are based on the average SCC across three integrated-assessment models and socio-economic
and emissions scenarios at discount rates of 2.5, 3, and 5 percent. The fourth value is the SCC
value for the 95th percentile at a 3 percent discount rate, which is meant to represent the
higher-than-expected economic impacts from climate change further out in the tails of the
SCC distribution. For the year 2050, the SCC estimates are 27, 71, 98, and 221 USD/tCO2 for
discount rates of 5, 3, 2.5 percent, and the 95th percentile at a 3 percent discount rate.

For monetizing the health impacts, we relied on two complementary costing methods, the
value of statistical life approach and the cost-of-illness approach. The value of statistical life
(VSL) is a measure for the willingness to pay for a mortality risk reduction defined as the
marginal rate of substitution between money and mortality risk in a defined time period (22).
The VSL does not represent the value of life itself, but rather the value of small risks to life
which can be estimated either from market decisions that reveal the implicit values reflected
in behaviour (revealed preference studies), or by using surveys which elicit respondents’
willingness to pay for small reductions in mortality risks directly (stated preference studies).

We based our valuation on a comprehensive global meta-analysis of stated preference surveys

of mortality risk valuation undertaken for the Organisation for Economic Co-operation and
Development (OECD) (21). Following OECD recommendations, we adopted a VSL base
value for the EU-27 of USD 3.5 million (1.75-5.25 million) and used the benefit-transfer
method to calculate VSLs in other regions (22). In the benefit-transfer method, the VSL base
value is adjusted by income (Y) subject to an elasticity of substitution (β):
𝑌𝑟 𝛽
𝑉𝑆𝐿𝑟 = 𝑉𝑆𝐿𝑏𝑎𝑠𝑒 ( )
𝑌𝑏𝑎𝑠𝑒

Following OECD recommendations, we used GDP per capita adjusted for purchasing power
parity (PPP) as a proxy for income, and we adopted an elasticity of 0.8 for benefit transfers to
high-income countries and an elasticity of 1.0 for benefit transfers to low and middle-income
countries (22). Baseline data on GDP per capita were sourced from the World Bank
Development Indicator database, and GDP projections for 2050 were based on projections by
the OECD for a “Middle of the Road” socio-economic development pathway (SSP Database,
available at: https://tntcat.iiasa.ac.at/SspDb). In line with World Bank methodology, we
adjusted the income classification of countries depending on their GDP per capita (adjusted
for purchasing power parity) in each period.

15
The error in applying the benefit-transfer method is expressed by the low and high range of
the VSL base value. We estimated the uncertainty in the monetized health impacts by using
standard methods of error propagation.

The current OECD guideline (22) recommends against differentiating the VSL for different
age groups. Our main valuation therefore consists of applying the region-specific VSL
estimates directly to changes in mortality. However, since mostly older people are affected by
changes in diets through changes in chronic-disease mortality, we also monetized the health
impact in terms of years of life lost by using the value of statistical life year (VSLY). We
calculated the VSLY for each region by expressing the VSL as the discounted net present
value of the VSLY throughout a lifetime:

𝑉𝑆𝐿𝑌𝑟
𝑉𝑆𝐿𝑟 = ∑ 𝑎𝑔𝑒
𝑎𝑔𝑒 (1 + 𝑑𝑟)

where we used a discount rate (dr) of 3% and a maximum age of 86 adopted from the Global
Burden of Disease standard lifetable. We used non-linear programming (GAMS, NLP solver)
to numerically solve the equation above for the VSLYs per region.

Our second costing method relied on the cost-of-illness approach which captures the direct
and indirect costs associated with treating a specific disease, including medical and health-
care costs (direct), and costs of informal care and from lost working days (indirect) (see e.g.
ref (23)). Because global cost-of-illness studies of cardiovascular disease and cancer do not
exist at present, we again used a cost transfer method, which is similar to the benefit transfer
method used in the value of statistical life approach and which has been used in other global
assessments of the global economic burden of cardiovascular disease and cancer (34).

We based our cost-of-illness estimation on a comparative assessment of the economic burden

of cardiovascular diseases (23, 24) and cancer (25) across the European Union. We adopted
the total cost estimate associated with CHD, stroke, and cancer for the EU in 2009, which
included direct costs (healthcare expenditure, health service utilization, expenditure on
medication) and indirect costs (opportunity costs of informal care, productivity costs due to
mortality and morbidity). We calculated the costs per death due to CHD, stroke, and cancer
using EU-wide death-by-disease statistics for the same year (24). The costs per death by
disease (d=CHD, stroke, cancer) in the EU and other regions (r) in the year 2050 were then
estimated by scaling the EU base values for direct and indirect costs by the ratio of health
expenditure per capita for direct costs, and by the ratio of GDP per capita (adjusted for
purchasing power parity) for indirect costs:
𝑑𝑖𝑟𝑒𝑐𝑡 𝑑𝑖𝑟𝑒𝑐𝑡
ℎ𝑒𝑥𝑝𝑟
𝐶𝑜𝐼𝑟,𝑑 = 𝐶𝑜𝐼𝐸𝑈,𝑑 ( )
ℎ𝑒𝑥𝑝𝐸𝑈
𝑖𝑛𝑑𝑖𝑟𝑒𝑐𝑡 𝑖𝑛𝑑𝑖𝑟𝑒𝑐𝑡
𝐺𝐷𝑃𝑟
𝐶𝑜𝐼𝑟,𝑑 = 𝐶𝑜𝐼𝐸𝑈,𝑑 ( )
𝐺𝐷𝑃𝐸𝑈

16
Productivity losses due to morbidity and mortality, which are a part of the indirect costs, were
only included for deaths occurring among those of working age (< 65 years old).

Baseline data on GDP per capita and health expenditure per capita were sourced from the
World Bank Development Indicator database, and GDP and population projections for 2050
were based on projections by the OECD and IIASA for a “Middle of the Road” socio-
economic development pathway (SSP Database, available at: https://tntcat.iiasa.ac.at/SspDb).

Projections for health expenditure per capita in 2050 are based on own projections. For those,
we linearly regressed past health expenditure per capita on past GDP per capita for the period
1995-2013, and then used the relationships to project future health expenditure per capita
based on future GDP trajectories. Most regions exhibited a good fit to the linear regression
model (p-values smaller than 0.01 (99% significance level) for n=141 out of 174 regions;
96% with p-values smaller than 0.001). For regions that did not exhibit a good statistical fit
(p-values larger than 0.01; n=33 out of 174 regions), we used WHO estimates of health
expenditure as percentage of GDP in 2010, and calculated future health expenditure by using
future GDP values, holding the percentage of health expenditure to GDP constant. (This is
likely to have yielded a conservative estimate as global health expenditure as a percentage of
GDP increased by approximately 7% in each of the last three five-year periods.)

We added a transfer error (uncertainty) of 30% to the cost-of-illness estimates based on sub-
sample comparison. A comparison of the costs per disease death for individual EU countries
in 2009 between estimates based on the cost transfer method and the original estimates
indicated a population-weighted deviation of 8-14% in total costs, and of 8-31% in direct
costs.

For diabetes, globally comparable health expenditure estimates were available (Zhang et al.,
2010) which we adopted directly and adjusted for potential double-counting of cost
components by using incremental cost estimates (Köster et al., 2011; ADA, 2013). Diabetes is
associated with a high risk of developing complications and co-morbidities, such as CHD,
stroke, blindness, renal failure, and amputation, which makes cost coding difficult and
diabetes-related healthcare spending difficult to determine precisely. We adopted estimates of
diabetes-related deaths and diabetes-related healthcare expenditure by country from the
Diabetes Atlas (6th Edition), and used those to calculate diabetes-related healthcare
expenditure per death by region. To avoid double counting of cost components, we adjusted
the estimates of diabetes-related health expenditure produced for the Diabetes Atlas by
estimates of the incremental cost components that are specifically attributable to diabetes. We
adopted an incremental-cost ratio of 50% which is the average of available incremental cost
estimates (46-57%) (Köster et al., 2011; ADA, 2013). Finally, we projected the healthcare
expenditure attributable to diabetes death forward to 2050 by multiplying the 2013 values by
the region-specific increase in healthcare spending per capita between 2013 and 2050. Based
on earlier version of the Diabetes Atlas (Zhang et al., 2010), we assumed an uncertainty
interval of 50% around the mean estimates.

17
Supplementary results

SI.5 Supplementary food system results

According to our analysis, the adherence to global dietary guidelines is weak in many regions
and projected to remain weak in the future (Figure S1). In the base year of 2005/07, the
majority of regions did not meet dietary recommendations for the consumption of fruits and
vegetables (90 out of 107 regions covered in the analysis) and sugar (65), and more than a
third of all regions did not meet dietary recommendations for red meat (41), and exceeded
recommendations for total energy intake (47). In the reference scenario in 2050, more regions
were projected to meet recommended minimum intakes of fruits and vegetables (with 74
regions not meeting recommendation), but less regions were projected to meet the
recommended maximum intakes of red meat (65), sugar (74), and total energy intake (91). As
a consequence, large changes in the food system would be necessary to achieve the dietary
patterns considered here (Table S6).

Figure S1: Number of regions (out of a total of 107) that do not meet global dietary guidelines
in the years 2005/07 and 2050. The dietary recommendations include min 5 portions of fruits
and vegetables per day (16), less than 50 g of sugar per day (16), max 300 g of red meat per
week (28), and 2200-2300 kcal per day depending on age and sex composition of the
population (29).

18
Table S7: Food consumption in the REF scenario (lower columns) by region, and in the
alternative diet scenario in aggregate (upper columns). Food consumption by food groups is
given in grams per day, except for total energy intake which is given in kilo-calories per day.
Red shading indicates that dietary recommendations (listed under HGD) are not met in
aggregate.
Total Red Fruit &
Region Sugar Oils Poultry Eggs Dairy Pulses Staples
kcal Meat Veg
2005/07(WLD) 2,318 66 341 59 32 31 22 215 17 453
HGD(WLD) 2,243 33 492 45 43 50 26 260 16 391
VGT(WLD) 2,243 0 545 45 43 0 26 260 80 352
VGN(WLD) 2,243 0 606 45 43 0 0 0 80 436
WLD 2,624 75 393 68 43 50 26 260 19 481
DPD 2,737 135 465 87 54 89 35 549 8 381
DPG 2,603 64 380 64 41 43 24 205 21 500
HIC 2,773 145 482 85 59 95 35 570 9 354
MIC 2,625 104 412 91 39 68 35 483 6 464
LAC 2,782 116 355 110 44 100 33 366 28 417
SSA 2,387 31 170 41 37 12 7 101 37 719
EMR 2,599 39 520 79 39 58 20 257 17 461
SAS 2,570 18 245 76 45 30 15 316 25 441
EAS 2,744 126 638 50 40 54 45 103 4 432

19
SI.6 Supplementary health results

Figure S2: Total number of avoided deaths (left axis) and years of life saved (right axis) in
2050 compared to 2005/07 by risk factor (Δweight: change in prevalence of overweight and
obesity combined, ΔC(fruit&veg): change in fruit and vegetable consumption; ΔC(red meat):
change in red meat consumption; Total: all risk factors combined; Total YLS: all risk factors
combined for the measure of years of life saved).

Figure S3: Total number of avoided deaths (left axis) and years of life saved (right axis) in
2050 compared to 2005/07 by cause of death: coronary heart disease (CHD), stroke, cancer,
type 2 diabetes (T2DM), and all causes combined (“Total” for avoided deaths, and “Total
YLS” for years of life saved).

20
Table S8: Total deaths in 2050 among population aged 20 and older by cause of death
(coronary heart disease (CHD), stroke, cancer, type 2 diabetes (T2DM), other causes of deaths
(other), and all causes combined) and region (developed countries (DPD), developing
countries (DPG); Western high-income countries (HIC), Western middle-income countries
(MIC), Latin America and the Caribbean (LAC), Sub-Saharan Africa (SSA), Eastern
Mediterranean (EMR), South Asia (SAS), East Asia (EAS)).
Region CHD Stroke Cancer T2DM Other All causes
WLD 14,062 12,904 12,846 2,615 43,024 85,451
DPD 3,648 2,051 3,984 409 7,292 17,384
DPG 10,415 10,853 8,863 2,205 35,732 68,068
LAC 907 629 1,108 443 3,096 6,183
SSA 771 839 583 322 5,791 8,306
EMR 1,261 691 637 165 2,600 5,354
SAS 3,739 2,529 1,646 517 12,257 20,687
EAS 3,685 6,134 4,843 742 11,840 27,243

Figure S4: Total number of avoided deaths in 2050 compared to 2005/07 by age group.

21
Figure S5: Health analysis of dietary change for the year 2050 relative to 2005/07 by risk
factor and region. Risk factors include change in fruit and vegetable consumption
(ΔC(fruit&veg)), change in red meat consumption (ΔC(red meat)), change in prevalence of
overweight and obesity combined (Δweight), and all risk factors combined (Total). Regions
include an aggregate of all world regions (WLD); developed countries (DPD), which include
Western high-income countries (HIC) and Western middle-income countries (MIC); and
developing countries (DPG), which include Latin America and the Caribbean (LAC), Sub-
Saharan Africa (SSA), Eastern Mediterranean (EMR), South Asia (SAS), and East Asia
(EAS). (Upper panel) Number of avoided deaths (in millions) in the dietary scenarios in 2050
relative to 2005/07 by risk factor and region. (Lower panel) Number of avoided deaths per
capita (in deaths per thousand people) in the dietary scenarios in 2050 relative to 2005/07 by
risk factor and region.

22
Figure S6: Health analysis of dietary change for the year 2050 relative to the reference
scenario by risk factor and region. Risk factors include change in fruit and vegetable
consumption (ΔC(fruit&veg)), change in red meat consumption (ΔC(red meat)), change in
prevalence of overweight and obesity combined (Δweight), and all risk factors combined
(Total). Regions include an aggregate of all world regions (WLD); developed countries
(DPD), which include Western high-income countries (HIC) and Western middle-income
countries (MIC); and developing countries (DPG), which include Latin America and the
Caribbean (LAC), Sub-Saharan Africa (SSA), Eastern Mediterranean (EMR), South Asia
(SAS), and East Asia (EAS). (Upper panel) Number of avoided deaths (in millions) in the
dietary scenarios in 2050 relative to the reference scenario by risk factor and region. (Lower
panel) Number of avoided deaths per capita (in deaths per thousand people) in the dietary
scenarios in 2050 relative to the reference scenario by risk factor and region.

23
Figure S7: Change in avoided deaths per capita relative to the reference scenario in 2050.

24
SI.7 Supplementary environmental results

Figure S8: Food-related GHG emissions (GtCO2-eq) in 2005/07 and projected to 2050 for
each dietary scenarios by commodity. The percentages within the bars refer to the percentage
of food-related GHG emissions to total GHG emissions, using historical emissions data for
2005/07 and an emissions pathway that would limit global temperature increase to below 2°C
(32).

25
Fig. S9: Environmental analysis of dietary change for the year 2050 by food group and region.
Regions include an aggregate of all world regions (WLD); developed countries (DPD), which
include Western high-income countries (HIC) and Western middle-income countries (MIC);
and developing countries (DPG), which include Latin America and the Caribbean (LAC),
Sub-Saharan Africa (SSA), Eastern Mediterranean (EMR), South Asia (SAS), and East Asia
(EAS). (Upper panel) Food-related greenhouse gas (GHG) emissions in the dietary scenarios
in 2050 by food group and region. (Lower panel) Food-related GHG emissions per capita in
the dietary scenarios in 2050 by food group and region.

26
Fig. S10: Environmental analysis of dietary change for the year 2050 relative to the reference
scenario by food group and region. Regions include an aggregate of all world regions (WLD);
developed countries (DPD), which include Western high-income countries (HIC) and Western
middle-income countries (MIC); and developing countries (DPG), which include Latin
America and the Caribbean (LAC), Sub-Saharan Africa (SSA), Eastern Mediterranean
(EMR), South Asia (SAS), and East Asia (EAS). (Upper panel) Changes in food-related
greenhouse gas (GHG) emissions in the dietary scenarios relative to the reference scenario in
2050 by food group and region. (Lower panel) Changes in food-related GHG emissions per
capita in the dietary scenarios relative to the reference scenario in 2050 by food group and
region.

27
Figure S11: Change in GHG emissions per capita relative to reference scenario in 2050.

28
SI.8 Supplementary valuation results

Figure S12: Healthcare-related benefits (USD billion) by region and cost component of total
healthcare benefits, including saved direct healthcare costs (direct) and indirect costs of
informal care (indirect(care)) and lost labour productivity (indirect(productivity).

Figure S13: Health benefits estimated by value-of-statistical-life (VSL) approach by region:

developed countries (DPD), developing countries (DPG); Western high-income countries
(HIC), Western middle-income countries (MIC), Latin America and the Caribbean (LAC),
Sub-Saharan Africa (SSA), Eastern Mediterranean (EMR), South Asia (SAS), East Asia
(EAS).

29
Figure S14: Value of environmental benefits by region: developed countries (DPD),
developing countries (DPG); Western high-income countries (HIC), Western middle-income
countries (MIC), Latin America and the Caribbean (LAC), Sub-Saharan Africa (SSA),
Eastern Mediterranean (EMR), South Asia (SAS), East Asia (EAS). The error bars denote
environmental benefits obtained by using different social costs of carbon which are associated
with different discount rates (main: 3%, low: 5%, high: 95th percentile of 3%).

30
Supplementary discussion

SI.9 Emissions feedback from reductions in mortality

In our emissions analysis, we accounted for the impacts that dietary changes have on food-
related GHG emissions. However, changes in mortality could generate feedbacks on
emissions which might counteract the diet-related impacts. For example, one would expect
that if people lived longer due to diet-reductions in mortality, they would also contribute a
greater amount of GHG over their life cycle. Here we analyse such feedbacks and find that the
additional emissions due to diet-related reductions in mortality could compensate about 4-6%
of the total reductions in food-related GHG emissions that are associated with dietary changes
(Table S9). One way of contextualizing these emissions feedbacks is by noting that the
number of avoided deaths (5.1-8.1 million, depending on the scenario) represent less than a
tenth of a percent of the projected population in 2050 (> 9 billion). Thus, changes in the
dietary habits of the whole population have a much greater impact on food-related GHG
emissions than the feedback on those emissions from the relatively small number of avoided
deaths.

Table S9: Sensitivity analysis of emissions feedback related to changes in mortality.

Scenario
Row Parameter
HGD VGT VGN

1 Years of life saved (YLS) (millions) 79 114 129

2 Food-related emissions per capita (tCO 2-eq per person per year) 0.89 0.46 0.37

3 Food-related emissions reductions (GtCO 2-eq) 3.3 7.2 8

4 Food-related emissions due to YLS Absolute emissions (MtCO2-eq) 71 53 48

(MtCO2-eq)
5 Percent of food-related emissions reduction (%) 2.2 0.7 0.6

Ratio of non-food-related GHG emissions under a 2°C emissions trajectory to food-

6 2.7 5.3 6.7
related emissions (see Fig. S8)

7 Total YLS-related emissions Absolute emissions (MtCO2-eq) 191 280 322

(MtCO2-eq)
8 Percent of food-related emissions reductions (%) 5.8 3.9 4
Notes: Rows 1-3 and 6 are main results; row 4 is calculated by multiplying rows 1 and 2; row 5 is calculated by dividing row 4 by row 3;
row 7 is calculated by multiplying rows 4 and 6; row 8 is calculated by dividing row 7 by row 3.

31
SI.10 Supplementary discussion of environmental results

In our environmental analysis, we projected food-related GHG emissions to increase from 7.6
GtCO2-eq in 2005/07 to 11.4 GtCO2-eq in the reference scenario in 2050; and we projected
dietary changes in the HGD, VGT, and VGN scenarios to decrease the reference emissions in
2050 by 3.3-8.0 GtCO2-eq (29-70%). In aggregate, our environmental results are in line with
the environmental literature on dietary change. Hedenus et al (13) projected food-related
GHG emissions to increase from 8 GtCO2-eq in 2000 to 12 GtCO2-eq in 2050, and that
dietary changes that ranged from replacing 75% of ruminant meat and dairy by other meat to
replacing 75% of all animal foods by pulses and cereals on kcal basis would reduce food-
related GHG emissions by 3.4-5.2 GtCO2-eq (net of technical mitigation in the agricultural
sector and increased productivity which would add another 2 and 1.7 GtCO2-eq to the
estimate). Tilman and Clark (12) projected food-related GHG emissions to increase from 8.43
GtCO2-eq in 2009 to 15 GtCO2-eq in 2050; and that adopting Mediterranean, pescatarian,
and vegetarian diets would reduce emissions by 4.22-8.44 GtCO2-eq (30-60%). Although we
adopted the same baseline GHG emissions factors as Tilman and Clark, our estimates are
slightly lower than theirs, because we also accounted for output-based productivity
improvements in agriculture which also improve (i.e., lower) emissions intensities, and we did
not account for the GHG emissions associated with the consumption of fish and seafood.
Another difference between our study and Tilman and Clark’s is that we use consumption
projections produced by FAO, whereas Tilman and Clark generate income-dependent dietary
projections. Our final environmental comparison is to Bajzelj et al (7) who projected that
adopting a dietary pattern based on dietary recommendations would decrease food-related
GHG emissions by 5.8-6.4 GtCO2-eq depending on the underlying reference scenario. In
contrast to our study, Bajzelj et al included land-use emissions (which is why reference
emissions cannot be compared) and their dietary scenario is largely based on advice from US
institutions and associations (Harvard Medical School, American Heart Association), which
makes it more stringent than our HGD scenario that is based on global recommendations.

32
SI.11 Supplementary discussion of valuation results

We are not aware of other studies that contrasted the value of environmental and health
benefits. However, some separate valuation studies exist. In our economic analysis, we
estimated the value of environmental benefits from avoided damages to range from 89-1,773
billion in 2050 (0.04-0.77% of global GDP) depending on the dietary scenario and the
discount rate used. Using an integrated assessment model, Stehfest et al (11) estimated
mitigation cost savings of 0.56% of GDP if dietary recommendations on healthy eating would
be adopted – an estimate that is within the range obtained here. With regards to health-related
costs, Bloom et al (34) estimated the economic burden of cancer and cardiovascular diseases
in 2030 to be USD 1.46 trillion using a cost-of-illness approach and USD 43.4 trillion using a
value-of-statistical-life approach. Re-doing our analysis for the year 2030 yielded reductions
in the number of deaths from stroke, CHD, and cancer of 12-20% and monetized health
benefits of USD 314 billion (219-408 billion), USD 445 billion (311-578 billion), and USD
500 billion (350-650 billion) in HGD, VGT, and VGN scenario, respectively, when using the
cost-of-illness approach, and of USD 10 trillion (5-15 trillion), USD 14 trillion (7-22 trillion),
and USD 16 trillion (8-24 trillion) in the HGD, VGT, and VGN scenario, respectively, when
using the value-of-statistical-life approach. Applying the percentage mortality reduction to the
economic estimates of Bloom et al yields values that fall within those ranges (USD 175-292
billion for the cost-of-illness estimate and USD 5.2-8.7 trillion for the value-of-statistical-life
estimate).

33