EAT–Lancet diet score requires minimum intake values to predict higher micronutrient adequacy of diets in rural women of reproductive age from five low- and middle-income countries

Abstract

The EAT–Lancet Commission promulgated a universal reference diet. Subsequently, researchers constructed an EAT–Lancet diet score (0–14 points), with minimum intake values for various dietary components set at 0 g/d, and reported inverse associations with risks of major health outcomes in a high-income population. We assessed associations between EAT–Lancet diet scores, without or with lower bound values, and the mean probability of micronutrient adequacy (MPA) among nutrition-insecure women of reproductive age (WRA) from low- and middle-income countries (LMIC). We analysed single 24-h diet recall data (n 1950) from studies in rural DRC, Ecuador, Kenya, Sri Lanka and Vietnam. Associations between EAT–Lancet diet scores and MPA were assessed by fitting linear mixed-effects models. Mean EAT–Lancet diet scores were 8·8 (SD 1·3) and 1·9 (SD 1·1) without or with minimum intake values, respectively. Pooled MPA was 0·58 (SD 0·22) and energy intake was 10·5 (SD 4·6) MJ/d. A one-point increase in the EAT–Lancet diet score, without minimum intake values, was associated with a 2·6 (SD 0·7) percentage points decrease in MPA (P < 0·001). In contrast, the EAT–Lancet diet score, with minimum intake values, was associated with a 2·4 (SD 1·3) percentage points increase in MPA (P = 0·07). Further analysis indicated positive associations between EAT–Lancet diet scores and MPA adjusted for energy intake (P < 0·05). Our findings indicate that the EAT–Lancet diet score requires minimum intake values for nutrient-dense dietary components to avoid positively scoring non-consumption of food groups and subsequently predicting lower MPA of diets, when applied to rural WRA in LMIC.

Ensuring food systems deliver improved human and planetary health is among the most pressing global challenges of the 21st century⁽¹⁾. Global food production threatens climate stability and ecosystem resilience and constitutes the single largest driver of environmental degradation and transgression of planetary boundaries⁽²⁾. Furthermore, low-quality diets are responsible for the greatest burden of disease worldwide⁽³⁾, affecting countries and population groups at all levels of socio-economic development⁽⁴⁾. To facilitate reversing these detrimental global trends, the EAT–Lancet Commission drew on state-of-the-art nutritional and environmental science as a yardstick for healthy diets from sustainable food systems⁽¹⁾. The report thus defined a ‘flexitarian’ universal reference diet and advocated a Great Food Transformation from business as usual towards win–win environmental sustainability and human health outcomes by the year 2050.

At present, academics and public health professionals are calling for accurate, robust and cross-cutting metrics (and data) to track diet quality and environmental sustainability goals at global, country and regional levels^(5,6) . Accordingly, Knuppel et al.⁽⁷⁾ constructed a novel EAT–Lancet diet score (0–14 points) and reported its inverse associations with risks of major health outcomes in a high-income adult population. EAT–Lancet diet score might proxy diet quality, which is one guiding principle of healthy sustainable diet indicators^(8,9) . Furthermore, in contrast to other diet quality scores, for example, Healthy Eating Index-2015⁽¹⁰⁾, Alternate Healthy Eating Index-2010⁽¹¹⁾ and Mediterranean Diet Score^(12,13) , the EAT–Lancet diet score is designed for global application and therefore aims to make meaningful comparisons across countries, populations and cultures. Moreover, the EAT–Lancet diet score extends previous universal diet quality metrics, based on adherence to WHO global nutrition guidelines, that is, Healthy Diet Indicators^(14,15) , by incorporating environmental health considerations⁽¹⁾ and simplifying food composition data requirements (no cholesterol, n-3 or n-6 PUFA components, etc.) to assess adherence to universal dietary recommendations in the Global South⁽¹⁶⁾.

This being said, in the EAT–Lancet Commission’s report, nutrient adequacy of the universal reference diet was calculated by assuming consumption (i.e. mean value of the proposed intake ranges) from each of the fourteen recommended dietary components and used food composition data primarily from US sources⁽¹⁾. Surprisingly, however, the healthy reference diet developed by Willett et al.⁽¹⁾ and the subsequent EAT–Lancet diet score by Knuppel et al.⁽⁷⁾ set minimum intake values for multiple nutrient-dense food groups at 0 g/d. A priori, the authors of this research acknowledged these intake/scoring ranges as a potential pitfall of the EAT–Lancet diet score in food and nutrition insecure low- and middle-income countries (LMIC). Lawrence et al.⁽¹⁷⁾ also argued for further development of the EAT–Lancet universal reference diet narrative, which must specifically clarify the inclusion of zero consumption recommendations (i.e. no minimum intake values) for various nutrient-dense dietary components. In particular, women of reproductive age (WRA) remain more susceptible than men to malnutrition and food insecurity, that is, higher prevalence of non-consumption of dietary components (0 g/d), across every continent⁽¹⁸⁾, but most notably in resource-poor settings^(19,20) . Hence, women’s nutrition and health in LMIC have received increased global political attention and resource allocation during the Millennium and Sustainable Development Goals Era^(21–23). At present, the United Nations Decade of Action on Nutrition 2016–2025 and the 2030 Agenda for Sustainable Development provide global and national stimuli to monitor and address global malnutrition outcomes, with specific targets for women.

To the authors’ knowledge, no previous study has assessed the association between the EAT–Lancet diet score and the (micro) nutrient adequacy of diets for rural WRA in LMIC. Thus, the EAT–Lancet diet score’s potential application to monitor and evaluate global adherence to healthy and sustainable food consumption in the Anthropocene is currently unknown (i.e. cross-cutting metric for Sustainable Development Goals and Paris Agreement targets). Therefore, the present study used local food composition tables and evaluated EAT–Lancet diet scores⁽⁷⁾, without or with (>0 g/d) minimum intake values for nutrient-dense dietary components, as a potential predictor of higher Mean Probability of Micronutrient Adequacy (MPA) of diets in rural WRA from five LMIC across Africa, Asia and Latin America.

Methods

Data sources

Cross-sectional secondary data from five LMIC were used in the present study (online Supplementary Table S1). Dietary intake data of non-pregnant non-lactating WRA (15–49 years) were collected using a quantitative and comparable single multiple-pass 24-h dietary recall method that identified food and drinks to at least the species level⁽²⁴⁾. Dietary intake data during the wet (lean) season were obtained from rural areas in the Democratic Republic of Congo (n 375⁽²⁵⁾), Ecuador (n 201⁽²⁶⁾), Kenya (n 361⁽²⁷⁾), Sri Lanka (n 20) and Vietnam (n 262). Dietary intake data from the dry (plenty) season were also available from Kenya (n 362⁽²⁷⁾) and Vietnam (n 369). All studies considered agricultural and wild sources of food and drinks, without assessment of dietary supplement intake. All data were collected between July 2009 and April 2015, and samples were representative of the village-level populations. Food composition data for Ca (mg), folate (μg), Fe (mg), vitamin A (μg retinol equivalents), vitamin C (mg) and Zn (mg), available for all five LMIC, were mostly sourced from national food composition tables, to capture the substantial variations in content and density of essential micronutrients between- and within-food species^(5,28,29) . In the event, food composition data were missing and best-matching values were obtained from similar settings, countries, or food and drinks (online Supplementary Table S1).

Ethics

Anonymised individual-level data and protocols are available at https://dataverse.harvard.edu/dataverse/DietarySpeciesRichness. All studies were approved by an ethics committee, except in Sri Lanka where the protocol was exempted from clearance. The present analysis was approved by the Ethics Committee of Ghent University Hospital (NR B670201422403). Our research was reported using the STROBE-nut checklist (online Supplementary Text 1 ⁽³⁰⁾).

EAT–Lancet diet scores

First, using data from the 24-h dietary recall method (n 1950), we calculated the EAT–Lancet diet score⁽⁷⁾ based on fourteen key recommendations in Willett et al.⁽¹⁾ (Table 1). The EAT–Lancet universal reference diet corresponds to the average energy needs (10·5 MJ/d or 2500 kcal/d) of a 30-year-old woman weighing 60 kg and whose physical activity level is between moderate and high (1·7–2·0). The serving ranges for each food group were derived from state-of-the-art scientific evidence with regard to human and planetary health impacts of foods in the Anthropocene⁽¹⁾. As recommended in Knuppel et al.⁽⁷⁾, diets were assigned one point for meeting the amount of dietary intake (g/d) and proportion of energy (MJ/d) for whole grains, recommended for each dietary component, resulting in a possible score ranging from 0 to 14 points.

Table 1. Construction of the EAT–Lancet diet score without or with minimum intake values

* Each dietary component contributed 0 or 1 point resulting in a total score ranging from 0 to 14 points.

† Food species consumed per food group are described in Lachat et al.⁽⁵⁾ (online Supplementary Table S5).

‡ Recommendations used for the associations with ischaemic heart disease, stroke, diabetes and all-cause mortality in Knuppel et al.⁽⁷⁾ (online Supplementary Table S1).

§ Intake values used for the assessment of nutrient adequacy in Willett et al.⁽¹⁾ (online Supplementary Table S4).

|| Reference diet refers to dry, raw weight. Equivalent dry weights assigned based on the third supplement to McCance & Widdowson’s The Composition of Foods, 4th edition (1988)⁽³¹⁾.

¶ Reference diet refers to dry, raw weight. Equivalent dry weights assigned based on the fifth supplement to McCance & Widdowson’s The Composition of Foods, 4th edition (1991)⁽³²⁾.

Nevertheless, as previously described, nutrient adequacy of the ‘flexitarian’ universal reference diet was calculated by assuming consumption (mean value of the proposed intake ranges, e.g. 29 g/d of chicken and other poultry; Table 1) from each of the fourteen recommended dietary components⁽¹⁾. However, the healthy reference diet promulgated by Willett et al.⁽¹⁾ set lower bound intake values for multiple nutrient-dense food groups at 0 g/d. A priori, we acknowledged the absence of minimum intake requirements for tubers and starchy staples (dietary component no. 2), dairy foods (no. 5) and protein sources (nos. 6–11), as a potential pitfall of the EAT–Lancet diet score for vulnerable populations in food and nutrition insecure settings. To illustrate, WRA consuming a monotonous diet composed of 436 g/d whole-grain hard red spring wheat (±5·9 MJ/d), 91 g/d added rapeseed oil (±3·4 MJ/d) and 30 g/d added granulated sugar (±0·5 MJ/d) would counterintuitively score 11 points from a maximum of 14.

Therefore, we re-calculated the EAT–Lancet diet score (Table 1) with minimum intake values (i.e. mean value of the proposed intake ranges, as used for the nutrient adequacy calculations in online Supplementary Table S4 of the EAT–Lancet Commission’s report⁽¹⁾), to avoid assigning zero consumption (0 g/d) of a dietary component as one point.

Mean probability of nutrient adequacy

As a measure of the micronutrient adequacy of diets, we calculated the probability of adequacy (PA) for Ca, folate, Fe, vitamin A, vitamin C and Zn and the MPA for each WRA over a 24-h period using the probability approach⁽³³⁾. We used the estimated average requirements (EAR) and CV from the FAO & WHO⁽³⁴⁾. For Fe requirements, which are known to be skewed for non-pregnant, non-lactating women, we used the IOM’s reference tables⁽³⁵⁾, but adjusted for absorption of 10 % on the basis of diet patterns, according to FAO and WHO guidance⁽³⁴⁾. For Zn, we use the International Zinc Nutrition Consultative Group’s EAR and CV⁽³⁶⁾, assuming low absorption (25 %). The EAR used for each micronutrient are reported in online Supplementary Table S3. MPA was calculated as the mean of the PA of the six individual micronutrients. In parallel to micronutrient PA, the MPA has a possible range of 0–1. In addition, as a measure of the micronutrient density of diets, we calculated energy-adjusted MPA (MPA/total energy intake (MJ/d)) for each WRA.

Statistical analysis

Data management and statistical analysis were conducted in Stata version 15.1⁽³⁷⁾. A two-sided significance level of P < 0·05 was applied for all analyses. WRA were considered equally representative; therefore, overall summary statistics (mean values and standard deviations; %) were calculated averages for WRA per country and across countries. We compared EAT–Lancet diet scores and MPA between seasons (i.e. Kenya and Vietnam only) using Welch’s independent-samples t test.

To assess the associations between EAT–Lancet diet scores, without or with minimum intake values, and MPA (or energy-adjusted MPA), linear mixed effects models (mixed; random intercept: country; random slope: varying association by country) were fitted, assuming an unstructured covariance matrix.

To examine any differences in EAT–Lancet diet scores, without or with minimum intake values, by season and total energy intake, we tested exploratory interaction terms between EAT–Lancet diet scores and season and total energy intake as a binary variable (i.e. 0 for <10·5 MJ/d and 1 for ≥10·5 MJ/d underpowered for the latter group in Ecuador and Sri Lanka). Furthermore, we repeated the analyses, for EAT–Lancet diet scores with each individual recommendation adjusted for the score (minus itself) to investigate whether one or more dietary recommendations were responsible for our associations⁽³⁸⁾. Moreover, EAT–Lancet diet scores were potentially associated with total energy intake (MJ/d)⁽³⁹⁾. Therefore, to evaluate the micronutrient density of the diet, we tested associations between EAT–Lancet diet scores, without or with minimum intake values, and standardised energy-adjusted MPA (percentage points/MJ; sd).

Results

Sample characteristics

24-h Dietary recall method data were obtained for 1219 (62·5 %) WRA in the wet and 731 WRA in the dry season (mean age 28·5 (sd 7·8) years). The difference between EAT–Lancet diet scores, without or with minimum intake values (mean 6·8 (sd 1·1) points), ranged between 6·0 (sd 1·1) in Ecuador and 7·5 (sd 0·8) in the Kenya. MPA (mean 0·58 (sd 0·22)) ranged between 0·45 (sd 0·18) in Sri Lanka and 0·64 (sd 0·18) in Vietnam (Table 2). Diets were particularly non-adhering to the EAT–Lancet diet score, without minimum intake values, with regard to whole grains (43·2 %), vegetables (28·8 %), fruits (16·5 %), nuts (3·3 %) and added fats (20·9 %) (online Supplementary Table S2). In contrast, diets were non-adhering to the EAT–Lancet diet score, with minimum intake values, for all key dietary components (<50 %), except added sugar (63·6 %) (online Supplementary Table S4). Diets were particularly inadequate for Ca (mean PA 0·12 (sd 0·30)) and folate (mean PA 0·31 (sd 0·43); online Supplementary Table S3). The EAT–Lancet diet score, without minimum intake values, was higher in the wet (lean) season (P = 0·030), whereas the EAT–Lancet diet score, with minimum intake values, was higher in the dry (plenty) season (P = 0·024). MPA was lower (P < 0·001) in the wet (lean) season, and thus dietary variables were not comparable across seasons when only countries with data for both seasons were used (i.e. Kenya and Vietnam; n 1354; Table 2).

Table 2. EAT–Lancet diet scores, without or with minimum intake values, and mean probability of adequacy in women of reproductive age by country and season

(Mean values and standard deviations)

* Two-sided Welch’s independent-samples t test significant by season at the 5 % level.

Associations between EAT–Lancet diet scores and MPA

The EAT–Lancet diet score, without minimum intake values, was consistently inversely associated with MPA across countries (Fig. 1; online Supplementary Fig. S1). Each one-point increase in the EAT–Lancet diet score, without minimum intake values, was associated with a 2·6 (sd 0·7) percentage points decrease in MPA (P < 0·001) (Table 3). The associations were not explained by one single dietary recommendation (non-consumption included in criteria for scoring one point), suggesting a negative cumulative effect of the EAT–Lancet diet score’s dietary components without minimum intake values (online Supplementary Table S6). However, 1923 (98·6 %) WRA achieved the recommendations (including 0 g/d) for dairy foods, 1724 (88·4 %) for eggs, 1773 (90·9 %) for fish, 1761 (90·3 %) for dry beans, lentils and peas, 1948 (99·9 %) and only 64 (3·3 %) for peanuts and tree nuts, suggesting that a subset of the recommendations contributed to the lower MPA associated with better adherence to the EAT–Lancet diet score without minimum intake values. The interaction term between the EAT–Lancet diet score, without minimum intake values, and season was significant in Kenya and Vietnam (P < 0·05). Nevertheless, the direction of the modified association between the EAT–Lancet diet score and MPA remained unchanged. Furthermore, the interaction term between the EAT–Lancet diet score, without minimum intake values, and total energy intake (<10·5 or ≥10·5 MJ/d) was also significant (P = 0·003), although associations between the EAT–Lancet diet score and MPA were non-significant (online Supplementary Table S5).

Fig. 1. Associations between mean probability of adequacy and EAT–Lancet diet score, without minimum intake values, in 1950 women of reproductive in five low- and middle-income countries. Kernel-weighted local polynomial smoothing plot. , Democratic Republic of Congo; , Sri Lanka; , Vietnam; , Ecuador; , Kenya.

Table 3. Associations between EAT–Lancet diet scores, with or without minimum intake values, and mean probability of adequacy in women of reproductive age by country and season

(β Values with their standard errors)

* P < 0·05, ** P < 0·01, *** P < 0·001.

† Mixed effects linear regression model random intercept: country; random slope: association between EAT–Lancet diet score and MPA by country.

In contrast, the EAT–Lancet diet score, with minimum intake values, was positively associated with MPA (Fig. 2; online Supplementary Fig. S2). Each one-point increase in the EAT–Lancet diet score, with minimum intake values, was non-significantly associated with a 2·4 (sd 1·3) percentage points increase in MPA (P = 0·07; Table 3). The associations were not explained by one single dietary recommendation, suggesting a positive cumulative effect (e.g. all vegetables, dairy products, and peanuts and tree nuts) of the EAT–Lancet diet score’s dietary components with minimum intake values (online Supplementary Table S6). The interaction term between the EAT–Lancet diet score, with minimum intake values, and season was significant in Vietnam only (P = 0·003), but did not change the positive associations between the EAT–Lancet diet score, with minimum intake values, and MPA across both seasons. Moreover, the interaction term between the EAT–Lancet diet score, without minimum intake values, and total energy intake (<10·5 or ≥10·5 MJ/d) was significant (P = 0·018), although the positive association between the EAT–Lancet diet score and MPA was significant in the <10·5 MJ/d group only (P = 0·001; online Supplementary Table S5).

Fig. 2. Associations between mean probability of adequacy and EAT–Lancet diet score, with minimum intake values, in 1950 women of reproductive in five low- and middle-income countries. Kernel-weighted local polynomial smoothing plot. , Democratic Republic of Congo; , Sri Lanka; , Vietnam; , Ecuador; , Kenya.

Further analysis indicated consistent positive associations between EAT–Lancet diet scores, without or with (except the Democratic Republic of Congo) minimum intake values, and standardised energy-adjusted MPA across countries (Figs. 3 and 4). Each one-point increase in the EAT–Lancet diet score, without minimum intake values, was associated with an increase of 0·20 (sd 0·03) in energy-adjusted MPA (P < 0·001; online Supplementary Table S7 and Fig. S3). In parallel, a one-point increase in the EAT–Lancet diet score, with minimum intake values, was associated with an increase of 0·16 (sd 0·06) in energy-adjusted MPA (P = 0·011; online Supplementary Fig. S4).

Fig. 3. Associations between standardised energy-adjusted mean probability of adequacy and EAT–Lancet diet score, without minimum intake values, in 1950 women of reproductive in five low- and middle-income countries. Kernel-weighted local polynomial smoothing plot. , Democratic Republic of Congo; , Sri Lanka; , Vietnam; , Ecuador; , Kenya.

Fig. 4. Associations between standardised energy-adjusted mean probability of adequacy and EAT–Lancet diet score, with minimum intake values, in 1950 women of reproductive in five low- and middle-income countries. Kernel-weighted local polynomial smoothing plot. , Democratic Republic of Congo; , Sri Lanka; , Vietnam; , Ecuador; , Kenya.

Discussion

The EAT–Lancet diet score, without minimum intake values for all nutrient-dense dietary components, was consistently negatively associated with MPA in all five countries. In contrast, the EAT–Lancet diet score, with minimum intake values, was positively associated with MPA in all countries, except in the underpowered sample from Sri Lanka (which only included twenty WRA). Therefore, our findings advocate minimum intake values (>0 g/d), to avoid the EAT–Lancet diet score predicting non-consumption of nutrient-dense dietary components and subsequent lower micronutrient adequacy of diets in resource-poor settings. In parallel to our findings, Willett et al.⁽¹⁾ also noted that the universal reference diet’s dietary recommendations might need to be adapted/flexible to the local culture, geographic, social or economic circumstances, as a strict global adoption of the EAT–Lancet diet might not be an optimal (ethical or equitable) choice for all^(17,40,41) .

Our results are in contrast to research by Knuppel et al.⁽⁷⁾, which indicated that a similar EAT–Lancet diet score, without minimum intake values for multiple nutrient-dense food groups, was associated with lower risks of major health outcomes, such as ischaemic heart disease and diabetes (but was not associated with stroke or mortality), in adults of the EPIC-Oxford cohort. However, we argue that analyses of the EAT–Lancet diet score, without minimum intake values, and nutritional outcomes in high-income countries characterised by overconsumption⁽³⁾, are unlikely to suffer from the same limitations as in LMIC, that is, capturing non-consumption of nutrient-dense food groups (0 g/d) and subsequent lower micronutrient adequacy of diets in populations crippled by the triple burden of malnutrition⁽⁴²⁾. To illustrate, our findings indicate that diets were micronutrient inadequate (MPA < 0·60) in the study population of rural WRA in the Democratic Republic of Congo, Ecuador, Kenya and Sri Lanka (excluding Vietnam). Furthermore, in parallel to previous research on country-level adherence to dietary guidelines^(43,44) , the proportion of WRA meeting the fourteen key dietary recommendations, with minimum intake values, was low. Therefore, future nutrition and environmental research might apply the EAT–Lancet diet scores, with minimum intake values, in prospective cohort studies in LMIC (e.g. Prospective Urban Rural Epidemiology study⁽⁴⁵⁾) and model better adherence to multiple environmental impacts at national or regional scales⁽⁴⁶⁾. This research shows that the EAT–Lancet diet score, with minimum intake values, might potentially serve as a cross-cutting sustainable healthy diet indicator, to facilitate monitoring and evaluation of women’s dietary (and environmental, not assessed in this study) risk factors, rather than detrimental outcomes alone (e.g. prevalence of anaemia in WRA).

Furthermore, EAT–Lancet diet scores, without minimum intake values, and MPA were significantly different between wet (lean) and dry (plenty) seasons in Kenya and Vietnam. Our results might be attributable to the seasonal changes in local production systems, increased food availability and micronutrient density of food species associated with the plenty season⁽⁴⁰⁾. These findings are distinct from Lachat et al.⁽⁵⁾, although expected considering previous systematic reviews reported considerable intra-annual variation in nutritional quality of adult’s diets in highly biodiverse areas^(47,48) . Indeed, we report higher intake (g/d) of beef, lamb, pork, chicken, other poultry and eggs in the dry season (results available on request). Moreover, the EAT–Lancet diet score, with minimum intake values, was more positively associated with MPA in the dry season, suggesting that it might be easier to increase micronutrient adequacy of diets in the plenty season. Our findings are potentially attributed to the observed higher prevalence of WRA reaching the EAT–Lancet diet score’s recommendations, with minimum intake values, for vegetables, dairy foods and fish in the dry season.

Advocating global adherence to a ‘flexitarian’ EAT–Lancet diet (with lower bound intake values for animal-source food groups set at 0 g/d) has generated mixed responses and criticism within the global research and development community⁽¹⁷⁾. The state-of-the-art scientific basis for a predominately plant-based (or low animal-source food) universal reference diet⁽¹⁾, focused on promoting a diversity of whole grains, vegetables, fruits, legumes, nuts and unsaturated oils, low-to-moderate amounts of seafood and poultry, and no or low quantities of red meat, processed meat, added sugar, refined grains and starchy vegetables, is controversial^{(38,45,49–51)} , socially polarised^(17,52) and has been questioned for vulnerable population groups in LMIC^(41,53–55) . To illustrate, increased intakes of nutrient-dense foods, such as animal-source foods, are known to provide vital nutritional benefits, including protein and essential micronutrients (e.g. Fe, Zn and B-vitamins) to WRA in LMIC^(56–61). To add to the intricacies/wicked problem of healthy sustainable diets, Hirvonen et al.⁽⁶²⁾ reported that adherence to the EAT–Lancet Commission’s universal reference diet, with minimum intake values (MJ/d), requires relatively larger quantities of higher-cost food groups, such as dairy products, eggs, meat, fish, fruits and vegetables than near-subsistence diets (or alternative diets with minimally adequate levels of essential nutrients), rendering the diet unaffordable for approximately 1·5 billion people, mostly in sub-Saharan Africa and South Asia. Therefore, in order to address global diet-related inequities⁽⁶³⁾ and improve adherence to sustainable healthy diets^(41,62) , we argue that more dedicated and comprehensive multi-sectoral nutrition-sensitive and nutrition-specific policies and programmes are required.

Although there is a need to develop simple a priori global indicators for healthy sustainable diets, to ensure policymakers and consumers understand how such diets improve individual and population health and conserve natural resources⁽⁶⁴⁾, using crude scores, for example, the EAT–Lancet diet score has inherent statistical limitations, including subjective selection of components and cut-offs, single dietary components considered as independent (i.e. correlated structure or substitution effects (e.g. vegetal and animal protein sources) not considered) and assumptions of linear additive effects⁽⁶⁵⁾. Moreover, given the complexity of the magnitudes of impact dietary components has on human and planetary health⁽⁶⁶⁾, sustainable healthy diet indices might be more informative when composed of a suite of nutritional and environmental metrics^(46,67,68) .

As was the case with previous studies^(5,26,27,69) , the first limitation of the present research was a lack of nutrient composition data of certain foods consumed (limiting analysis to six micronutrients available across all five countries). The composition of various indigenous, wild, neglected and underutilised food species was often not available and was substituted with nutrient values from similar foods. Second, we were unable to use a cut-off for fibre intake to accommodate for the emphasis of the universal reference diet on whole grains. Third, we used only a single multiple-pass 24-h dietary recall method per WRA. Although this method is appropriate to estimate population average dietary intakes, it does not allow accounting for intra-person variability⁽⁷⁰⁾ and thus estimation of usual intake⁽³³⁾. Fourth, analyses of EAT–Lancet diet scores pertain only to a reference WRA, and recommendations differ among women depending on their height, weight, physical activity, pregnancy and lactation status. Further assessments of the applicability of EAT–Lancet diet scores in diets with a higher contribution of foods obtained from (peri-) urban markets or processed foods and diverse population groups are warranted.

Although the EAT–Lancet Commission’s dietary recommendations provide a valuable roadmap for healthy sustainable diets, which can be used to tailor (inter)national food-based dietary guidelines, the EAT–Lancet diet score requires minimum intake values, for all nutrient-dense food groups, to avoid positively scoring non-consumption (i.e. 0 g/d) of dietary components and subsequently predicting lower micronutrient adequacy of diets in rural WRA from LMIC across Africa, Asia and Latin America.