Defining diet quality: a synthesis of dietary quality metrics and their validity for the double burden of malnutrition

1Poor diet quality is a leading and preventable cause of adverse health globally, which includes both maternal and child health (MCH) and non-communicable diseases (NCDs)., The UN Sustainable Development Goals (SDGs) outline global consensus on social, economic, environmental, and health targets to be met by 2030, with most goals concerned with nutrition including one goal to end malnutrition. Yet rates of progress toward achieving the SDGs have been slow, and accelerated momentum is needed.,,,, To develop sound strategies and monitor progress toward these goals, the assessment of global dietary quality is essential.

2Although a variety of dietary metrics have been developed and used to summarise various components of a diet (eg, adequacy, quality, diversity),, there remains an absence of widely used, validated metrics to define the double burden of malnutrition, and to compare effectively across country settings. This absence at least partly relates to a historical distinction between global nutrition efforts for MCH versus NCDs.,, Malnutrition for MCH has traditionally been considered as insufficient caloric or other nutrient intake leading to insufficient physical growth (stunting), rapid weight loss or failure to gain weight (wasting), cognitive impairment, exacerbation of anaemia and blindness, or weakening of the immune system resulting in increased risk of infectious diseases and mortality. Malnutrition for NCDs has historically been considered as excess consumption of certain nutrients such as fat, sugar, and salt as well as calories., Yet, malnutrition for MCH and NCDs are frequently coexisting consequences of poor diet quality within populations, households, or individuals across the lifespan., A unified global dietary quality metric (hereafter referred to as dietary metric) would aid in policy and programme decision making around improving diet in which components that contribute to both MCH and NCDs would be represented in the same assessments, and that the relative contributions of malnutrition for MCH and NCDs could be characterised and compared across settings in a standardised way.

3The optimal dietary metric for assessing malnutrition for both MCH and NCDs remains uncertain. Previous reviews have described the development and characteristics of dietary metrics.,,,,,,, Some reviews qualitatively summarised the evidence for dietary metrics in relation to a single health outcome (eg, type 2 diabetes, obesity), and others focused on several MCH or NCD outcomes.,,,, However, to our knowledge, the comprehensive reviews of dietary metrics in relation to several MCH or NCD outcomes,,,,, were done more than a decade ago, and no review has focused on both MCH and NCD outcomes together. To understand whether any existing dietary metrics are validated against MCH or NCDs, or both, and what researchers and health organisations are using to measure various aspects of the dietary contribution to MCH and NCDs, we extensively reviewed and assessed the validity of existing multinational dietary metrics for MCH and NCD health outcomes.

4Many global efforts are focused on tackling the double burden of malnutrition including the UN Sustainable Development Goals. However, these targets might not be achieved without a practical and valid dietary metric to reduce malnutrition across global contexts.We identified 19 dietary metrics, including seven developed for maternal and child health (MCH), 12 developed for non-communicable diseases (NCDs) and none developed or applied for both.All MCH dietary metrics used foods only, while most NCD metrics included foods and nutrients. The most frequent components for both MCH and NCD dietary metrics were vegetables, fruits, grains, roots, and tubers.When the validity was assessed, we found that four NCD metrics had convincing evidence of protective associations with specific NCD outcomes, primarily mortality, cardiovascular disease, type 2 diabetes, and total cancer. The remaining eight NCD dietary metrics and all seven MCH dietary metrics did not.We found that few dietary metrics have been validated against MCH or NCD health outcomes, and none for both indicates. More work is needed to validate existing and novel dietary metrics for MCH and NCD, an approach that is likely critical to the achievement of global nutrition and health targets.

5Many global efforts are focused on tackling the double burden of malnutrition including the UN Sustainable Development Goals. However, these targets might not be achieved without a practical and valid dietary metric to reduce malnutrition across global contexts.We identified 19 dietary metrics, including seven developed for maternal and child health (MCH), 12 developed for non-communicable diseases (NCDs) and none developed or applied for both.All MCH dietary metrics used foods only, while most NCD metrics included foods and nutrients. The most frequent components for both MCH and NCD dietary metrics were vegetables, fruits, grains, roots, and tubers.When the validity was assessed, we found that four NCD metrics had convincing evidence of protective associations with specific NCD outcomes, primarily mortality, cardiovascular disease, type 2 diabetes, and total cancer. The remaining eight NCD dietary metrics and all seven MCH dietary metrics did not.We found that few dietary metrics have been validated against MCH or NCD health outcomes, and none for both indicates. More work is needed to validate existing and novel dietary metrics for MCH and NCD, an approach that is likely critical to the achievement of global nutrition and health targets.

6Many global efforts are focused on tackling the double burden of malnutrition including the UN Sustainable Development Goals. However, these targets might not be achieved without a practical and valid dietary metric to reduce malnutrition across global contexts.

7We identified 19 dietary metrics, including seven developed for maternal and child health (MCH), 12 developed for non-communicable diseases (NCDs) and none developed or applied for both.

8All MCH dietary metrics used foods only, while most NCD metrics included foods and nutrients. The most frequent components for both MCH and NCD dietary metrics were vegetables, fruits, grains, roots, and tubers.

9When the validity was assessed, we found that four NCD metrics had convincing evidence of protective associations with specific NCD outcomes, primarily mortality, cardiovascular disease, type 2 diabetes, and total cancer. The remaining eight NCD dietary metrics and all seven MCH dietary metrics did not.

10We found that few dietary metrics have been validated against MCH or NCD health outcomes, and none for both indicates. More work is needed to validate existing and novel dietary metrics for MCH and NCD, an approach that is likely critical to the achievement of global nutrition and health targets.

11We defined a dietary metric as a metric derived from nutrients or food or food groups, or both, with the aim of measuring dietary diversity (number of different foods consumed), or nutrient adequacy (achievement of recommended intakes of energy or essential nutrients). We included dietary metrics with the reported intended use of relating diet to MCH (micronutrient adequacy, mortality in children <5 years, maternal mortality, underweight, stunting, wasting, infectious diseases, diarrhoeal disease) and NCD health outcomes (all-cause mortality, cardiovascular disease, type 2 diabetes, gestational diabetes; total cancer and subtypes, and anthropometrics in adults and children), and that quantified the level of intake of foods or nutrients consumed or the achievement of recommended intakes. A detailed description of the health outcomes is shown in the appendix (p 1). We selected the health outcomes in accordance with WHO's categories of malnutrition. Given the interest in dietary metrics that can be used globally, we included dietary metrics that were used for this purpose in at least three countries. We did not include dietary metrics that included non-dietary factors as a core component, that only summarised a single nutrient or food, or that were developed with a primary focus on non-health outcomes (eg, sustainability). We did not include dietary metrics derived primarily from statistical data clustering approaches, such as cluster or factor analysis, because these are often not generalisable. We also did not include dietary metrics at the population level (eg, food supply) because we were interested in individual-level and household-level dietary metrics.

12Given the abundance of potential dietary metrics, we did not systematically identify all possible metrics. First, we compiled a list of dietary metrics through expert contacts, who were faculty members at the Friedman School of Nutrition Science, Tufts University, with expertise in food policy, agriculture, economics, international nutrition programming, humanitarian emergency relief, clinical nutrition, and epidemiology. This initial list was then used to search PubMed and Google to identify studies and development reports from government agencies, health organisations, and non-profit organisations up until Sept 21, 2018, that included these dietary metrics, met our inclusion criteria, and were published in English. These searches were complemented by hand-searching of the citations of all identified articles and reports to identify additional potential dietary metrics. We documented all identified dietary metrics that both did and did not meet the scope of our inclusion criteria (appendix p 5).

13To assess validity of each relevant dietary metric, we systematically searched PubMed from Oct 11, 2000, to April 17, 2020, to identify meta-analyses or narrative reviews (including systematic reviews without a meta-analysis, but not umbrella reviews) evaluating these dietary metrics in relation to MCH and NCD health outcomes. The search terms and restrictions are described in full in the appendix (pp 2–4).

14VM did the literature searches, assessed titles and abstracts of all identified studies and reviewed and extracted relevant data by hand using a standardised electronic spreadsheet. The intended purpose, food and nutrient components, scoring, reference period, validity, and reliability was tabulated on each dietary metric. For each published meta-analysis or narrative review, data were extracted on the number of studies included, pooled relative risks and corresponding uncertainty, sample sizes, number of events, follow-up durations, unit of exposures, and population characteristics. In most cases, the total number of participants was not reported in the original meta-analyses and was computed from summary tables of the individual studies. Because narrative reviews do not provide pooled estimates, the extractions were done for each individual study included in the narrative review. When more than one meta-analysis or narrative review was identified for each dietary metric-health outcome relationship, we included all published meta-analyses or narrative reviews. Questions or uncertainties related to article screening and extraction were resolved by discussion with another investigator (DM).

15Two investigators (VM and then either PW, RM, or DM), independently and in duplicate, used the most comprehensive (with the greatest number of participants and studies) or recent meta-analysis when available, followed by the most comprehensive (with the greatest number of participants and studies) or recent narrative review to assess and grade the evidence for validity of dietary metrics in relation to MCH and NCD health outcomes. In cases in which the most recent meta-analysis or narrative review was not the most comprehensive, we selected the most comprehensive. Each assessor extracted data on the number of studies and the direction of effects across studies (positive association, null association, negative association). If the number of studies with null associations was equal or greater to the number of studies with positive or negative associations, the relationship was classified as no association. For both meta-analyses and narrative reviews, we considered the direction of effects from each included individual study. The validity of each dietary metric was assessed using two criteria: first, the number of studies, and second, the consistency of evidence from meta-analyses and narrative reviews of prospective cohort studies or randomised controlled trials. Consistency was defined as the association is repeatedly observed in different populations and circumstances. For a consistent relationship, at least half of the associations were in the same direction and for an inconsistent relationship, fewer than half of the associations were in the same direction. Three assessment categories were established: consistent evidence for the dietary metric-health outcome relationship (positive, null, negative) from five or more studies; inconsistent evidence for the dietary metric-health outcome relationship from five or more studies; and little evidence from less than five studies. For assessing validity of each dietary metric against foods, nutrients, and other non-dietary metrics (eg, biomarkers, food insecurity indicators) we used the published papers or development reports that described the development process of the dietary metric. All disagreements among reviewing investigators regarding the assessment and grading of validity were resolved through discussion.

16In total, we identified 19 dietary metrics used in three or more countries each to assess diet quality in relation to various health outcomes (Table 1, Table 2, Table 3). Seven dietary metrics were primarily used for MCH: Dietary Diversity Score (DDS), Food Consumption Score (FCS), Food Variety Score (FVS), Household Dietary Diversity Score (HDDS), Infant and Young Child Minimum Dietary Diversity (IYCMDD), Minimum Dietary Diversity for Women (MDD-W), and Women's Dietary Diversity Score (WDDS) and Individual Dietary Diversity Score (IDDS). Twelve dietary metrics were most commonly used for NCDs: Alternative Healthy Eating Index (AHEI), Dietary Approaches to Stop Hypertension (DASH), Dietary Guidelines for Americans Adherence Index (DGAI), Dietary Inflammatory Index (DII), Dietary Quality Index International (DQI-I), Healthy Eating Index-2010 (HEI), Mediterranean Diet Quality Index for Children and Teenagers (KIDMED), Mediterranean Diet Score (MED), Prospective Urban Rural Epidemiology Diet Score (PURE), Recommended Foods Score (RFS), WHO Healthy Diet Metric (WHO-HDI), and World Cancer Research Fund and American Institute for Cancer Research Dietary Recommendations (WCRF-AICR). No dietary metrics were identified that were used to assess both MCH and NCDs.

17Of the seven MCH dietary metrics, two were assessed at the household-level (FCS, HDDS) and five at the individual level (DDS, FVS, IYCMDD, MDD-W, and WDDS and IDDS). All 12 NCD dietary metrics were assessed at the individual level. Most of the MCH dietary metrics were developed for describing micronutrient or caloric adequacy (FCS, IYCMDD, MDD-W, and WDDS and IDDS). Metrics measuring dietary diversity (DDS, FVS) or household food access (HDDS) were less common. NCD dietary metrics were designed to describe adherence to US (DGAI, HEI, RFS) or international (WHO-HDI, WCRF-AICR) dietary guidelines, diet patterns (AHEI, DASH, KIDMED, MED), or foods associated with chronic disease risk (DQI-I, PURE). One NCD dietary metric was designed to measure inflammatory potential (DII).

18The foods and nutrients included in each dietary metric is shown in figure 1. The median number of included food groups or nutrients per dietary metric is 9·5 (IQR 7·25–12·0). The MCH dietary metrics (8·5, IQR 7·25–9·75) included fewer food groups or nutrients compared with the NCD dietary metrics (11·5, IQR 7·75–20·25). All MCH dietary metrics (excluding the FVS which counts the number of unique foods eaten without specifying food groups or nutrients) used only food groups. Alternatively, six NCD dietary metrics used food groups only, and six used foods and nutrients. None of the identified dietary metrics used only nutrients.

19Among all dietary metrics (excluding the FVS), the most frequent foods or nutrients included were vegetables (18 metrics); fruits (17 metrics); dairy products (15 metrics); and grains, roots, and tubers (14 metrics). Among MCH dietary metrics, grains, roots, and tubers, fruits, vegetables, dairy products, and meat and fish were included in all dietary metrics; other frequent foods were legumes and nuts (five metrics), and eggs (four metrics). The most common foods in NCD dietary metrics were vegetables (12 metrics) and fruits (11 metrics), followed by dairy products (nine metrics), grains, roots, and tubers (eight metrics), legumes and nuts (eight metrics), and meat (eight metrics). Uncommon food groups included spices, condiments, and beverages; fast foods; and sugar-sweetened beverages. Unhealthy food groups (eg, sugar-sweetened beverages, sugar, sweets, candy, fast food, or empty calorie foods) were included in two MCH dietary metrics (FCS, HDDS) and six NCD dietary metrics (DASH, AHEI, DQI, HEI, KIDMED, WCRF-AICR). Of the eight NCD dietary metrics that included nutrients (AHEI, DASH, DGAI, DII, DQI-I, HEI, MED, WHO-HDI), the most frequent nutrients were sodium (six of eight metrics), total fat, saturated fat, and fibre (four of eight each).

20All dietary metrics were intended to assess habitual diet, five MCH dietary metrics asked about intake in the past 24 hours, and all NCD dietary metrics asked about diet over an extended period, often the past year.

21Methods of scoring varied considerably. Five MCH dietary metrics (FVS, HDDS, IYCMDD, MDD-W, and WDDS and IDDS) and four NCD dietary metrics (DDS, KIDMED, RFS, WHO-HDI) assigned a binary value if the foods or nutrients were consumed during the reference period or recommended intakes were met. Several other NCD dietary metrics (AHEI, DGAI, WCRF-AICR) scored with more than two categories. To identify scoring thresholds, three NCD dietary metrics (MED, DASH, PURE) used the intake distributions in the population in which the metric was being used (ie, using medians and quintiles). One MCH (FCS) and three NCD dietary metrics (DII, DQI-I, HEI) used complex, component-specific scoring.

22Positive value only scores were used by all MCH dietary metrics and most NCD dietary metrics, except the KIDMED and DII. Four NCD dietary metrics (DASH, KIDMED, MED, PURE) assigned scores based on the perceived directional benefit of the foods or nutrients, or both, with both lower intake of unhealthy items and greater intake of healthy items receiving higher scores. Unequal scoring weights were used by one MCH dietary metric (FCS) and three NCD dietary metrics (DQI-I, DII, HEI).

23Once the overall dietary metric score was determined, assessment of the score varied. Most dietary metrics (15 of 19) used continuous assessments (ie, higher is better). Two MCH dietary metrics (ICYMDD, MDD-W) specified binary cutoffs, while one MCH (FCS) and one NCD dietary metric (KIDMED) provided ordinal (multi-category) cutoffs.

24Reliability, defined as repeated measure validity, was assessed for only two NCD dietary metrics (appendix pp 6–13). The HEI metric showed reasonable reliability for all components except sodium and dairy. The KIDMED metric showed moderate to excellent test-retest reliability.

25From their development reports, three MCH dietary metrics were validated against foods or nutrients (IYCMDD, MDD-W, and WDDS and IDDS) and two MCH metrics against other non-dietary metrics (FCS, HDDS; appendix pp 6–13). Three NCD dietary metrics were validated against foods or nutrients (HEI, DGAI) or biomarkers (DII).

26Our systematic search identified 48 meta-analyses or narrative reviews,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, that assessed validity of these dietary metrics against health outcomes (figure 2). These reports included 126 dietary metric-health outcome relationships for 13 of the 19 dietary metrics. Nearly all dietary metric-health outcome relationships (116 of 126) were for NCD dietary metrics, with only ten for MCH dietary metrics. Table 3 describes the identified associations of each dietary metric and MCH and NCD health outcome relationship from published meta-analyses, and the evidence for associations based on both meta-analyses and narrative reviews is shown in figure 3.

27The most commonly studied NCD dietary metrics were the MED (29 dietary metric-relationships), DASH (19 dietary metric-relationships), HEI (18 dietary metric-relationships), and AHEI (17 dietary metric-relationships). The most frequently studied outcomes were cancer (53 relationships), anthropometrics in children (21 relationships), cardiovascular disease (18 relationships), and anthropometrics in adults (13 relationships). Among our outcomes of interest, we did not find meta-analyses or narrative reviews of any of the dietary metrics in relation to major MCH outcomes including micronutrient adequacy, mortality for children younger than 5 years, maternal mortality, wasting, or infectious diseases. We also did not find meta-analyses or narrative reviews evaluating any dietary metric against both MCH and NCD outcomes.

28Four of the MCH dietary metrics (FCS, HHDS, MDD-W, and WDDS and IDDS) did not have any identified meta-analyses or narrative reviews assessing their relationship with MCH or NCD health outcomes (figure 3). We found little evidence from few studies for the DDS and stunting and diarrhoea, the IYCMDD and stunting, and the FVS and child underweight.

29Four of the NCD dietary metrics had consistent evidence for associations with NCD health outcomes (figure 3). The MED was found to be inversely associated with all-cause mortality, cardiovascular disease, cardiovascular disease mortality, coronary health disease, total and ischaemic stroke, type 2 diabetes, and cancer mortality. Both AHEI and HEI were associated with lower risk of all-cause mortality, cardiovascular disease, cardiovascular disease mortality, cancer, and cancer mortality, while DASH was inversely associated with all-cause mortality, cardiovascular disease, type 2 diabetes, cancer, cancer mortality, body-mass index (BMI), and weight loss in adults.

30We found inconsistent evidence for MED and cancer, BMI, waist circumference, and weight loss in adults; for KIDMED and BMI, waist circumference and overweight or obesity in children; and for HEI and colorectal cancer. Six NCD dietary metrics (AHEI, DASH, DQI-I, WHO-HDI, HEI, RFS) had little evidence for associations with breast cancer. For two NCD dietary metrics (DGAI, PURE), no meta-analyses or narrative reviews were identified of relationships with MCH or NCD health outcomes.

31Several dietary metrics had evidence (consistent or inconsistent) showing no association with NCD health outcomes including: DASH with total stroke; DII with breast cancer; HEI with BMI in adults and children; KIDMED with BMI, waist circumference, and overweight or obesity in children; and MED with breast cancer (figure 3).

32When we looked at what characteristics of the dietary metrics might be more predictive of validity of health outcomes we found that the four NCD metrics (AHEI, DASH, HEI, MED) with consistent evidence of associations were developed to describe adherence to dietary guidelines or diet patterns, consisted of food and nutrient components, all included healthy plant foods (fruits, vegetables, whole grains, nuts and legumes) and dairy, and most included red and processed meat, or sodium.

33In this comprehensive review of dietary metrics for assessing diet quality globally, we identified 19 dietary metrics, including seven for MCH and 12 for NCD outcomes; and none developed or applied for both. The dietary metrics varied substantially in their composition and scoring, with MCH dietary metrics generally focused on a few key foods (grains, fruits, vegetables, dairy products, meat, and fish) and NCD dietary metrics incorporating a more diverse mix of foods and nutrients, or both. Importantly, most of these dietary metrics were not validated against health outcomes in meta-analyses or narrative reviews. Only four NCD dietary metrics (MED, AHEI, HEI, DASH) had convincing evidence of protective associations, mainly for all-cause mortality, cardiovascular diseases, type 2 diabetes, total cancer, and cancer mortality. These four dietary metrics had little evidence of associations for anthropometrics in adults or children, and were not validated against MCH outcomes (micronutrient adequacy, mortality in children younger than 5 years, stunting, wasting, and infectious and diarrhoeal disease) in published meta-analyses or narrative reviews. The remaining eight NCD dietary metrics and all seven MCH dietary metrics were either not convincingly validated against most MCH and NCD health outcomes or published meta-analyses and narrative reviews assessing their validity were not identified. Additionally, our findings show that no dietary metrics currently exist designed or validated to characterise the double burden of malnutrition. To our knowledge, these findings provide the most current and extensive synthesis of specific multinational dietary metrics for risk of MCH and NCDs.

34Among MCH dietary metrics, the IYCMDD is the most widely used and is routinely collected in studies in low-income and middle-income countries, such as the Demographic and Health Surveys and UNICEF Multiple Indicator Cluster Surveys, as a component of the WHO's Minimum Acceptable Diet for children aged 6–23 months.,, Despite its frequent use, we identified one meta-analysis assessing the association of the IYCMDD with one MCH health outcome, and no meta-analyses or narrative reviews with NCD outcomes. We are aware of selected individual studies that have reported beneficial associations of IYCMDD with child growth outcomes (eg, stunting, wasting, height-for-age Z score, and weight-for-height Z score).,,,,,,,, However, such evidence has not been systematically reviewed nor summarised. The IYCMDD assesses binary recall of seven broad food groups over the previous 24 hours. Although data collection is quick and undemanding for research staff and participants, the simplicity of the IYCMDD is also a potential limiting factor in predicting health outcomes. Ironically, the existing widespread use of the IYCMDD could impede development and validation of new metrics for MCH or the double burden of malnutrition, because a substantial proportion of available dietary data in low-income and middle-income countries is restricted to IYCMDD's broad food groups. Our findings show that much work is needed to validate the IYCMDD for uses other than its intended purpose of measuring micronutrient adequacy. In the absence of robust validation of the IYCMDD against health outcomes, studies and organisations should consider more rigorously validated questions on diet quality.

35Among the other MCH dietary metrics, we identified only two narrative reviews for the DDS,, and four for the FVS.,,, No meta-analyses or narrative reviews were identified for the remaining MCH dietary metrics (FCS, HDDS, MDD-W, and WDDS and IDDS). We expect that individual studies will have assessed the relationship between these dietary metrics and MCH outcomes, and our findings highlight the need to pool and synthesise available individual studies of these dietary metrics in relation to both MCH and NCD outcomes in adults and children.

36Our investigation also found that although numerous dietary metrics are used to assess the dietary risks of NCDs, most do not have convincing evidence of associations that have been summarised in published narrative reviews or meta-analyses. Only four metrics (MED, AHEI, HEI, DASH) had generally consistent evidence of associations with specific NCD outcomes. For outcomes such as anthropometrics and certain cancer subtypes, further validation is needed. Additionally, none of these four NCD dietary metrics have been assessed in meta-analyses or systematic reviews of important MCH outcomes.

37Previous reviews of dietary metrics have focused solely on describing dietary metric features and dietary metric development,,,,,,, or the evidence for dietary metrics in relation to a single disease., Other reviews looked more broadly at dietary metrics and MCH or NCDs,,,,, but these are more than a decade old and did not also consider both MCH and NCDs together. Consistent with our findings, these previous studies generally found that MCH dietary metrics were not validated against MCH or NCD health outcomes, and that only a modest number of NCD dietary metrics predicted some chronic diseases. Our findings build upon and greatly extend these previous reviews by providing the most extensive contemporary summary of the evidence for associations of dietary metrics with both MCH and NCD outcomes.

38Future studies should consider analyses to validate existing dietary metrics with both MCH and NCD outcomes. Additionally, our findings suggest the need for development of novel dietary metrics designed from the start to assess and monitor both MCH and NCDs. Novel dietary metrics could be entirely new, or adaptations of existing dietary metrics and should be developed with several key considerations. First, comprehensive reviews of existing dietary metrics and health outcome relationships can be used to inform the features of new metrics (eg, foods or nutrients, scoring, cutoffs). We found that the four NCD dietary metrics with convincing evidence for associations were comprised of food and nutrient components, included healthy plant foods and dairy, and often included red and processed meat, or sodium. This finding might suggest that these components are important aspects of dietary metrics. However, the validity of most remaining dietary metrics has not been assessed against health outcomes, and it is not known whether dietary metrics without these components are less predictive of validity than those with these components. Second, systematic evaluations to summarise the aetiologic effects and optimal levels of dietary factors for MCH outcomes might provide hypotheses to inform the selection of foods and nutrients in new dietary metrics. In the past 10 years the growing availability of standardised, global dietary data has provided greater clarity on actual food and nutrient intakes,, and the relationships between foods and nutrients, and cardiovascular disease, type 2 diabetes and cancer are well documented, but similar associations for many MCH outcomes have not been systematically assessed. Third, new dietary metrics need to acknowledge and reconcile potentially opposing effects of some foods and nutrients on MCH and NCD outcomes. For example, animal source foods (eg, meat, fish, eggs, dairy) have been shown to have beneficial effects on micronutrient deficiencies, child growth outcomes, and cognitive function,,, but some of these foods, such as processed meats, are negatively associated with NCDs., Although we did not include sustainability metrics in our Review, an important consideration for future dietary metric development is the incorporation of the measurable effect of different food production systems on the environment in dietary metrics. Lastly, an influential consideration for the widespread adoption of existing dietary metrics by health agencies and governments is likely dependent on cultural appropriateness and ease of collecting and scoring these dietary metrics.

39Several challenges have likely limited validation of dietary metrics against both MCH and NCD. Ideally, prospective cohort studies with detailed and accurately measured data on both dietary habits and health outcomes should be leveraged. However, populations with MCH outcomes often do not have strong dietary assessment methods and longitudinal data. Most studies on MCH dietary metrics collect information on dietary intake over the previous 24 hour period because the priority is on identifying trends, comparisons, and rankings of outcomes within and between groups, which is generally of interest in setting and evaluating programme targets, and a longer reference period is not necessarily appropriate for these purposes and might be more difficult to operationalise. Moreover, these studies frequently assess diet qualitatively and the absence of quantitative data limit opportunities to construct and validate existing NCD dietary metrics. Ecological analyses are an alternative study design, which bring their own limitations but could address both MCH and NCD across low-income, middle-income, and high-income countries. In addition to these practical data challenges, one of the most important limitations of previous work might have been the general segregation of diet related MCH versus NCD research and policy.

40Our Review has several strengths. We did systematic searches for meta-analyses and narrative reviews of validation against health outcomes, making it less likely that we missed any existing syntheses of dietary metric and health outcome relationships. We included a broad range of health outcomes for MCH and NCDs, informing the modern priorities for the double burden of malnutrition. We evaluated the evidence for health associations independently and in duplicate, including the grading of consistency, reducing the potential for bias.

41Limitations should be considered. First, we focused on major MCH and NCD health outcomes, and there might be other relevant health outcomes that were not included, such as cognitive function, bone and joint disorders, and NCD risk factors (hypertension, dyslipidaemia, and glucose control). Second, our systematic reviews for each dietary metric in relation to MCH and NCD health outcomes was limited to published meta-analyses and narrative reviews and did not separately assess individual studies. However, meta-analyses and narrative review are more likely to provide accurate and reliable conclusions compared to single studies, minimising publication bias. Lastly, the evidence for validity came mostly from observational studies and some trials, and these are subject to residual confounding, which could overestimate associations, and measurement error in both diet assessment methods and outcome ascertainment, which might result in an underestimation of the relationships between the dietary metrics and MCH and NCD health outcomes.

42In summary, we identified seven international dietary metrics for MCH, none of which have been validated against health outcomes in meta-analyses or narrative reviews, and 12 dietary metrics for NCDs, of which only four have convincing evidence for validation against specific NCD outcomes. These findings highlight important gaps and major opportunities in global analyses of diet quality relating to malnutrition, which are highly relevant to the achievement of the UN SDGs, other global nutrition targets, and corresponding effective policy actions.

43We searched PubMed for articles published between Oct 11, 2000, and April 17, 2020, using the terms “diet quality”, “diet metric”, or “diet score” with “malnutrition”, “undernutrition”, “maternal health”, “child health”, “stunting”, “wasting”, “mortality”, “non-communicable disease”, “chronic disease”, “cardiovascular disease”, “diabetes”, “cancer”, or “obesity” with “meta-analysis”, “systematic review”, or “narrative review”. We used search terms in English but did not apply any language restrictions. We screened articles by title and abstract to identify full-text reports that were relevant to the study aims. We also screened citation lists for these full-text reports to identify other relevant articles. Articles were considered relevant if they reported the relation between dietary metrics and malnutrition. Several previous reviews qualitatively summarised the evidence for dietary metrics in relation to a single health outcome, and other reviews focused on several maternal and child health (MCH) or non-communicable disease (NCD) health outcomes. A few comprehensive reviews of dietary metrics in relation to several MCH or NCD outcomes were completed more than a decade ago, and no review has focused on both MCH and NCD outcomes.

44We searched PubMed for articles published between Oct 11, 2000, and April 17, 2020, using the terms “diet quality”, “diet metric”, or “diet score” with “malnutrition”, “undernutrition”, “maternal health”, “child health”, “stunting”, “wasting”, “mortality”, “non-communicable disease”, “chronic disease”, “cardiovascular disease”, “diabetes”, “cancer”, or “obesity” with “meta-analysis”, “systematic review”, or “narrative review”. We used search terms in English but did not apply any language restrictions. We screened articles by title and abstract to identify full-text reports that were relevant to the study aims. We also screened citation lists for these full-text reports to identify other relevant articles. Articles were considered relevant if they reported the relation between dietary metrics and malnutrition. Several previous reviews qualitatively summarised the evidence for dietary metrics in relation to a single health outcome, and other reviews focused on several maternal and child health (MCH) or non-communicable disease (NCD) health outcomes. A few comprehensive reviews of dietary metrics in relation to several MCH or NCD outcomes were completed more than a decade ago, and no review has focused on both MCH and NCD outcomes.