Table 1

Data sources, modelling approaches and validation methods used to estimate adult intake levels of key foods worldwide, by region, country, age and sex in 1990 and 2010

Dietary factorData sourcesStatistical methods used for pooling and modelling data from diverse global sources
Individual-level surveysNational FAO food balance sheets†Modelling approachSurvey-specific covariates‡Validity
Regions covered (of 21)*Years coveredSurveys, countries (of 187) and global population covered
FruitsAC, AE, APH, AS, ASE, AUS, CAR, EURC, EURE, EURW, LAC, LAS, LAT, NA, NAM, OC, SSC, SSE, SSS, SSW1980–2009A total of 204 surveys, of which 137 had individual-level data (113 of those had age-specific and sex-specific estimates) and 67 were household-level surveys, were collected from 109 countries and represented 87% of the world's adult populationCalculated fruit intake (derived from FAO data on fruits) consumed per capita per day in 187 countries in each year from 1990 to 2010DisMod3, a Bayesian hierarchical method, was used to pool data from multiple sources and model missing data using informative time-varying covariates, borrowing information across geographical region and time period while also incorporating uncertainty due to measurement error and model specification Models were fit using a randomised Markov chain Monte Carlo algorithm based on the Adaptive Metropolis step functionMetric (primary vs secondary metric), representativeness (nationally representative vs subnational), diet assessment method (diet recalls/records or FFQ vs household availability/budget survey)
Country-specific: lag-distributed national per capita income (inflation and purchasing power parity adjusted), FAO factor variables 1, 2, 4
Models were assessed for convergence of Markov chain Monte Carlo iterations. DisMod3 was validated using goodness-of-fit tests and out-of-sample predictive validity tests, in which 10% of data were held out of the model. Qualitative evaluation for foods was conducted by comparing the estimated foods with known high-quality data and assessing their face validity through contact with country experts
Vegetables§AC, AE, APH, AS, ASE, AUS, CAR, EURC, EURE, EURW, LAC, LAS, LAT, NA, NAM, OC, SSC, SSE, SSS, SSW1980–2009A total of 204 surveys, of which 137 had individual-level data (113 of those had age-specific and sex-specific estimates) and 67 were household-level surveys, were collected from 109 countries and represented 87% of the world's adult populationCalculated vegetable intake (derived from FAO data on vegetables) consumed per capita per day in 187 countries in each year from 1990 to 2010Metric (primary vs secondary metric), representativeness (nationally representative vs subnational), diet assessment method (diet recalls/records or FFQ vs household availability/budget survey)
Legumes§AE, APH, AS, ASE, AUS, CAR, EURC, EURE, EURW, LAC, LAS, LAT, NA, NAM, SSE, SSS, SSWA total of 138 surveys, of which 72 had individual-level data (62 of those had age- and sex-specific estimates) and 66 were household-level surveys, were collected from 64 countries and represented 81% of the world's adult populationCalculated legume intake (derived from FAO data on legumes) consumed per capita per day in 187 countries in each year from 1990 to 2010Metric (primary vs secondary metric), representativeness (nationally representative vs subnational), diet assessment method (diet recalls/records or FFQ vs household availability/budget survey)
Nuts and seedsAE, APH, AS, ASE, AUS, CAR, EURC, EURE, EURW, LAC, LAS, LAT, NA, NAM, SSE, SSS, SSW1980–2009A total of 126 surveys, of which 61 had individual-level data (54 of those had age-specific and sex-specific estimates) and 65 were household-level surveys, were collected from 53 countries and represented 74% of the world's adult populationCalculated nut and seed intake (derived from FAO data on nuts/seeds) consumed per capita per day in 187 countries in each year from 1990 to 2010Representativeness (nationally representative vs subnational), diet assessment method (diet recalls/records or FFQ vs household availability/budget survey)
Whole grainsAE, APH, ASE, AUS, CAR, EURC, EURW, LAS, LAT, NA, NAM, SSS1987–2009A total of 35 surveys, all of which had individual-level data with age-specific and sex-specific estimates, were collected from 25 countries and represented 41% of the world's adult populationCalculated whole grain intake (derived from FAO data on barley, rye and other cereals) consumed per capita per day in 187 countries in each year from 1990 to 2010Representativeness (nationally representative vs subnational)

SeafoodAE, APH, AS, ASE, AUS, CAR, EURC, EURE, EURW, LAC, LAT, NA, NAM, SSE, SSS, SSW1980–2009A total of 115 surveys, of which 48 had individual-level data (40 of those had age-specific and sex-specific estimates) and 67 were household-level surveys, were collected from 52 countries and represented 54% of the world's adult populationCalculated PUFA n-3 intake (derived from FAO data on PUFA n-3) consumed per capita per day in 187 countries in each year from 1990 to 2010Representativeness (nationally representative vs subnational)

Red meats, unprocessedAC, AE, APH, AS, ASE, AUS, CAR, EURC, EURE, EURW, LAC, LAS, LAT, NA, NAM, SSE, SSS, SSW1980–2009A total of 154 surveys, of which 87 had individual-level data (69 of those had age-specific and sex-specific estimates) and 67 were household-level surveys, were collected from 74 countries and represented 83% of the world's adult populationCalculated red meat intake (derived from FAO data on red meats) consumed per capita per day in 187 countries in each year from 1990 to 2010Metric (primary vs secondary metric), representativeness (nationally representative vs subnational), diet assessment method (diet recalls/records or FFQ vs household availability/budget survey)
Processed meatsAE, APH, ASE, AUS, CAR, EURC, EURE, EURW, LAC, LAS, LAT, NA, NAM, SSS 1980–2009A total of 127 surveys, of which 60 had individual-level data (58 of those had age-specific and sex-specific estimates) and 67 were household-level surveys, were collected from 54 countries and represented 54% of the world's adult populationCalculated red meat intake (derived from FAO data on red meats), pig meat intake (derived from FAO data on pig meats) and animal fat intake (derived from FAO data on animal fats) consumed per capita per day in 187 countries in each year from 1990 to 2010Metric (primary vs secondary metric), representativeness (nationally representative vs subnational), diet assessment method (diet recalls/records or FFQ vs household availability/budget survey)

  • *Based on 21 GBD study regions including APH, Asia Pacific, high income; AC, Asia, Central; AE, Asia, East; AS, Asia, South; ASE, Asia, Southeast; AUS, Australasia; CAR, Caribbean; EURC, Europe, Central; EURE, Europe, Eastern; EURW, Europe, Western; LAA, Latin America, Andean; LAC, Latin America, Central; LAS, Latin America, Southern; LAT, Latin America, Tropical; NAM, North Africa/Middle East; NA, North America, high income; OC, Oceania; SSC, Sub-Saharan Africa, Central; SSE, Sub-Saharan Africa, East; SSS, Sub-Saharan Africa, Southern; and SSW, Sub-Saharan Africa, West.

  • †The FAO food balance sheets capture a country's net annual food availability based on reported local production, imports and exports. We calculated 14 composite diet composition variables from FAO food balance sheets specific to the overall 2010 NutriCoDE list of dietary factors of interest (with the exception of sodium). The 14 standardised FAO nutrients or food groups represented the majority of food available for human consumption in 187 countries in each year from 1990 to 2010.

  • ‡Both survey-specific and national-level covariates were incorporated in the model. Primary inputs were the survey-level intake data and the diet composition variables from FAO food balance sheets, including all available country-specific, time-specific, age-specific and sex-specific consumption levels (mean, distribution), data on the numbers of participants in each strata; and survey-level indicator covariates (sampling representativeness, dietary assessment method, type of dietary metric). Surveys carried out between 1980 and 1997 were used to inform the 1990 time period, and surveys carried out between 1997 and 2010 and the 2010 period. Time-varying country-level covariates (available in all years, including 2010) further informed the estimates, including LDI44 (inflation and purchasing power parity adjusted); and national dietary patterns characterised by scores on four factors from a principal component analysis of the 14 FAO diet composition variables.18 Taking into account that many of the food covariates are very collinear (eg, red meat, pig meat and animal fats), and that consuming more of one food necessitates consuming less of other types, we used dimension reduction through principal component analysis to reduce the 14 standardised FAO nutrients or food groups into four factor variables, which were included in the model to improve country-level predictions: factor 1 included red meats, animal fats and pig meats; factor 2 included n-3 polyunsaturated fats, n-6 polyunsaturated fats, whole grains, nuts and vegetables; factor 3 included fruits, legumes and nuts; and factor 4 included sugars, stimulants and saturated fats (from oils). The FAO covariates were used in the per cent natural logarithm form, that is, the per cent of total kilocalorie that is comprised of a particular food. A space-time smoothing procedure was used to generate a full time series of intake estimates. Income and education were used as covariates in the space-time model to improve predictions in instances of missing data. For education, the age standardised mean number of years of education for ages 25+ by sex as a continuous variable was used.45 For income, the estimated and normalised lag-distributed income based on the international dollar as a continuous variable was used.34 For countries that had split or merged during the time series (1990–2010), we split/merged these countries into constituent countries using a growth rate method to generate as close to a full time series as possible for all countries. For model description (DisMod3, eAppendix) and model fits (eFigure 7) see data supplements.

  • §Vegetable and legume intake were estimated separately using Dismod3, and subsequently summed, given that studies evaluating disease risk typically summed or used overlapping categories of vegetable and legumes (eg, green beans included as vegetables).

  • FAO, Food and Agriculture Organization of the United Nations; FFQ, food frequency questionnaire; GBD, Global Burden of Diseases, Injuries, and Risk Factors; LDL, lag-distributed national per capita income.