Consumption Levels of Dietary Fats and Oils in 1990 and 2010
This work was performed by the Nutrition and Chronic Diseases Expert Group (NutriCoDE) as part of the 2010 Global Burden of Diseases, Injuries, and Risk Factors (GBD) Study. Our methods for identification, access, and selection of dietary risk factors and data have been reported. Because we used de-identified national datasets, this research was reviewed by the Harvard School of Public Health institutional human subjects committee and deemed to be exempt from human subjects research requirements. To generate valid, comparable estimates of consumption of fats and oils around the world, we used consistent methods across regions, countries, age and sex subgroups, and time to:
We reviewed the evidence for aetiological effects on chronic diseases including coronary heart disease, stroke, type 2 diabetes, and cancers, based on multiple criteria for assessing causality of diet-disease relationships. We required convincing or probable evidence for effects on clinical events (such as myocardial infarction) rather than simply on physiological risk factors (such as blood cholesterol concentration). The detailed results for our assessments of aetiological effects of dietary fats and oils have been reported. We identified aetiological effects of omega 6 polyunsaturated fat as a replacement for saturated fat, seafood omega 3 fats, and trans fats on coronary heart disease. We did not identify convincing or probable evidence for aetiological effects of these fats on stroke, diabetes, or cancers; or of total fat, monounsaturated fat, plant omega 3 fats, or dietary cholesterol (evaluated mainly through egg consumption) on coronary heart disease, stroke, type 2 diabetes, or cancers.
We performed systematic searches for individual level dietary surveys in all countries. Surveys with evidence for selection bias or measurement bias were excluded. Using standardised criteria and methods, we searched multiple online databases from March 2008 to September 2010 without date or language restrictions. From these searches, we identified comparably few appropriate published data sources. Thus, from March 2008 to July 2012, we also used extensive personal communications with researchers and government authorities throughout the world, including authors of published nutrition studies and nutrition authorities in a given country, inviting them to be corresponding members of the NutriCoDE group (Fig 1). For countries lacking identified national or subnational individual-level surveys by these methods, we searched for other potential data sources, including individual level surveys from large cohorts, the WHO Global Infobase, the WHO Stepwise Approach To Surveillance (STEPS) database, and budget survey data at the household level. Given our aim to evaluate chronic diseases, we focused on data from adults (aged ≥20 years). The results of our search strategy by dietary factor, time, and region have been reported. A total of 266 surveys in adults representing 113 of 187 countries and 82% of the global population were identified (Fig 1).
(Enlarge Image)
Figure 1.
Flow diagram of systematic search for nationally representative surveys of food and nutrient intake
Data retrieval followed the 2010 Global Burden of Diseases study’s comparative risk assessment framework, collecting quantitative data on consumption in 16 age- and sex-specific subgroups across 21 world regions (see eTable 1 of Data Supplement) and two time periods (1990 and 2010). Most published or publically available dietary data were limited or not in the relevant format. For 173 surveys, 99 corresponding members provided original raw data to us or re-analysed their raw data according to our specifications, providing age and sex stratified dietary results in specified metrics and units using a standardised electronic format (Appendix 1 of Data Supplement). Optimal and alternative metrics and units were defined for each dietary factor, with optimal units matching those of studies used to evaluate relationships with disease risk as well as major dietary guidelines. Based on these criteria, dietary factors were evaluated as percentage energy (saturated fat, omega 6 polyunsaturated fat, trans fat) or as mg/day standardised using the residual method to 2000 kcal/day (dietary cholesterol, seafood omega 3 fat, plant omega 3 fat). The surveys providing data on dietary fats and oils are listed in eTable 2 of Data Supplement.
For each survey, we extracted data on survey characteristics, dietary metrics, units, and mean and distribution (such as standard deviation) of consumption of each dietary fat and oil, by age and sex (eTable 2). Data were double checked for extraction errors and assessed for plausibility. We assessed survey quality by evaluating evidence for selection bias, sample representativeness, response rate, and validity of diet assessment method. Measurement comparability across surveys was maximised by using a standardised data analysis approach that (1) accounted for sampling strategies within the survey by including sampling weights (if available), (2) used the average of all days of dietary assessment to quantify mean intakes, (3) used a corrected population standard deviation to account for within person variation versus between person variation, (4) used standardised dietary metrics and units of measure across surveys, and (5) adjusted for total energy to reduce measurement error and account for differences in body size, metabolic efficiency, and physical activity.
Our systematic approaches to survey identification and data retrieval identified gaps in data for certain countries, certain dietary fats or oils across countries, or time periods. Furthermore, even using the systematic data retrieval and standardisation methods described above, identified surveys and measures were not always comparable—for example, varying in representativeness, urban or rural coverage, age groups, dietary instruments, or dietary metrics. To address missing data, incomparability, and related effects on uncertainty of dietary estimates, we developed an age integrating Bayesian hierarchical imputation model (Appendix 2 of Data Supplement). This model estimated the mean consumption level and its statistical uncertainty for each age, sex, country, and year stratum. For each dietary factor, primary inputs were the survey-level quantitative data, including country-, time-, age-, and sex-specific consumption levels (mean, distribution); data on the numbers of subjects in each strata; survey level indicator covariates for sampling representativeness, dietary assessment method, and type of dietary metric; and country, region (21 Global Burden of Diseases regions), and super-region (7 Global Burden of Diseases groupings of regions) random effects. Additional country-level, time-varying (year-specific) covariates, which were available in all years including 2010, included lagged distributed income and food disappearance data derived and standardised from United Nations Food and Agricultural Organization food disappearance balance sheets, including 17 nutrients or food groups and four factors derived from principal components analysis of these 17 variables. The final model covariates for each dietary risk factor are presented in Table 1 . The final national, regional, and global estimates were calculated as population-weighted averages of the corresponding age- and sex-specific strata. Using these methods, we quantified the consumption levels of saturated fat, omega 6 polyunsaturated fat, trans fat, dietary cholesterol, seafood omega 3 fat, and plant omega 3 fat, and among men and women in 187 countries in 1990 and 2010.
The model included additional offset and variance components to account for differences between primary and secondary dietary metrics, national and subnational surveys, and individual and household dietary data—in each case allowing greater influence of the former. Model validity across different iterations was evaluated using cross validation, randomly omitting 10% of the raw data and comparing the imputed intakes with the original raw data. Sources of uncertainty were identified and incorporated, including from missing country data, sampling uncertainty of original data sources, and additional uncertainty associated with suboptimal metrics, subnational samples, or household level surveys. Using simulation (Monte Carlo) analyses, we drew 1000 times from the posterior distribution of each exposure for each age, sex, country, and year stratum; computed the mean exposure from the 1000 draws; and the 95% uncertainty intervals as the 2.5th and 97.5th centiles of the 1000 draws. Absolute and relative differences in exposure between 1990 and 2010 were calculated at the draw level to account for the full spectrum of uncertainty. We used Spearman correlations to evaluate interrelations between specific dietary fats and oils.
To place observed consumption levels in context and allow separate assessment of impact on disease, for each dietary factor we characterised the optimal, yet feasible, consumption levels ( Table 2 ). This was based on both the mean observed consumption associated with lower disease risk in meta-analyses of clinical endpoints and the mean national consumption levels observed in at least two or three countries around the world. As another criterion, we also considered whether such characterised optimal consumption levels were broadly consistent with current major dietary guidelines.
Methods
Study Design
This work was performed by the Nutrition and Chronic Diseases Expert Group (NutriCoDE) as part of the 2010 Global Burden of Diseases, Injuries, and Risk Factors (GBD) Study. Our methods for identification, access, and selection of dietary risk factors and data have been reported. Because we used de-identified national datasets, this research was reviewed by the Harvard School of Public Health institutional human subjects committee and deemed to be exempt from human subjects research requirements. To generate valid, comparable estimates of consumption of fats and oils around the world, we used consistent methods across regions, countries, age and sex subgroups, and time to:
Identify specific dietary fats and oils with evidence for largest public health impact, based on strength of evidence for aetiological effects on coronary heart disease, stroke, type 2 diabetes, or cancers
Systematically search for nationally representative data sources from around the world on individual level dietary consumption of these fats and oils, including by age and sex
Retrieve data, including assessment of quality and representativeness, maximisation of measurement comparability and consistency, and ascertainment of missing data and uncertainty
Estimate consumption levels by region, country, age, sex, and time, accounting for missing data, incomparability of measured values, and sampling and modelling uncertainty
Characterise optimal consumption levels of each dietary fat and oil, based on observed intakes associated with lowest disease risk and observed mean national consumption levels globally, to place the observed intakes in context and enable quantification of relevant attributable disease burdens.
Identification of Dietary Fats and Oils With Largest Public Health Impact
We reviewed the evidence for aetiological effects on chronic diseases including coronary heart disease, stroke, type 2 diabetes, and cancers, based on multiple criteria for assessing causality of diet-disease relationships. We required convincing or probable evidence for effects on clinical events (such as myocardial infarction) rather than simply on physiological risk factors (such as blood cholesterol concentration). The detailed results for our assessments of aetiological effects of dietary fats and oils have been reported. We identified aetiological effects of omega 6 polyunsaturated fat as a replacement for saturated fat, seafood omega 3 fats, and trans fats on coronary heart disease. We did not identify convincing or probable evidence for aetiological effects of these fats on stroke, diabetes, or cancers; or of total fat, monounsaturated fat, plant omega 3 fats, or dietary cholesterol (evaluated mainly through egg consumption) on coronary heart disease, stroke, type 2 diabetes, or cancers.
Systematic Searches for National Dietary Data
We performed systematic searches for individual level dietary surveys in all countries. Surveys with evidence for selection bias or measurement bias were excluded. Using standardised criteria and methods, we searched multiple online databases from March 2008 to September 2010 without date or language restrictions. From these searches, we identified comparably few appropriate published data sources. Thus, from March 2008 to July 2012, we also used extensive personal communications with researchers and government authorities throughout the world, including authors of published nutrition studies and nutrition authorities in a given country, inviting them to be corresponding members of the NutriCoDE group (Fig 1). For countries lacking identified national or subnational individual-level surveys by these methods, we searched for other potential data sources, including individual level surveys from large cohorts, the WHO Global Infobase, the WHO Stepwise Approach To Surveillance (STEPS) database, and budget survey data at the household level. Given our aim to evaluate chronic diseases, we focused on data from adults (aged ≥20 years). The results of our search strategy by dietary factor, time, and region have been reported. A total of 266 surveys in adults representing 113 of 187 countries and 82% of the global population were identified (Fig 1).
(Enlarge Image)
Figure 1.
Flow diagram of systematic search for nationally representative surveys of food and nutrient intake
Data Retrieval and Standardisation
Data retrieval followed the 2010 Global Burden of Diseases study’s comparative risk assessment framework, collecting quantitative data on consumption in 16 age- and sex-specific subgroups across 21 world regions (see eTable 1 of Data Supplement) and two time periods (1990 and 2010). Most published or publically available dietary data were limited or not in the relevant format. For 173 surveys, 99 corresponding members provided original raw data to us or re-analysed their raw data according to our specifications, providing age and sex stratified dietary results in specified metrics and units using a standardised electronic format (Appendix 1 of Data Supplement). Optimal and alternative metrics and units were defined for each dietary factor, with optimal units matching those of studies used to evaluate relationships with disease risk as well as major dietary guidelines. Based on these criteria, dietary factors were evaluated as percentage energy (saturated fat, omega 6 polyunsaturated fat, trans fat) or as mg/day standardised using the residual method to 2000 kcal/day (dietary cholesterol, seafood omega 3 fat, plant omega 3 fat). The surveys providing data on dietary fats and oils are listed in eTable 2 of Data Supplement.
For each survey, we extracted data on survey characteristics, dietary metrics, units, and mean and distribution (such as standard deviation) of consumption of each dietary fat and oil, by age and sex (eTable 2). Data were double checked for extraction errors and assessed for plausibility. We assessed survey quality by evaluating evidence for selection bias, sample representativeness, response rate, and validity of diet assessment method. Measurement comparability across surveys was maximised by using a standardised data analysis approach that (1) accounted for sampling strategies within the survey by including sampling weights (if available), (2) used the average of all days of dietary assessment to quantify mean intakes, (3) used a corrected population standard deviation to account for within person variation versus between person variation, (4) used standardised dietary metrics and units of measure across surveys, and (5) adjusted for total energy to reduce measurement error and account for differences in body size, metabolic efficiency, and physical activity.
Quantification of Global, Regional, and National Distributions
Our systematic approaches to survey identification and data retrieval identified gaps in data for certain countries, certain dietary fats or oils across countries, or time periods. Furthermore, even using the systematic data retrieval and standardisation methods described above, identified surveys and measures were not always comparable—for example, varying in representativeness, urban or rural coverage, age groups, dietary instruments, or dietary metrics. To address missing data, incomparability, and related effects on uncertainty of dietary estimates, we developed an age integrating Bayesian hierarchical imputation model (Appendix 2 of Data Supplement). This model estimated the mean consumption level and its statistical uncertainty for each age, sex, country, and year stratum. For each dietary factor, primary inputs were the survey-level quantitative data, including country-, time-, age-, and sex-specific consumption levels (mean, distribution); data on the numbers of subjects in each strata; survey level indicator covariates for sampling representativeness, dietary assessment method, and type of dietary metric; and country, region (21 Global Burden of Diseases regions), and super-region (7 Global Burden of Diseases groupings of regions) random effects. Additional country-level, time-varying (year-specific) covariates, which were available in all years including 2010, included lagged distributed income and food disappearance data derived and standardised from United Nations Food and Agricultural Organization food disappearance balance sheets, including 17 nutrients or food groups and four factors derived from principal components analysis of these 17 variables. The final model covariates for each dietary risk factor are presented in Table 1 . The final national, regional, and global estimates were calculated as population-weighted averages of the corresponding age- and sex-specific strata. Using these methods, we quantified the consumption levels of saturated fat, omega 6 polyunsaturated fat, trans fat, dietary cholesterol, seafood omega 3 fat, and plant omega 3 fat, and among men and women in 187 countries in 1990 and 2010.
The model included additional offset and variance components to account for differences between primary and secondary dietary metrics, national and subnational surveys, and individual and household dietary data—in each case allowing greater influence of the former. Model validity across different iterations was evaluated using cross validation, randomly omitting 10% of the raw data and comparing the imputed intakes with the original raw data. Sources of uncertainty were identified and incorporated, including from missing country data, sampling uncertainty of original data sources, and additional uncertainty associated with suboptimal metrics, subnational samples, or household level surveys. Using simulation (Monte Carlo) analyses, we drew 1000 times from the posterior distribution of each exposure for each age, sex, country, and year stratum; computed the mean exposure from the 1000 draws; and the 95% uncertainty intervals as the 2.5th and 97.5th centiles of the 1000 draws. Absolute and relative differences in exposure between 1990 and 2010 were calculated at the draw level to account for the full spectrum of uncertainty. We used Spearman correlations to evaluate interrelations between specific dietary fats and oils.
Characterisation of Optimal Consumption
To place observed consumption levels in context and allow separate assessment of impact on disease, for each dietary factor we characterised the optimal, yet feasible, consumption levels ( Table 2 ). This was based on both the mean observed consumption associated with lower disease risk in meta-analyses of clinical endpoints and the mean national consumption levels observed in at least two or three countries around the world. As another criterion, we also considered whether such characterised optimal consumption levels were broadly consistent with current major dietary guidelines.
SHARE
