Design and study population

The current study used dietary information from Ensanut MC 2016, a nationally representative probabilistic multistage stratified cluster survey of 29 795 individuals from 9 474 randomly selected households, constructed to make distinctions between urban and rural areas, as well as among four geographic regions.^10,11 This survey was carried out between May and October 2016, including information for 24HR in a random subsample of ∼15% of participants (n=4 341).

For this study, we included individuals with complete diet information, from which pregnant or lactating women (n=85) and preschool children (n=535) were excluded, as were implausible values for main macronutrients and micronutrients (n=79; first dietary recall n=75; secondary dietary recall n=4). Our final sample included 3 646 individuals, of whom 1 081 were schoolchildren (aged 5-11 years; mean=8.51 years), 1 228 were teenagers (aged 12-19 years; mean=15.63 years), and 1 337 were adults (aged ≥20 years; mean=47.09 years).¹⁰ Around 8% of participants had a second dietary recall (3.98% of schoolchildren; 10.99% of teenagers; and 8.45% of adults) to estimate within-person variance estimates to adjust nutrient intake distributions¹²(figure 1).

Figure 1 Analytical sample of the 2016 Ensanut MC 24-hour dietary recall (24HR) 

The Research, Ethics and Biosafety Committees of the Mexican National Institute of Public Health (Instituto Nacional de Salud Pública, INSP) approved the protocol for Ensanut MC 2016. Written informed consent was obtained from all study participants aged ≥18 years, as well as from the father, mother, or guardian of participants under 18 years. Informed assent was also collected for children and adolescents between 5 and 17 years.

Diet information

A standardized 24HR was applied for data collection through an automated software^* of an adapted version of the original method designed by the United States Department of Agriculture.¹³ This method was selected for capturing more accurate information from the interviewees through five iterative steps that complement each other in order to improve the memory in relation to food intake and thus, reduce underreporting. In Mexico, it is identified as the Automated Software utilized for the multiple-pass 24HR (MP-24HR). This method was complemented with information regarding the type of foods consumed (food preparations or food-items), place and activity while consuming the reported foods, cooking method applied, waste fractions, and whether food recipes were prepared at home or bought in a food stand, supermarket, restaurant or others. The information above is briefly described in figure 2; more details about the adapted method are described elsewhere.¹⁴ Previously trained and standardized personnel applied the MP-24HR.

Data editing and processing

Information obtained from the MP-24HR was edited and processed in four main phases that make up a series of iterative processes:

Phase 1. Processing-review and correction of context information. Coding errors regarding mealtime and detailed food records were reviewed and corrected (steps 3 and 4 of figure 2). For step 3 of the multiple pass 24HR, the coding of mealtime, place and activity was verified. For the 24HR’s step 4, regarding food details, information on reported foods and recipes was reviewed and corrected (such as: name coding, type of food, raw and net weights, portions prepared, leftovers, cooking methods, etc.) and when new foods were identified (foods not previously compiled in the food composition database), these were included, and their nutritional information was obtained and added to the Mexican Food Database (BAM, Spanish acronym). Coding errors in food items and preparations were also corrected (e.g. the quick list included “fried fish” with the type code 3, “standardized recipe”, which is when the person didn’t know in detail the preparation of the recipe; however, for some reason, the interviewer selected “raw fish” in the list catalogue). Those cases which were clearly errors were corrected by selecting the correct option from the list, in this case the “fried fish, standardized recipe” option. Inconsistencies for food details were also corrected when possible.

Figure 2 Multiple-pass 24-hour dietary recall description (24HR), adapted for Mexico 

Phase 2. Processing - Estimation of grams of food consumed. Grams of food consumed were estimated based on a database with 14 types of household measures (such as: small piece, medium piece, large piece, teaspoon, measuring cup, etc.) constructed from standardized measurements obtained in a field workshop with kitchen scales. Implausible values for each food-item in terms of grams consumed were identified according to a reference based on its specific distribution by age and sex. It is worth noting that this is an iterative process, so when implausible values were identified, information for those individuals was reviewed again in phase 1. Afterwards, the data were processed in four stages:

1) Processing - Imputation of grams of salt and condiments used in recipes. Since salt and seasoning measurement in the field is complex, our approach was through a workshop with trained nutritionists who measured the amounts of salt or condiments used in recipes frequently reported by the Mexican population. This information was used for imputation in recipes in general (in both disaggregated and standardized recipes).

2) Processing - Estimation of net grams consumed. For every food item reported in its raw weight (when applicable), the edible portion factor was multiplied to obtain net weights. For beverages, we applied density factors when necessary to convert their volume in milliliters into grams, using a food-matching approach based on the Food and Agriculture Organization (FAO) and the international network of food data systems (INFOODS) density database, which was carried out by researchers from the INSP for the BAM.¹⁵

3) Cleaning information of grams of food consumed: Outliers in terms of grams of food consumed were defined and identified as >4 standard deviations above the average of each food-item, by age and sex. For outlier imputations, a random value between the 95-99^thpercentile was used (as in previous Ensanut), considering the characteristics of the food/recipe, mealtime, age group and sex,¹⁶ in order to keep those values in the upper tail of the distribution, within plausible ranges. For missing data, imputations were made using the mean value according to the characteristics mentioned above. It is worth noting that, in all cases, ≤1% of observations were imputed.

Phase 3. Processing - Energy and nutrient estimations. Nutrient and energy estimations were carried out using the BAM-a compilation of the frequently consumed foods in the country-in its 18.1.1 version, made up of a total of 1978 foods, recipes, and beverages, being an indirect compilation from several food composition databases^3,4,5,6 made for the Center for Nutrition and Health Research of the INSP.^* Additionally, in order to estimate potential nutrient losses by different cooking methods, we used an indirect compilation of NRFs made for BAM. For the NRF compilation, we mainly used the Unites Estates Departament of Agriculture (USDA), Bergstrom’s and Bognar’s publications.^17,18,19 In those cases where information was not available for reported food-items, we used the average value of the NRF compilation from the European Food Information Resource (EuroFIR), which is group-specific.²⁰ All the NRFs were applied at the ingredient level when the nutrient composition in the cooked version was not available, according to international recommendations made by EuroFIR.²¹ The NRFs used to adjust estimations of nutrient intake were for the following cooking methods: boiled, roasted, fried, baked, and stewed.

Phase 4. Energy and nutrient data cleaning. Regarding data editing, we prioritized nutrients identified as relevant based on the evidence in the area of nutrition in terms of their contents of dietary energy, fiber, carbohydrates, proteins, lipids, vitamin A, vitamin C, vitamin D, vitamin B12, iron, zinc and calcium. We added the energy, fiber, and nutrient content from foods/ingredients at individual level. For intake revision, we used as a reference for adequacy the equations developed by the Institute of Medicine (IOM), according to body mass index (BMI, normal weight, overweight or obese) and age group.²² The physical activity factor we considered for estimating energy requirements was light physical activity in school-aged children, adolescents and adult men; and in adult women, sedentary activity was considered (based on previous evidence of national surveys).¹⁴ For micronutrients, individual intake was divided by the estimated average requirement (EAR) of the Dietary Reference Intakes (DRI).^23,24,25,26 For macronutrients, we used the ratio between individual intake contribution to dietary energy and the upper value of the acceptable contribution range (55% for carbohydrates and 30% for lipids). Finally, outliers were identified based on the ratio of each nutrient’s intake to the average requirements for specific population groups. For both macro and micronutrients, based on each nutrient’s ratio distribution, <-3 and >+3 standard deviation values were marked as outliers or implausible values by sex and age group, and these outliers were removed from analysis.

Afterwards, a review of energy and nutrient estimates was carried out. When implausible values were identified, the case was revised again in phases 1 and 2 to correct the information. For the above, we reviewed the complete 24HR for selected cases, in order to search for possible errors in the estimations of grams of food consumed. This was done by comparing it to the energy and nutrient intake distributions according to the individual’s age and sex. If errors in terms of grams of food consumed were identified, we proceeded to impute the food-specific mean value of intake by age and sex as described above.

On the other hand, in order to describe the effect of the use of NRFs, we selected the following micronutrients previously known to have varying degrees of loss due to cooking methods: vitamin A (µg of retinol equivalents), vitamin C (mg), vitamin D (international units), thiamin (mg), riboflavin (mg), niacin (mg), dietary folate equivalents (µ DFE), vitamin B12 (µg), zinc (mg), calcium (mg) and magnesium (mg). Afterwards, using the methodology previously described, we estimated intakes of these micronutrients with and without NRFs. Values above 1.5 times the 99th percentile were imputed with a random value between the 95th and 99th percentiles, according to the individual’s age and sex, as in the last Ensanut 2012.¹⁶

All the processes were performed using the Stata V.13.1 Software.^*

Study variables

We estimated the usual dietary intake estimations with and without the application of NRF for vitamin A (µg of retinol equivalents), vitamin C (mg), vitamin D (international units), thiamin (mg), riboflavin (mg), niacin (mg), dietary folate equivalents (µ DFE), vitamin B12 (µg), zinc (mg), calcium (mg), and magnesium (mg).

The prevalence of micronutrient inadequacy was calculated using as a reference the EARs cutoff from the DRIs for the nutrients mentioned above.^23,24,25,26 We estimated the proportion of the population that was below the EAR for each specific vitamin and mineral by age group and sex, with and without NRFs.

Frequency of vitamin and mineral loss through cooking methods by food group

Average vitamin and mineral loss through cooking methods was calculated in order to provide information regarding the most affected food groups in the Ensanut MC 2016.^* Food groups were obtained by selecting food items that represented 70% of those reported in Ensanut MC 2016. Selected foods were classified into 12 categories: vegetables, tubers, corn products, fish and seafood, cereals and breads, legumes, dairy products, fats and oils, fruits, cold meats, breakfast cereals, and meats and eggs. For each food group, frequency of reported cooking methods was obtained. Mean differences between estimated nutrient intake with and without NRFs were calculated and transformed into relative percentages for each nutrient.

Statistical analysis

The main characteristics of the study population were estimated considering the survey design, using the SVY module for complex samples of the Stata software V.13.1.

The usual intake of selected micronutrients was estimated by age groups, using the Iowa State University method,²⁷ through the PC-Side program version 1.0 (Iowa State University).^‡

This method estimates the distribution of the usual intake of nutrients by removing the effect of intra-individual variability and leaving the variability of daily intake between individuals. Likewise, the inadequate prevalence of selected micronutrients was estimated with PC-Side using the age and sex specific EAR cutoffs of the IOM in two scenarios by age group: with and without NRFs.^22,23,24,25 Because our age groups of interest differed from those of the EARs, we scaled the individual intake, as it has been previously described in terms of age and sex, in order to use a single cutoff within each group.²⁸ The scaling process consisted in selecting the group with the largest sample size within age and sex and scaling the individual intake for the rest of the groups. For example, in the adult group aged >20 years, there were two cutoffs for vitamin C according to sex, of which the group of females had a larger sample size; therefore, we selected the female cutoff and scaled the individual intake by males in order to be able to apply the same cutoff. The differences between groups (with and without NRFs) regarding mean usual intakes were analyzed with paired T-test’s; this was done using the Individual Usual Intakes obtained from PC-Side in Stata. Also, the differences in prevalence of nutrient inadequacy were calculated using the McNemar’s test. P values <0.05 were considered statistically significant. All the statistical analyses were performed using the Stata software V.13.1.

Usual intake

Significant differences were identified with respect to the mean usual intake of vitamins and minerals, based on the application of the NRFs (table I). For all age groups, differences in estimated usual intakes with and without NRFs were small for some nutrients, such as minerals, with differences ranging from 0.5% for calcium up to 2.9% for magnesium. However, differences in the usual intake of vitamins were relevant, ranging from 1.55% for vitamin D to 14% for folate and 34.4% for B12 (table I). By age group, differences in the usual intake of vitamins and minerals were similar to those mentioned above.

Prevalence of nutrient inadequacies

Two out of three school children had a lower calcium intake than recommended, and approximately one out of three had a lower intake of vitamin A. The prevalence of inadequacy (with NRFs) for other nutrients in school children ranged from 4.4 (thiamin) to 71.2% (calcium) and 100% (vitamin D). In adolescents, nutrients with the highest prevalence of inadequacy (with NRFs) -excluding vitamin D- were calcium, vitamin A and magnesium, followed by vitamin B12, folate, vitamin C, riboflavin, and thiamin. Adults had the highest prevalence of inadequacy (with NRFs) for most nutrients. All age groups had a vitamin D inadequacy of 100% (table II).

Regarding the prevalence of nutrient inadequacies with and without NRFs by age groups, we identified that, as with the usual intake estimations, the differences were small for some nutrients such as calcium, zinc, vitamin A and D. For vitamin C and magnesium, the differences were approximately 3-6%. Thiamin, niacin, vitamin B12 and folate were the most affected nutrients, with differences of up to 55.5% in the prevalence of inadequacy. Globally, the prevalence of nutrient inadequacies was underestimated when NRFs were not applied, with all differences being statistically significant except for vitamin D (table II).

Percent nutrient loss per food group

The intake of vitamins and minerals depends in part on the amount of food consumed, but it also depends on the way the food item is cooked. Figure 3A shows the proportion of cooking methods reported in the preparation of some of the main food items of the Ensanut MC 2016: 73% of vegetables, 96% of tubers, 94.7% of meats and eggs, 89% of fish and shellfish, 84% of legumes, 58% of sausages, and 43% of breakfast cereals were consumed in a cooked form (boiled, roasted, fried or baked). Figure 3B shows the average percentage of nutrient loss for some food groups, identifying vegetables, meats and eggs, fish and seafood and legumes as the most affected groups.

Figure 3A Proportion of cooking methods used for the main food groups by the Mexican population 

Figure 3B Percentage of mean vitamin and mineral loss in main food groups by cooking method among the Mexican population