The 2012 and 2016 Mexican National Health and Nutrition Surveys (ENSANUT 2012 and 2016, for its acronym in Spanish) are probabilistic population-based surveys with a multistage, stratified sampling design, representative at the national and regional levels and for rural and urban areas [23, 24]. Briefly, ENSANUT 2012 was conducted between October 2011 and May 2012, and information was collected from 50,528 households, for a household response rate of 87% [24]. ENSANUT 2016 was conducted between May and September 2016, and information was collected from 9,479 households, for a household response rate of 77.9% [23].
Detailed dietary information was obtained for random subsamples in both surveys using 24-h recalls (n = 10,885 for ENSANUT 2012 and n = 4,341 for ENSANUT 2016). For the present analysis, we excluded children younger than one year of age (n = 420), children older than one year of age who were being breastfed (n = 109), pregnant and lactating women (n = 245), and individuals with implausible intake (n = 183). Plausible intake was defined as energy intake between ± 3 standard deviations of the ratio of energy intake and energy requirement, estimated using the Institute of Medicine equations for body weight maintenance. A detailed description of the methods used to identify implausible intake is described elsewhere [25, 26]. Furthermore, we also excluded individuals who consumed less than 50 kcal (n = 3) or more than 6,000 kcal (n = 24) per day to account for slight variations in dietary collection and cleaning methods between the surveys. Only those with intakes greater than 6,000 kcal were recorded for 2012, and those with intakes less than 50 kcal were recorded for 2016. Thus, the study sample included 14,242 individuals, composed of preschool children (1–4 y), school-aged children (5–11 y), adolescent (12–19 y) and adult (≥ 20 y) men, and nonpregnant, nonlactating adolescent and adult women with complete socioeconomic information (n = 10,062 for ENSANUT 2012 and n = 4,180 for ENSANUT 2016).
Dietary assessmentThe 24-h recall was collected by trained interviewers between Monday and Sunday using an automated 5-step multiple-pass method [27]. Participants ≥ 15 years old were asked to report all foods and beverages consumed the previous day. For children and adolescents younger than 15 years, the person responsible for food preparation in the household was asked to provide information regarding their intake, with children or adolescents complementing the interview by reporting food eaten away from home. Interviewers assisted participants to avoid omissions and were provided with a food scale, measuring cups and serving spoons to help with the estimation of portion sizes. Tortilla and other typical foods of specific regions were weighted to capture variability from different regions of the country. Intake could be reported as individual foods or beverages (e.g., chips or water) or mixed dishes/beverages (e.g., stew or coffee with sugar). Mixed dishes and beverages could be either disaggregated to their ingredients if the participant knew the amounts used in their preparation or recorded as a standard preparation if eaten away from home or if the amounts used were unknown. For the purposes of this analysis, standard preparations were disaggregated to their ingredient level. For instance, a beef stew was disaggregated to beef, potato, carrot, tomato, etc. Energy and nutrient content was estimated using the 2016 food composition table compiled by the National Institute of Public Health (Nutrient Database, Compilation of the Mexican National Institute of Public Health, unpublished material, 2016).
EAT-Lancet diet quality indexWe used an EAT-Lancet index, developed by Stubbendorff et al., to assess adherence to the EAT-Lancet reference diet, with some minor adaptations [28]. A total of 14 components were included in the index construction, including components that should be consumed in adequate amounts or “emphasized intake” and components that should be consumed in moderation or “limited intake”. The emphasized components were 1) vegetables, 2) fruits, 3) whole grains, 4) legumes, 5) seafood, 6) nuts, and 7) unsaturated oils. The limited components were 1) beef and lamb, 2) pork, 3) poultry, 4) eggs, 5) dairy, 6) potatoes and 7) added sugar. Scores between 3 and 0 were assigned for each component, with 3 indicating compliance and 0 indicating noncompliance. Hence, the score ranged from 0 to 42 points. Whole grains were defined as grains with ≥ 10 g of total fiber per 100 g of carbohydrates and refined otherwise [29].
Stubbendorff's EAT-Lancet index considered intake in grams as recommended by the EAT-Lancet reference diet. For example, compliance for vegetables consisted of 300 g, which would result in 3 points. The amount of grams per food group from the EAT-Lancet reference diet was only estimated for an intake of 2,500 kcal/day, representing the diet of an average adult male. However, since we included all age and sex groups, we recalculated the amounts of grams from Stubbendorff's EAT-Lancet index to a percent of contribution to total daily energy intake to account for lower energy requirements of children and women. For instance, 300 g of vegetables would represent a 3.2% contribution.
Furthermore, the EAT-Lancet reference diet recommends an added sugars intake from 0 to 31 g (< 5% kcal). The EAT-Lancet index by Stubbendorff et al. assigns 3 points for less than 31 g of added sugars, 2 points from 31 to 62 g, 1 point from 62 to 124 g, and 0 points for more than 124 g. We considered these criteria to be too permissive, given that 1 point (62 to 124 g) would represent a contribution between 10 and 20%; therefore, we modified the cutoff points for added sugars as follows: < 5%, 3 points; 5 to 7.5%, 2 points; 7.5 to 10%, 1 point; and > 10%, 0 points (Supplemental Table 1).
Food and beverage pricesWe retrieved monthly nominal prices of food and beverages from the National System of Statistical and Geographical Information (INEGI, for its acronym in Spanish), which are used as the imput to calculate the Consumer Price Index (CPI). The CPI measures average weighted price changes of a basket of goods and services that are commonly purchased by urban households [30]. INEGI collected prices from 46 cities distributed across the 32 Mexican states between 2011 and 2016. These cities have a population of > 20,000 inhabitants, including the ten most populated urban zones in the country, representing ~ 66% of the Mexican population. In each city, prices were obtained from a sample of 16,000 sales points (stores, markets, and other vendors), excluding food services. Food and beverage prices are collected byweekly and reported on a monthly basis throughout the year from the different points of sale, and monthly averages are reported on the INEGI website.
Given that INEGI provides data with greater detail about most food and beverage items (e.g., brand and package size) than does the ENSANUT food composition table, many INEGI items were linked to a single food code from ENSANUT. The exception was a few categories from ENSANUT, which were collected at the brand level; thus, these items were matched with a specific item from INEGI (e.g., ready-to-eat cereals). The INEGI data reported prices for unprocessed foods per kilogram of gross weight; thus, we adjusted for refusal to match the net intake in grams. Moreover, some items are reported in other units, such as pieces, a handful, and liters. For those items, we estimated the weight using ENSANUT’s portion and weights table and the density of liquids. To bring food and beverage nominal prices to their equivalent in real prices, we used rthe CPI provided by INEGI whose reference month is July 2018.
Matching of prices with ENSANUT itemsWe started with 2,133,141 unique prices from 2011 to 2016. For water, we only took prices from 20-L jugs, excluding prices from water bottles (n = 20,342), to consider water intake within households [31]. We also identified items which prices per year had a coefficient of variation above 80, and reviewed them individually. A total of 976 unique prices were excluded for possible errors. Thus, the remaining 2,111,823 unique prices were used to estimate average prices.
To account for seasonal and regional variability, we estimated average prices at four levels: 1) quarterly averages at the regional level according to the four regions used in ENSANUT (north, center, south, and Mexico City), 2) annual averages at the regional level, 3) national annual averages, and 4) national total average from 2011 to 2016. Any average that had fewer than five observations was excluded. Of all the foods and beverages reported in ENSANUT, 85.8% were matched with quarterly averages at the regional level, 3.5% were matched with annual averages at the regional level, 6.4% with national annual averages, and 3.7% with national total averages. The remaining 0.6% of the reported items in the ENSANUT lacked a direct match, so these items were instead matched to a group or subgroup average.
Statistical analysisWe estimated proportions and 95% confidence intervals (95% CIs) to describe the sample by sociodemographic characteristics, including age group, sex, SES, region, place of residence, and survey. The SES index was determined through principal component analysis, considering household characteristics and assets. Subsequently, households were classified into tertiles. Region was divided into North, Central, Mexico City, and South. Place of residence was categorized as rural if the location had fewer than 2,500 inhabitants and urban otherwise. To adjust the cost by total energy intake, we estimated the cost residuals centered at the mean by regressing the diet cost on energy intake and then calculated the residual of the regression, followed by adding the mean of the diet cost to the calculated residual, hereafter referred to as diet cost residuals. This way the cost comparison is not influenced by differences in total energy intake that can be observed between age, sex, or socioeconomic groups.
We dichotomized the diet cost residuals at the median to define a low-cost and a high-cost diet. Then, we used multivariate linear regression models to compare the EAT-Lancet index and its 14 components between individuals consuming a low-cost diet and those consuming a high-cost diet. Similarly, we compared the energy intake and cost per 100 kcal among consumers of the index components by diet cost categories. The previous models were adjusted by age group, self-reported sex, SES tertile, region, place of residence, survey, and total energy intake (except for the energy intake comparison). In the tables, we present the predicted means of the EAT-Lancet index, energy intake and cost per 100, which were estimated using the Stata’s margins command.
Finally, we assessed the association between diet cost residuals and the EAT-Lancet index using a pooled multivariate linear regression model. To further assess whether the relationship differed by SES, we repeated the same model but included an interaction term between the EAT-Lancet index and SES tertiles. All analyses were conducted in Stata version 14.1 (College Station, TX: StataCorp LLC) and were weighted to be nationally representative and to account for the complex survey design.
留言 (0)