Comparison of Dietary Intake Before vs After Taxes on Sugar-Sweetened Beverages and Nonessential Energy-Dense Foods in Mexico, 2012 to 2018

This cross-sectional study compares dietary intake before and after taxes on sugar-sweetened beverages (SSBs) and nonessential energy-dense foods in Mexico.


Introduction
Taxation is an effective policy to address the health and economic burden caused by the excessive intake of unhealthy foods and beverages. [1][2][3] Worldwide, the number of countries or local jurisdictions implementing diet-related taxes is increasing; by 2022, more than 45 countries and jurisdictions had implemented taxes on sugar-sweetened beverages (SSB). 4 Mexico was one of the first countries to implement diet-related taxes. Since 2014, the Mexican government implemented a 1 peso-per-liter tax (approximately 10% price increase) on SSBs and an 8% ad valorem tax on nonessential energydense food. 5 The SSB tax comprises all industrialized nonalcoholic beverages (including concentrates and powders) with added sugar. The food tax includes nonessential foods with an energy density of 275kcal/100 g or more, such as chips and snacks, candies and sweets, chocolate, puddings, peanut and hazelnut butter, ice cream, ice pops, and cereal-based products with added sugar.
Previous studies in Mexico found that after the tax, prices increased by the amount of the tax, particularly in urban areas. 6 Studies evaluating up to 3 years after tax implementation reported a 7.6% decrease in SSB purchases, while taxed food decreased 6.0%. [7][8][9][10][11] Reductions were larger in lower socioeconomic status (SES) groups, in urban areas, and in households that used to have larger purchases of taxed products. 9,10 In a cohort study, the probability of becoming a nonconsumer of SSBs increased by 5%. 12 As in Mexico, studies from other countries concur that taxes are effective to achieve a reduction in the purchase of taxed products. A meta-analysis including 16 tax policies from all over the world found an average reduction of 15% in sales of SSBs after implementation of taxes. 4 Most tax evaluation studies, including the ones from Mexico, have used purchase or sales data. 4 Commercial consumer panels (eg, Nielsen or Kantar) have important advantages in evaluating beverage and food taxes because household purchases are collected longitudinally, frequently (ie, biweekly), and in great detail (ie, at the brand level). 13 Thus, evaluations with consumer panels are more solid and can give more conclusive results than other data sources. However, an important limitation of these data is that information is aggregated at the household level so changes by individual characteristics such as age or sex cannot be identified. In addition, purchase data do not capture total dietary intake; hence, it misses the ability to evaluate total intake.

Dietary surveys conducted with 24-hour dietary recalls (24HRs) or Food Frequency
Questionnaires (FFQ) can fill purchase data gaps by capturing total intake and at the individual level.
Few previous efforts to monitor taxes have been conducted with a representative dietary survey. 14,15 Yet, a challenge with dietary surveys is their large degree of measurement error, limiting the ability to capture absolute intake and attribute changes to small taxes. 16 An additional weakness is that the periodicity of data collection is not frequent, and in the case of Mexico, surveys have been conducted in different seasons. Despite this, looking at the relative contribution of taxed items over time in dietary surveys is helpful to gain new insights. Our aim was to estimate the adjusted contribution of taxed and untaxed beverages and foods, and the nutritional dietary profile (added sugars, saturated fats, and energy density) in the National Health and Nutrition Surveys (ENSANUT for its acronym in

Study Population
We analyzed dietary information from probabilistic and cross-sectional surveys representative of the Informed consent was obtained from participants aged 18 years or older and from the parent or guardian of participants aged younger than 18 years. We adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cross-sectional studies where applicable. The survey protocol for each ENSANUT was approved by the ethics committee of the National Institute of Public Health. The description of the methods has been published elsewhere. [17][18][19] For this analysis, we excluded children aged less than 1 year, breastfed children aged less than 1 year, and pregnant or lactating women. Additionally, we excluded individuals with extreme energy intakes according to the survey's established protocols (see eFigure 1 and eFigure2 in Supplement 1). [20][21][22]

Dietary Data
Dietary data from the ENSANUT 2012 and 2016 were collected using 2 instruments in different subsamples, a 24HR and a semiquantitative FFQ; in 2018, only FFQ was collected. The sample size for each instrument as well the collection or not of 24HR depended on the budget available. The subsamples for each 24HR or FFQ are nationally representative. Both 24HR and FFQ were collected in person by trained interviewers. The 24HR used the multiple-pass method. 20 Interviewers recorded the types and amounts of all food items consumed in the previous day. The FFQ inquires about the frequency, portion size, and the number of portions consumed of 140 food items during the previous 7 days. 21 Data from 24HR and FFQ were categorized into comparable food groups. For foods, complex mixed dishes were disaggregated into individual ingredients, and beverages were kept as preparations. Following tax law definitions, items were grouped into 4 main groups: taxed beverages, untaxed beverages, taxed foods, and untaxed foods. Additional subgroups were also analyzed. Energy and nutrient intakes were estimated in all surveys with the most recent foodcomposition tables compiled by the National Institute of Public Health. 23 Added sugars were estimated as described previously by Sánchez-Pimienta et al. 24 Energy density was estimated by dividing the energy content of foods consumed by their weight (kcal/100 g). To estimate the weight, water content in mixed dishes (eg, soups, rice, pasta) was imputed according to a database of standard recipes.

Covariates
Sociodemographic and macroeconomic changes can be associated with dietary behaviors; thus, we accounted for these. We included sex, age, area of residence (urban [Ն2500 inhabitants] or rural

Statistical Analysis
We estimated proportions or means for sociodemographic and macroeconomic variables and evaluated their changes over time with a χ 2 or t test accordingly. To address the lack of comparable temporality between surveys, which primarily affects the intake of beverage during warmer months, we analyzed foods and beverages separately and focused on the percentage contribution instead of absolute intake. We estimated the contribution of taxed and untaxed beverages to the total volume of beverages (% vol), and the contribution of taxed and untaxed foods to total energy from foods (% kcal). The rationale for looking at volume instead of energy was that some beverages, such as water, are noncaloric and could not be captured in the % kcal. This analysis was conducted among all and by subpopulation groups (age, sex, urban or rural area, and SES).
We estimated the content of added sugar in all beverages (kcal/100 mL) and all foods (% kcal from foods) consumed, the content of saturated fat (% kcal from foods), and energy density (kcal/100 g from foods) in all foods consumed. Then we split the nutritional content and computed the share of taxed and untaxed items (ie, kcal of added sugar consumed in taxed beverages per 100 mL of all beverages consumed).

Results
In total, 17 239 participants were analyzed from 2012, 18 974 from 2016, and 30 027 from 2018. The main characteristics of the study population are shown in Table 1   In the case of foods (Table 2), according to the 24HR, the energy contribution to total foods of taxed foods decreased 3.0 (95% CI, −4.2 to −1.8) percentage points. Sweet bread from bakeries did    24HR and FFQ, the contribution of whole grains, processed meats, other animal sources, and sugar and desserts increased. The remaining food groups did not change (24HR), refined grains and legumes and nuts decreased, and all the remaining food groups increased in at least 1 survey (FFQ).
According to the 24HR, the total beverage added sugar, food added sugar, food saturated fat, and food energy density decreased after the tax ( Table 3)

Discussion
We evaluated the contribution of taxed and untaxed beverages and foods from 2012 (pretax) to 2016 and 2018 (posttax) with data from 24HR and FFQ. Despite small size taxes in Mexico and large measurement error of dietary data, we found an overall decrease in the contribution of taxed beverages to the total volume of beverages consumed, in the contribution of taxed foods to the total energy of foods consumed, and in content of unhealthful nutritional components, particularly added sugars, in total beverages and total foods.
By subpopulations and food subgroups, our findings were aligned with previous evaluations of the Mexican tax. Studies on commercial consumer panels and/or household expenditure surveys found that among those with higher SES or living in rural areas, the response was lower for SSB or null for energy-dense food 7,8,10 ; or that taxed carbonated beverages had a lower reduction and taxed sweet bread from bakeries did not decrease. 7,10 In our study, we found no change in higher SES, rural areas, carbonated beverages, and sweet bread from bakeries with the 24HR; but with the FFQ, we found statistical differences across all subpopulations and food subgroups. This suggests that the 24HR might be more precise and detailed to capture variations by subpopulations and food subgroups. Yet, in comparison with the FFQ that captures intake of last the 7 days, the 24HR has a larger random measurement error that affects statistical significance. Worldwide, the lower change among those with higher SES has been inconsistent 3,26 ; for instance in Chile 27 and the UK, 28 the opposite was found. The finding of a larger decrease in noncarbonated vs carbonated was also observed in Chile, 27 but the opposite was found in Barbados 29 and Seattle. 30 The lower change in rural areas observed in Mexico could be associated with the industry strategy of not completely passing the tax to consumers in rural areas. 31 Finally, the finding that sweet bread from bakeries (eg, unpackaged) did not decrease could be due to a tax exemption given to small bakeries.
An important contribution of our study was the possibility to look at changes by age group and sex. According to the 24HR, taxed beverages did not change among children aged less than 5 years, adults, and men, but according to the FFQ, taxed beverages decreased in all age groups and sexes. It is likely that adults and men were less responsive because proportionally they consume fewer noncarbonated beverages (see eTable in Supplement 1), the beverages that decreased the most after the tax. Moreover, for both taxed beverages and taxed foods, we found that the age group with the largest intake before the tax had the largest decline after the tax (adolescents for beverages and children aged <5 years for foods). This is consistent with previous evaluations of the taxes in Mexico in which households with the highest baseline intake of taxed items were the ones that decreased the most after the tax. 9,32 Another contribution was the ability of our study to look at changes in a wider range of untaxed categories. For instance, previous studies found an increase in bottled water, 7 yet it was not possible to identify if this was used to prepare homemade sugary drinks. Homemade aguas frescas are heavily consumed in Mexico as it is unusual to have main meals with plain water. We found that along with an increase in the contribution of water to total beverage volume, there was also an increase in homemade sugary drinks (only statistically significant with FFQ). Yet, the overall content of added sugar in beverages decreased. In the case of untaxed foods, we found that food subgroups that increased their contribution according to both 24HR and FFQ were whole grains, processed meats, other animal sources, and sugars and desserts. Processed meats and sugars and desserts are unhealthy foods that are potentially substitutes of taxed energy-dense foods. Despite this, the overall intake of added sugars from foods decreased throughout, and the overall content of saturated and did not include changes in absolute amounts (ie, mL or kcal). Dietary surveys are prone to a large degree of measurement error, making it difficult to assess changes in energy intake in absolute terms. 16,35 For instance, it is difficult to know how much of the 73 kcal per day that decreased from 2012 to 2016 in the 24HR is due to measurement error. In a previous analysis, we evaluated the intraindividual change among a subsample of the ENSANUT 2012 with 2 repeated 24HRs and found that energy intake was not fully substituted, as on the days in which participants did not consume taxed beverages or foods, total energy intake decreased. 36 Thus, a decrease in total energy intake associated with the taxes is a possibility.

Limitations and Strengths
National dietary surveys provide useful information for tax evaluation, yet they are limited in important ways. The degree of measurement error is large, as explained before. Also, before the tax, there were no repeated surveys in short and frequent intervals, so we did not account for previous ongoing trends, and there are substantial lags between our studied periods and the implementation

Conclusions
Our analysis of national dietary nutrition surveys, despite its limitations, yielded similar conclusions to prior evaluations of taxes in Mexico using different data sources, contributing to the consistency of results. As a new contribution, we documented some increases in the contribution of unhealthy untaxed items after the tax, but these did not offset the decreases in the overall content of unhealthful nutritional components, particularly added sugars, in total beverages and foods. Our results also suggest that changes in the contribution of taxed beverages was heterogenous by age and sex. Altogether, our findings suggest that despite their small size, taxes in Mexico have contributed to improve the quality of dietary intake, and thus, these should be augmented to or beyond levels recommended by the World Health Organization. Our results reiterate that taxes remain a key part of the strategy for obesity and chronic disease prevention.