We used data from the Mexico City Representative Diabetes Survey 2015 (MCRDS, 2015). This survey has a probabilistic multistage stratified cluster sampling design. Participants were selected through cluster sampling, using basic geostatistical areas (AGEB in Spanish acronym) as the primary sampling unit. From each AGEB, systematic sampling was conducted to six houses within six blocks. Two adults aged 20–69 years were systematically selected in each house. Trained personnel conducted face-to-face interviews using validated questionnaires between May and June 2015. The response rate for the original study was 71.4%. The MCRDS 2015 collected information on demographics, lifestyles, diet habits, and chronic diseases of 1,313 adults from Mexico City (22). We excluded adults with invalid or incomplete information on diet and those with potentially implausible intakes by food group and total energy > 3 S.D. (n = 299). Hence, the final analyses included 1,023 individuals.

We first described the sociodemographic and lifestyle characteristics of the study sample. Second, we used Pearson's chi-squared test to examine the bivariate relationship of the density of small grocery and convenience stores and sociodemographic variables with the MxAHEI. Then we performed multivariable Poisson models to test the association of small grocery and convenience store densities with MxAHEI (overall and by component). The models were adjusted for sex, age, educational attainment, employment status, car possession, socioeconomic status, urban marginalization degree, and density of other types of stores. We considered the MxAHEI as the outcome variable using the high score as the reference category. We used the variance inflation factor (VIF) to check for multicollinearity. VIF values ranged from 1.22 to 3.26 suggesting no linear relationship among the predictors (36). The significance level was established at alpha 0.05. However, as recommended, we also described and discussed results that, even though not statistically significant, provide a broader picture of the findings (37). Statistical analyses were performed using Stata 14 (Stata corp LLC, College Station, TX, USA) (38). We used survey commands to account for survey design and weighting to generate nationally representative results.

Our findings of an inverse association not statistically significant between a higher density of small grocery stores and diet quality score can be explained by several reasons. First, the results seem to be consistent with what these establishments offer. Bridle-Fitzpatrick et al., studied the food environment in a city in a northern state of Mexico (Mazatlán, Sinaloa); they found that small grocery stores sold a small selection of fresh fruit and vegetables and large quantities of sugar-sweetened beverages (SSBs) and packaged snacks (20). Second, although the number of convenience stores has increased, the small grocery stores (tiendas de abarrotes) continue to be the most predominant food retailer in Mexico; there are 67 small retail stores per chain convenience store (12). Small grocery stores have great cultural importance, affordable prices, are located nearby homes, and offer credit to the customers (15), which might promote the intake of unhealthy and energy-dense foods at a lower cost. Moreover, it has been documented that small grocery stores account for an important proportion of the sales of transnational food companies in Mexico, including SSBs, baked goods, snacks, beer, and dairy (39).

Although not statistically significant, we observed an inverse association between small grocery store density and legumes intake. Even though small grocery stores offer dry and canned legumes, we do not exclude the possibility that people purchase their legumes from other stores like supermarkets (40). However, we do not have information on purchases of legumes by type of food store. Moreover, the estimations could be a reflection of the low intake of legumes in Mexico. Aburto et al. (41), found that legumes had the lowest energy contribution (3.8%) of the total energy intake among the Mexican population, especially in urban localities (42).

We did not observe an association between the density of small grocery stores and other diet components (vegetables, whole fruits, whole grains, seeds, red and processed meats, long-chain (n-3) fats (EPA+DHA), SSBs, trans-fat, alcohol). One possible explanation of this finding is that these foods are found in fresh food stores in Mexico (e.g., fruit and vegetable stores, butchers, poultry shops, fishmongers, and other specialized shops) (17, 20), and their purchases are likely independent of what is acquired in small grocery stores.

Although not statistically significant, we found an inverse association between convenience stores and the overall diet quality. In the same line, Lind et al., found that higher availability of convenience stores within 500 meters network buffers was associated with an unhealthy diet among 16 years and older Danish (32). Also, a longitudinal study in young adults in the United States found an association between a higher density of convenience stores and a low score quality of the diet for participants living within a 3-kilometer distance along the street network (7). One potential explanation for our not statistically significant results could be that the number of convenience stores in Mexico is still small relative to small grocery stores (12, 21). However, the relative proportion of convenience stores continues to rise (12, 15, 17, 21). Therefore, monitoring the association between convenience stores and diet intake will be needed. Also, the density of convenience stores is more likely to be related to specific food components than the overall diet.

Specifically, we found a not statistically significant inverse association between convenience store density and the vegetable component score. This finding is consistent with the study by Zhang et al. (43), who observed that a higher density of convenience stores was associated with an increased likelihood of infrequent vegetable consumption in the Chinese population (41). Another study performed in an urban multiethnic population in the United States found that the presence of a convenience store in the neighborhood was negatively associated with vegetable intake among Latinos (7). Although convenience stores represent a small proportion of the total stores (1.3% in 2016), it is the fastest-growing store type in Mexico (12). The availability of convenience stores could negatively influence the consumption of healthy food products because convenience stores typically carry fewer whole grains, legumes, fruit, and vegetables than other retail food outlets. A recent study in Mexico showed that the density of convenience stores was associated with higher purchases of ultra-processed food and SSBs, probably linked to their low prices and large availability in these stores (15).

We also found that a high and medium density of convenience stores was statistically associated with a higher intake of polyunsaturated fat, excluding long-chain fat. This finding can be explained by the fact that vegetable oils, highly accessible in convenience stores, are a major source of no-long-chain fat in Mexico (43). Although polyunsaturated fats can have a protective role against cardiovascular disease, this effect could be detrimental when used to fried the food (44, 45). Future analysis will be needed to determine the type of polyunsaturated fats purchased by the retail food store and how they are consumed.

Our study has several limitations. First, this was a cross-sectional study; therefore, it is not possible to determine the causal relationship between grocery and convenience store density and diet quality. However, the findings highlight the potential of the food environment to influence the diet of adults in Mexico City. Future longitudinal studies will be needed to confirm our results. Second, the data analyzed are not recent. However, the association between the density of food outlets and diet quality is not expected to change significantly over a short time. Third, we do not rule out the possibility of selection bias since the missingness in diet information might be related to the diet itself. Fourth, our findings could be only extrapolated to areas with populations with a socioeconomic status distribution similar to our study sample, which was different from those observed in the excluded sample (Supplementary Table 4). Moreover, although the original study might be representative of adults from Mexico City, this might not be true by geographic region. Given the nature of our analyses, it is likely that more individuals were clustered in some buffers than others. Therefore, our no statistically significant results might be partially explained by the low variability in the exposure and outcome due to the geographic distribution of our study sample. The dichotomization of the outcome variable can also explain the low variability in the MxAHEI scores across buffers and, therefore, the non-significant results. Nevertheless, we considered the results informative given the directions of the estimations. Fifth, we estimated store density to quantify the availability of food stores by using geographic information methods only (46). However, we did not conduct any store audit or checklist to assess the food availability in the food outlets of interest in more detail (47). Sixth, we did not consider other types of food stores, such as informal or mobile food outlets (as the DENUE data do not include information about informal food vendors), which could impact the diet, as shown in other studies (17, 18). Moreover, as mentioned before, we focused on small grocery and convenience stores because of their proximity to people's homes and their supply of products with immediate access (12, 17, 20). Seventh, although Euclidean buffer zones have limitations concerning street network buffers, both metrics have similar correlation results in terms of the level of pseudo individual density (at the same distance), and the correlation increased so when accounting for larger neighborhoods (400 m r_s = 0.667 and 800 m r_s = 0.667 p < 0.001) (28). Eighth, we focused on the food environment around homes in a buffer of 500 meters; however, likely individuals do not shop (or not always) at small grocery or convenience stores near home. Nineth, one of the main limitations of the SFFQ is it might underestimate the food intake; however, it is not a limitation in our study since the MxAHEI considers the total energy intake. Finally, we recognized that the MxAHEI scores calculated might not reflect the variability in dietary intake by seasonality since the survey was conducted between May and June. Future studies considering food information from different seasons would provide further insights into the potential association between the food environment and diet quality. For future studies, it is important to consider combining multiple evaluation techniques, including individual factors and social contexts of Mexican populations, for example, promotion, price, availability of homemade food/beverages, and acceptability, which have been linked with unhealthy diets (19). Therefore, counting more of these variables would help better understand the mechanisms for which the food environment potentially affects the food intake.

This study's strength includes using an economic census database to estimate the density of convenience stores rather than self-report of stores. Also, this is one of the first studies that explore the association between the density of two types of food retails (small grocery and convenience stores) and the diet quality among Mexican adults. This study contributes to the literature on the community food environment.