Effect of Mobile Food Environments on Fast Food Visits

Poor diets contribute significantly to global morbidity and mortality. The food environment, particularly the prevalence of fast food outlets (FFOs) and convenience stores (often termed 'food swamps'), is hypothesized to influence dietary choices. Previous research, however, has largely focused on static food environments around homes, yielding mixed results. This study addresses this gap by leveraging large-scale mobility data to examine the relationship between exposure to mobile food environments and the frequency of fast-food visits. The researchers hypothesize that exposure to FFOs throughout daily routines, rather than just near home, significantly impacts food choices. Understanding this relationship is critical for developing more effective interventions aimed at improving dietary health. The study's focus on mobile food environments is particularly relevant given that a substantial portion of food consumption occurs away from home, a factor often neglected in previous studies. The authors also note the limited impact of past interventions that focused solely on improving neighborhood food environments, highlighting the need for a more comprehensive understanding of the influence of food environments across daily mobility patterns. By utilizing large-scale mobility data, the study aims to establish a more nuanced understanding of the causal link between food environment exposure and fast-food consumption, providing a foundation for designing more effective public health interventions. The selection of FFO visits as the primary outcome is justified by the established link between fast food consumption and diet-related diseases, and because previous work using similar data has demonstrated a correlation between FFO visits and self-reported fast-food intake, obesity, and type 2 diabetes.

Existing research on the relationship between food environments and dietary choices has predominantly focused on static, neighborhood-level environments, specifically examining food deserts (areas with limited access to healthy foods) and food swamps (areas saturated with unhealthy food options). While some studies have linked exposure to food swamps with unhealthy eating and diet-related diseases, the findings are often mixed and frequently non-significant. The majority of these studies employ cross-sectional designs, hindering the establishment of causal relationships. Policy interventions, such as investments in healthy food retail in underserved areas or fast food bans, have largely demonstrated minimal impact on diet quality or disease outcomes. This lack of success highlights the limitations of focusing solely on residential food environments, especially considering the growing trend of food acquisition and consumption occurring away from home. Studies utilizing tracking technologies to map individual food acquisition behaviors are emerging, but these have been limited in scale and duration, failing to capture habitual patterns or achieve sufficient statistical power to draw robust conclusions. The current study addresses these limitations by combining large-scale mobility data with longitudinal observations of individual behavior, allowing for a comprehensive investigation of the effect of mobile food environments on dietary choices.

This study uses a large, anonymized, privacy-preserving dataset encompassing the mobility patterns of 1.86 million individuals across 11 U.S. metropolitan areas over a 6-month period (2016-2017). Data was obtained from Spectus, a location intelligence company. The dataset includes information on users' locations throughout the day, allowing for the examination of food environments beyond residential neighborhoods. Food outlet visits were identified by analyzing stays near points of interest (POIs) obtained from Foursquare, categorized into FFOs (fast food outlets) and FOs (all retail food outlets). The researchers calculated the ratio of FFOs to all FOs within a 1km radius around each location (φ(x)), representing the local food environment. They also calculated a time-weighted measure of mobile food environment exposure (φm) for each user. To analyze the association between mobile food environments and FFO visits, they defined a binary outcome variable (yit) indicating whether a user visited an FFO at a given time. A logistic regression model was used to assess the impact of the food environment (context) on the decision to visit an FFO, controlling for individual preferences and daily variations. The model included a fixed effect for each user and day to account for individual differences and daily patterns. The study focuses on lunch visits due to the concentration of food-related activities during that time window. To address concerns about confounding variables, the researchers further employed a natural experiment using individuals who changed their habitual pre-lunch locations during the study. This allowed them to study the effect of changing mobile food environment on FFO visits. A changepoint analysis was performed to find users whose habitual context shifted within the observation period, and a Bayesian structural time-series model was used to compare the post-change behavior of those who shifted to contexts with different levels of FFO exposure against those who remained in similar contexts. The study also included an analysis of visits to Department of Motor Vehicles (DMV) locations, considering the DMV's food environment as the context for subsequent food outlet choices, leveraging the quasi-random placement of DMV offices across the cities as a natural experiment. Finally, the researchers used the model's results to simulate the impact of different policy interventions aimed at altering the ratio of FFOs to other food outlets in specific areas. Four strategies were compared: Food Swamp intervention (targeting areas with the highest ratio of FFOs), Low Food Access intervention (targeting food deserts with high FFO ratios), Food Hotspots intervention (targeting areas with the highest number of FFO visits), and Behavior-Environment intervention (targeting areas where the intervention would have the greatest impact based on the combined effects of the food environment and individual preferences), as predicted by the model. Latent topic analysis was used to characterize the types of POIs prevalent in areas targeted by the Behavior-Environment intervention.

The study's key findings include: 1. Most fast food visits (6.8%) occur outside of a user's home census tract, highlighting the importance of mobile food environments. 2. The median distance from home to a fast-food visit was 6.74 km, significantly farther than the distance to other food outlets. 3. A positive correlation exists between average daily exposure to FFOs in mobile environments (φm) and the overall proportion of FFO visits (μ). 4. Logistic regression analysis revealed a significant association between exposure to high ratios of FFOs in the context preceding a lunch visit and the odds of visiting an FFO. A 10% increase in the proportion of FFOs in the context increased the odds of an FFO visit by 20%. 5. This effect was consistent across different times of day, days of the week, income levels, and metro areas. 6. The analysis of individuals changing their habitual contexts demonstrated that the effect of the food environment was not solely due to visits to new places; habitual exposure to FFO-rich environments led to an increase in FFO visits, while exposure to FFO-poor environments led to a decrease in visits. 7. An analysis of visits to DMVs, considered as a natural experiment with more randomly distributed contexts, also showed a significant association between the surrounding food environment and FFO visits. 8. Simulations of policy interventions demonstrated that a Behavior-Environment intervention, which targets areas with the most significant predicted impact based on both environmental context and individual preferences, would be far more efficient (1.93x to 3.85x) than interventions solely based on home neighborhoods or food swamps. This intervention, applied to a small percentage of FFOs (0.22%), could avert 0.56% of FFO visits. 9. The most effective interventions targeted areas with high concentrations of POIs related to Malls, Industry/Factory, Airport, and Office, suggesting that proximity to workplaces and shopping areas strongly influence food choices.

The study's findings strongly support the hypothesis that mobile food environments exert a substantial influence on fast food consumption, surpassing the impact of home neighborhood environments. The large-scale, longitudinal design, coupled with semi-causal analyses, provides compelling evidence for a causal relationship. The consistent effect observed across diverse demographics underscores the pervasive nature of this influence. The success of the Behavior-Environment intervention strategy, outperforming traditional approaches, highlights the importance of considering both environmental context and individual behaviors when designing interventions. This strategy's focus on areas where food choices are most susceptible to environmental influence suggests a shift from solely focusing on residential areas to potentially more effective interventions in commercial and work environments. The observed heterogeneity in intervention effectiveness across different cities suggests that future research should consider city-specific factors and their influence on food choice behavior. The strong correlation between exposure to mobile food environments and visits to other types of restaurants further validates the generalizability of the study's findings. This has significant implications for public health policy, suggesting a more nuanced, behaviorally informed approach to intervention design.

This study demonstrates a significant effect of mobile food environments on fast food consumption, surpassing the influence of home neighborhoods. The Behavior-Environment intervention strategy, integrating both environmental context and individual preferences, shows a potential for substantially greater efficiency than traditional approaches. This suggests focusing future interventions on areas like workplaces and shopping centers, where food decisions are significantly influenced by environmental factors. Future research should explore further refinements to intervention strategies, considering individual preferences and other factors like food delivery services and changing post-pandemic behaviors.

The study acknowledges limitations such as the reliance on anonymized data, potential misclassification of FFO visits (especially drive-thrus), and the use of proxy measures for food intake. While FFO visits strongly correlate with fast food intake, the study cannot directly measure actual consumption. The study's timeframe (2016-2017) might not fully reflect post-pandemic changes in mobility and food consumption patterns. Further, the focus on FFOs may not fully represent the impact of mobile environments on overall dietary choices; future research should examine a broader range of food outlets.