TL;DR
This study demonstrates that aggregated smartphone mobility data can effectively predict fast food consumption and diet-related diseases like obesity and diabetes at a neighborhood level. By analyzing millions of location pings, researchers found that patterns of visiting fast food outlets are stronger indicators of health outcomes than traditional self-reported dietary habits.
What Problem Does This Paper Address?
Understanding how food environments influence health is crucial, but research has often focused on static, home-based environments. This overlooks the dynamic nature of daily life, where people encounter various food options as they move. Traditional methods for measuring diet, like self-reporting, are prone to bias and recall errors. This paper addresses the gap by exploring whether large-scale, passively collected human mobility data can provide a more objective and comprehensive view of exposure to food environments and its link to diet-related diseases.
Methods
The core of this study lies in linking two distinct datasets:
- Health Survey Data: The 2011 Los Angeles County Health Survey (LACHS) provided individual-level data on self-reported fast food intake, obesity, and diabetes, along with sociodemographic information.
- Mobility Data: A large dataset of anonymized smartphone geolocations (Spectus) from over 243,000 users in Los Angeles County (LAC) was used. This data was processed to identify "stays" at locations and then aggregated to the neighborhood level (LAC Neighborhoods - LACN).
Key methodological steps included:
- Defining Food Outlets: A Foursquare Points of Interest (POI) database was used to identify food outlets, with fast food (FF) outlets specifically defined as limited-service restaurants serving predominantly ultra-processed, low-nutrient, energy-dense foods.
- Mobility Data Aggregation: Individual user visits to FF outlets were aggregated to the LACN level. This was necessary due to privacy constraints and to link with the LACHS data. Two primary neighborhood-level mobility variables were created:
- FF visits/time: Percentage of observed daily periods with visits to FF outlets.
- FF visits/food: Percentage of total food outlet visits that were to FF outlets.
- Linking and Analysis: The aggregated mobility data was linked to the LACHS respondents based on their residential LACN. Statistical analyses, including multinomial and binary logistic regressions, were used to assess the associations between mobility-derived FF visit measures, self-reported FF intake, obesity, and diabetes. Akaike Information Criterion (AIC) and Akaike weights were used to compare model fit.
Datasets
- 2011 Los Angeles County Health Survey (LACHS): A population-based survey providing individual-level health and demographic data. The analytic sample consisted of 5,447 adults.
- Spectus Mobility Dataset: Anonymized geolocations from 243,644 smartphone users in Los Angeles County, collected between October 2016 and March 2017. The analytic sample for mobility data comprised 234,995 users.
- Foursquare POI Database: Used to identify and geolocate food outlets.
- U.S. Census Bureau (American Community Survey): Used for sensitivity analyses to assess the stability of neighborhood-level demographic characteristics over time.
Key Insights and Ideas
- Mobility as a Proxy for Exposure: Aggregated smartphone mobility data, specifically the frequency and proportion of visits to fast food outlets, serves as a robust and objective proxy for an individual's exposure to fast food environments.
- Neighborhood-Level Impact: The study highlights that neighborhood-level patterns of fast food visits are strongly associated with individual health outcomes, suggesting that the collective behavior within a community influences individual health.
- Superiority over Self-Report: Mobility-derived measures of fast food exposure were found to be comparable to, or even better predictors of, obesity and diabetes than traditional self-reported fast food intake. This underscores the limitations of self-report methods for dietary assessment.
- Dynamic vs. Static Environments: The research emphasizes the importance of considering dynamic "activity spaces" rather than just static home environments when studying the impact of food environments on health.
- Privacy-Preserving Aggregation: The study successfully demonstrates how to aggregate individual mobility data to a neighborhood level while preserving user privacy, making it a viable approach for public health research.
Key Results
- Neighborhood-level measures of FF outlet visits were strongly and significantly associated with self-reported FF intake, obesity, and diabetes.
- A 10% increase in FF visits/time was associated with a 16% increase in the odds of obesity and a 15% increase in the odds of diabetes.
- A 10% increase in FF visits/food was associated with a 13% increase in the odds of obesity and an 11% increase in the odds of diabetes.
- For obesity, frequent self-reported FF intake was associated with a 26% greater odds compared to no intake, while mobility measures showed stronger associations (e.g., FF visits/time associated with 16% increased odds).
- Model fit comparisons (Akaike weights) indicated that mobility variables (FF visits/time and FF visits/food) were better predictors of obesity and diabetes than self-reported FF intake frequency.
Limitations and Future Work
- Ecological Fallacy: The study acknowledges the risk of ecological fallacy, as aggregated neighborhood-level data is used to infer individual behaviors and outcomes. Individual variations within neighborhoods are obscured.
- Time Gap: The LACHS data (2011) and mobility data (2016-2017) are from different time periods, which could introduce incompatibility. Sensitivity analyses showed minimal impact from this gap.
- Smartphone User Bias: The study relies on smartphone users, who may not perfectly represent the entire population, particularly lower-income, older, or non-white individuals.
- Measurement Error: Gaps in smartphone data collection (e.g., when phones are out of service) can lead to under-sampling of visits. The study addressed this by using relative measures but acknowledges potential remaining error.
- Defining "Visit": The study defines a "stay" as at least 5 minutes in a location, potentially missing very brief FF outlet visits (e.g., drive-throughs).
- Consumption Uncertainty: While visits are observed, actual consumption of fast food during those visits is inferred, not directly measured.
- Food Delivery Not Included: The study focuses on in-person visits and does not account for fast food accessed via delivery services.
Future Work: The authors suggest that future research should explore:
- More granular spatial aggregation of mobility data to smaller areas while maintaining privacy.
- Generalizability of findings to different urban contexts.
- Leveraging more recent and densely sampled mobility data.
- Developing methods to characterize the nutritional quality of food consumed at FF outlets.
- Investigating "food deserts" and "food swamps" by considering lived environments beyond the home.
- Using mobility data for "natural experiments" to understand causal mechanisms.
- Developing more effective food environment policies and interventions.
Summary updated