Integrating human activity into food environments can better predict cardiometabolic diseases in the United States

This study addresses whether integrating human mobility into measurements of food environments improves the prediction of cardiometabolic disease (CMD) prevalence. Prior research links obesogenic foodscapes to poor diet and CMDs, but findings using location-based indices (e.g., mRFEI) are inconsistent due to confounding, scale issues, and neglect of actual consumer movement. The authors propose that food procurement activity—where people actually visit and shop—precedes diet and health outcomes and should be incorporated. They develop a Retail Food Activity Index (RFAI) using large-scale GPS-based visit data to food outlets across the United States, and evaluate whether this activity-based index better explains variation in CMD prevalence than traditional location-based metrics, with implications for more effective and equitable policies and interventions.

The paper situates its contribution within a literature showing mixed associations between local food environments and obesity/CMDs, with reviews documenting positive, negative, null, and mixed results for fast food access and health outcomes. Methodological issues include exposure measurement error, heterogeneity in obesity indicators, and the modifiable areal unit and uncertain geographic context problems. Prior work emphasizes that individuals’ activity spaces often extend beyond residential neighborhoods and that mobility and behavioral factors influence diet, suggesting that static residential measures may mischaracterize exposures. Studies indicate socioeconomic and racial/ethnic disparities in both access and behaviors; however, evidence on diet quality across groups is nuanced, with findings such as relatively better diet quality among some Hispanic and Black populations when accounting for covariates. Policy evaluations of geographic access interventions often show limited impact on diet/health, underscoring the need for approaches that incorporate behavior and mobility.

Design: Nationwide observational study constructing an activity-based food environment index (RFAI) at the U.S. census tract level and comparing it to the location-based modified Retail Food Environment Index (mRFEI). Associations with CMD prevalence were assessed using multivariable regression and multiple sensitivity analyses. Data sources: - mRFEI (CDC DNPAO, initial release 2011): Percentage of healthy food retailers among all qualified retailers within or 0.5 miles of a tract. Healthy retailers defined via 2007 NAICS as supermarkets and larger grocery stores (445110; supermarkets ≥50 employees, larger groceries 10–49), warehouse clubs (452910), and fruit/vegetable markets (445230). Less healthy: limited-service restaurants (722211), small groceries (445110 with ≤3 employees), and convenience stores (445120). - Human mobility and POIs (SafeGraph Core Places and Patterns, 2018–2019): Anonymized, aggregated OD flows from ~10% of GPS-enabled U.S. devices. Home inferred from 6 weeks of nighttime data (Geohash-7) and mapped to census geographies. Included 359,365 food retailer POIs and 94,256,870 visit records (destination POI, origin home tract, annual visit counts). Food retailer categories (2017 NAICS): Healthy—supermarkets and grocery stores (445110; all employment sizes), warehouse clubs (452311), fruit/vegetable retailers (445230). Less healthy—convenience stores (445120), limited-service restaurants (722513). Visits aggregated by origin tract. Mobility metrics computed: geodetic distance from tract centroid to POI; percent of visits within 0.5 miles of home tract boundary. - CMD prevalence (CDC PLACES 2019 model-based estimates): Adult prevalence for high blood pressure, coronary heart disease, diagnosed diabetes, high cholesterol, and obesity at the tract level. - Covariates: 2018 CDC Social Vulnerability Index (SVI) theme percentile ranks (socioeconomic status; household composition and disability; minority status and language; housing type and transportation); ACS 2013–2018 demographics (percent female, minority, low income, less than high school education, ages <5 or >64, median family income); 2020 Decennial Census (population, population density, racial composition); USDA Food Access Research Atlas (2019) urban indicator and food desert status (low-income, low-access by distance thresholds). Total number of food retailers per tract (log-transformed). Per-capita visit volume (log) used to account for potential mobility coverage bias. Indices: - RFAI (activity-based): For each home tract, RFAI = visits to healthy food retailers / (visits to healthy + visits to less healthy), expressed 0–100. - mRFEI (location-based): Percentage of healthy retailers among all retailers in/near the tract per CDC definition (0–100). Statistical analyses: - Descriptive spatial analyses of visit distances and within-0.5-mile visit proportions; mapping performed in ArcGIS Pro 3.0.2; other visualizations in R 4.2.2 (ggplot2). - Determinants of food-shopping distance: Multivariate linear regression with log-median distance traveled as outcome; quantile regression with median distance as outcome. Predictors: SVI theme percentiles, urban indicator, food desert indicator, population density (and log per-capita visits for coverage bias). Alternative specs replaced SVI minority theme with indicators for predominantly non-Hispanic White, non-Hispanic Black, and Hispanic tracts (≥50%). - Determinants of RFAI and mRFEI: Separate multivariate linear regressions with each index as outcome and the same sociodemographic and geographic predictors. - Associations with CMD prevalence: Separate multivariate linear regressions for each CMD, using either RFAI or mRFEI as the key exposure, adjusting for all aforementioned covariates plus percent female, minority, low income, less than high school education, age structure, median family income, total retailers (log). Sensitivity analyses: (a) Generalized additive models to assess non-linear RFAI–CMD relationships. (b) Constructed an updated location-based index using 2018–2019 SafeGraph retailers (classifying all supermarkets/groceries as healthy) and repeated analyses. (c) Aggregated to county level and repeated analyses to assess scale effects. (d) County-level spatial error regression to account for spatial autocorrelation. (e) Stratified tract-level analyses by predominant racial/ethnic group. Two-sided t-tests with 95% CIs reported. Analyses conducted in STATA 17.0.

- Mobility patterns: Only 20.8% (SD 13.9%) of food retailer visits occurred within 0.5 miles of residents’ home tracts. Median travel distance to food retailers was 3.70 miles (IQR 2.61–7.02). Visit distribution by distance: 12.8% (SD 13.9%) within 1 mile; 37.1% (SD 19.4%) 1–5 miles; 16.8% (SD 12.8%) 5–10 miles; 12.8% (SD 12.5%) 10–20 miles; 20.5% (SD 14.8%) >20 miles. - Determinants of distance traveled: Doubling (100% increase) in population density associated with 27.6% (95% CI 27.4–28.0) shorter median distance. Urban residents traveled 25.5% (95% CI 24.2–26.9) shorter distances than non-urban. One-percentile increases in SVI themes associated with: household composition/disability −0.055% (95% CI −0.070 to −0.041) in distance; minority status/language +0.249% (95% CI 0.234–0.264); housing/transportation −0.164% (95% CI −0.177 to −0.151). Predominantly non-Hispanic White tracts traveled 15% (95% CI 14–16.1) shorter distances, while predominantly non-Hispanic Black tracts traveled 3.5% (95% CI 2.1–4.9) farther, conditional on covariates. - RFAI vs mRFEI: The indices had low correlation (r=0.069, 95% CI 0.061–0.076), indicating limited concordance between activity- and location-based measures. - Sociodemographic associations with RFAI: Higher social vulnerability in socioeconomic status and housing/transportation associated with lower RFAI by 0.0581 (95% CI 0.055–0.061) and 0.023 (95% CI 0.021–0.025) units per percentile, respectively. Food desert tracts had 1.499 (95% CI 1.355–1.644) units lower RFAI. Doubling population density associated with +0.569 (95% CI 0.526–0.611) units in RFAI. Minority status vulnerability associated with higher RFAI: +0.056 (95% CI 0.053–0.058) units per percentile. - Racial/ethnic disparities: Predominantly Hispanic tracts had higher RFAI (+2.992 units, 95% CI 2.773–3.209) than others; predominantly non-Hispanic Black tracts also higher (+0.463, 95% CI 0.231–0.694); predominantly non-Hispanic White tracts lower (−1.199, 95% CI 1.035–1.363). For mRFEI, predominantly Hispanic tracts were higher (+1.166, 95% CI 0.834–1.497), predominantly non-Hispanic Black lower (−1.919, 95% CI 1.567–2.271), and predominantly non-Hispanic White higher (+0.309, 95% CI 0.060–0.559) than others. - Associations with CMDs: One interquartile range (IQR) increase in RFAI associated with lower prevalence of obesity (−0.629%, 95% CI −0.674 to −0.585), high cholesterol (−0.171%, 95% CI −0.206 to −0.135), and high blood pressure (−0.521%, 95% CI −0.571 to −0.471) after covariate adjustment. Corresponding mRFEI associations were weaker: obesity −0.174% (95% CI −0.204 to −0.143), high cholesterol −0.064% (95% CI −0.089 to −0.039), high blood pressure −0.261% (95% CI −0.295 to −0.226). Sensitivity analyses (updated location-based index, county-level aggregation, spatial error models, non-linear GAMs, and racial/ethnic stratification) generally supported linear, robust effects of RFAI, especially for obesity and high blood pressure.

Findings demonstrate that accounting for human mobility via an activity-based index (RFAI) yields stronger and more consistent associations with CMD prevalence than a traditional location-based index (mRFEI). The majority of food retailer visits occur beyond immediate residential neighborhoods, challenging the assumption that local storefront composition sufficiently captures dietary exposure. RFAI identifies socioeconomic and racial/ethnic disparities distinct from mRFEI, revealing that behavioral patterns can differ from static access metrics. The stronger RFAI–CMD associations suggest that food procurement behavior is a critical antecedent in the pathway from environment to diet and health. Policy implications include rethinking reliance on geographic access levers alone—given the high share of non-local visits, altering in-neighborhood retail mix may have modest impact on behaviors. Instead, RFAI can help target investments and interventions (e.g., CSA, mobile markets, co-ops), inform hospital Community Health Needs Assessments, and guide tailored education, affordability, and transportation strategies. Racial/ethnic patterns (e.g., higher RFAI in predominantly Hispanic and non-Hispanic Black tracts despite differing mRFEI) underscore behavioral heterogeneity and the need for culturally appropriate, context-specific interventions.

The study introduces the Retail Food Activity Index (RFAI), an activity-based measure derived from large-scale GPS mobility data, and shows it differs markedly from the location-based mRFEI and better predicts the prevalence of multiple CMDs (notably obesity, high blood pressure, and high cholesterol) at the census tract level across the U.S. The work advances food environment research by integrating human behavior and mobility, offering a scalable tool to identify communities with less healthy food procurement patterns and to inform policy and interventions. Future research should: (1) refine retailer classification by integrating multiple data sources and store audits; (2) leverage more recent, post-pandemic mobility to assess shifts in food procurement; (3) employ longitudinal, quasi-experimental, or natural experiment designs to probe causality; (4) integrate individual-level behavioral and dietary data; and (5) develop public, interactive platforms to operationalize RFAI for decision-makers in the U.S. and other countries.

- Ecological inference limitation: Mobility data aggregated to tract-level cannot reveal individual or household behaviors; translating aggregate patterns to individuals risks ecological fallacy. - Data representation and bias: SafeGraph sampling, while broadly representative at county/state scales, may underrepresent certain groups (e.g., Hispanic, low-income, lower education), with variation across space and time. - Observational design: Causality cannot be established; the pathway from food activities to CMDs involves multiple intervening behaviors not captured here. - Measurement differences: RFAI and mRFEI use different business classifications and criteria (e.g., mRFEI considers grocery employment size; RFAI does not), potentially contributing to discrepancies. - Scope of behavior captured: RFAI reflects visit destinations, not the quality or quantity of foods purchased or consumed. - Temporal scope: Mobility restricted to 2018–2019 to avoid COVID-19 effects; patterns may have changed during/after the pandemic. - Structural and cultural factors: Unique barriers and culturally specific factors influencing diet in minority communities are not directly measured and may mediate relationships.