Studying the Relationships Between Food Acquisition and Health Outcomes During COVID-19

Chris Carey, Maia Guo, Nuoyi Wang
New York University, Center for Urban Science + Progress

Background

Food Insecurity and COVID-19

Food insecurity is the lack of consistent access to food adequate to sustain a healthy and active lifestyle.

2019 2020
1.2 million
 
2.0 million
(67%)
NYC residents experiencing food insecurity
Food Insecurity and Health Outcomes

Food insecurity has been consistently associated with negative health outcomes, including diet-related diseases, mental health, and comorbities with COVID-19 (Gundersen & Ziliak, 2015; CDC COVID-19 Response Team, 2020).

Food Acquisition Behaviors and Nutrition During COVID-19

Food secure NYC residents were also impacted by changes in food acquisition behavior by COVID-19. "NYC Food 20/20" (2020) reported the following impacts on food retail and services:

Reported Impacts on Food Retail
Reported Impacts on Food Service
  • Reduced frequencies of food shopping
  • Lower availability of supermarket staples
  • Lower priced items out of stock
  • Lower availability of essential items in low-income neighborhoods
  • Reduced restaurant spending
  • Higher closures of independent restaurants run by women, immigrants, and minorities
  • Lower impact on chain restaurants with greater financial capital

According to CUNY School of Public Health (2020) survey data, 54% of respondents reported having a less healthy diet than pre-pandemic. Lower income households also reported consuming more packaged food than higher income households, which has generally lower nutritional value than fresh foods.

Problem Statement

During the COVID-19 pandemic, vulnerable populations in NYC had excessive difficulty in safely obtaining affordable and nutritious food as typical avenues for food access were massively disrupted for sustained periods of time.
Hypothesis

The public health dangers and policy responses to COVID-19 exacerbated inequitable food access, and thus disproportionately increased food acquisition patterns associated with negative health outcomes.

Research

The authors sought to inform future food resiliency strategies in NYC by connecting mobility data to the existing scientific basis of associations between nutrition, food insecurity, and diet-related diseases.

Data

POI and weekly patterns to model food acquisition behaviors through aggregated food location visitations.
Demographic and socioeconomic attributes to identify subpopulations with similar characteristics.
410
GB of Data
36,195
Food retail and service POIs in NYC
3,664,420
Trip records to those POIs
21
Subcategories of those POIs
Selected POI categories assessed for overall nutritional value:
Supermarkets and Grocery Stores Healthy
Delis and Convenience Stores Less Healthy
Fast-Food Restaurants Unhealthy

Methodology

Analysis Metrics
Estimated Visitor Count
POI Category Visitation Proportion
Contact Density Index (CDI)
The adjusted weekly visitor count for each home CBG by the ratio of device count to CBG population.
The proportion of visitations made to a particular POI category.
The amount of contact exposure to other individuals based on POI visitor count, floor area, and dwell time.
To reflect the sampled data on the overall population.
To model dietary dependency on a particular POI category and by extension, the overall nutritional value of visitors’ diets.
To evaluate disproportionate COVID-19 exposure risks encountered during food acquisition.
Analytic Approaches
Population Clustering
Time Series Analysis
Data Visualization
Principal Component Analysis (PCA) and k-means clustering were performed with 2019 trip and census data aggregated by CBG to produce four subpopulations with similar demographic and socioeconomic characteristics.
POI category visitation proportions and CDI were compared over time periods by cluster and home CBG income percentile. Rolling averages and seasonality alignments were applied to observe valid trends.
An interactive web application was built to visualize metrics by CBG over selected time periods. Query results were displayed in a choropleth integrated with distribution and segmentation plots.
Limitations
Data Ethics and Impact Considerations
All analyses were limited by potential disparities in modeling real-world behavior through mobility data.
In the absence of a standardized privacy-conscientious framework for mobile phone data, the privacy protections provided by SafeGraph Inc. were relied upon.
Subject to:
  • Varing privacy-based aggregation techniques.
  • Varying device penetration rates.
  • Spatial or temporal autocorrelation.
  • Data dimensionality in k-means clustering.
  • Itinerary reconstruction was prohibited by only exposing anonymized visitation counts.
  • Spatial and temporal coarsening, minimum device count thresholding, differential privacy techniques, and Lapacian noise were added.
  • Groups were analyzed at coarse aggregation levels which combined multiple demographic and socioeconomic factors, to make the clear isolation of specific minority groups unachievable.

Results

Analysis Results
Visitation Totals
  • Figure 2a - Three-week average of COVID-19 cases in NYC and weekly food location visitations.
  • Figure 2b - Three-week average of COVID-19 cases in NYC and percent change in weekly food location visitations from 52 weeks prior.
Clusters
Income Level Education Level Families with Elderly & Children Ethnic Composition Families Receiving Food Assistance Benefits POI Count
Cluster #0 Moderate Moderate More elderly More Black and African American Moderate More Supermarkets, more Fast-Food Restaurants
Cluster #1 Higher Higher Fewer More White Fewer More Food Service, fewer Supermarkets
Cluster #2 Lower Lower More children More evenly-distributed More More Supermarkets, more Fast-Food restaurants
Cluster #3 Moderate Moderate More elderly More White and Asian Moderate Fewer Supermarkets
Income Level Education Level Families with Elderly & Children
Cluster #0 Moderate Moderate More elderly
Cluster #1 Higher Higher Fewer
Cluster #2 Lower Lower More children
Cluster #3 Moderate Moderate More elderly
Ethnic Composition Families Receiving Food Assistance Benefits POI Count
Cluster #0 More Black and African American Moderate More Supermarkets, more Fast-Food Restaurants
Cluster #1 More White Fewer More Food Service, fewer Supermarkets
Cluster #2 More evenly-distributed More More Supermarkets, more Fast-Food restaurants
Cluster #3 More White and Asian Moderate Fewer Supermarkets
Table 1 - Demographic and socioeconomic characteristics and POI Count summarized from the analytical plots. (The POI Count indicates the number of different types of POIs, which is counted by the unique place keys within the cluster area).
  • Figure 4a - Income, education, poverty, SNAP, and age characteristics of each cluster.
  • Figure 4b - Race composition of each cluster.
  • Figure 4c - POI Count (the number of different types of POIs) by category in different clusters.
POI Category Visitation
  • Figure 5a - Weekly visitations to food retail or service categories over time (three-week average).
  • Figure 5b - Weekly visitations to various POI categories over time (three-week average).
  • Figure 6 - Selected POI category visitation distributions by home CBG income percentile; median percentage (top), and percent change in median percentage (bottom).
  • Figure 7 - Selected POI category visitation distributions by demographic and socioeconomic cluster between March 16 - November 23, 2020 compared to 52 weeks prior; supermarkets (top), delis (center), and fast-food restaurants (bottom).

CDI


Interpretation
Consistency with Reported Food Acquisition Behaviors

Visitation totals showed trends consistent with NYC survey data from May 2020 which reported that 64% of respondents shopped less frequently and 49% ate more packaged food compared to before the pandemic (CUNY School of Public Health, 2020); food retail visitation totals were 50% lower in May 2020 year-over-year, but formed a greater proportion.

Disproportionate Supermarket Resiliency

Shoppers from low-income neighborhoods and clusters with higher rates of children and older residents had the greatest supermarket dependency. Low-income shoppers also reported the unavailability of essential and lower-priced food items at greater rates (CUNY Urban Food Policy Institute, 2020) despite having the smallest increases in supermarket dependency.

This suggested disparity in supermarket resiliency which exacerbated harm to high-poverty communities and individuals with diet-related diseases associated with food insecurity (Gundersen and Ziliak, 2015) who were already at greater risk from COVID-19 (Arasteh, 2021; CDC COVID-19 Response Team, 2020).

Increased Dependencies on Unhealthy Eating Establishments

Fast-food dependency temporarily increased citywide as non-fast food restaurant visitation proportions declined. CUNY School of Public Health (2020) surveys found that compared to pre-pandemic behavior: 54% of respondents reported having a less healthy diet, and lower-income, Black, and Latinx households consumed more packaged food at higher rates. These reports were consistent with observed fast-food and deli visitation proportions in total and by clusters.

CDI Disparity Followed COVID-19 Infection Disparity

Clusters and boroughs with higher CDIs also included neighborhoods with the highest COVID-19 infections per capita (NYC Health, 2021). CDI model limitations prevented drawing policy recommendations specific to the food sector with confidence. However, CDI could potentially act as a broad diagnostic metric for outlier detection in future work.

Policy Implications
Preventing Food Supply Disruption

Clusters #0 and #2 had the highest supermarket dependency, and thus were the most susceptible to food supply disruption when stay-at-home orders closed many food service options. This signaled that the greatest need to mitigate supermarket supply disruption was located in central and south Brooklyn, central and east Queens, and the Bronx. Expanding food supply in such neighborhoods may be the most effective direction of food insecurity reduction efforts.

Supporting Non-Fast Food Restaurants

After the initial stay-at-home order, most clusters and high-income neighborhoods increased their fast-food visitation proportions. These chain restaurants had the greatest financial resources to remain open. Meanwhile, over 1,000 NYC restaurants permanently closed with women and minority-owned businesses impacted disproportionately. Preserving greater nutritional choice to support stronger immune responses could be accomplished by ensuring immediate financial support to non-fast food services in future public health emergencies.

Data Visualization
Visualization Application

Conclusions

Survey data showed that shoppers from low-income and minority neighborhoods in NYC were more likely to experience supermarket food supply disruption, while mobility data showed that these same shoppers were also more dependent on supermarkets for a larger proportion of their diet.

Therefore, a comprehensive food resiliency strategy to reduce food insecurity and the resulting negative health outcomes during public health emergencies should:

The rise in food insecurity in NYC and its associated negative health outcomes require persistent monitoring for adverse long-term health impacts, recognition of the disparity in the city’s food resiliency system, and the collective support of New Yorkers to strengthen its weaknesses.


Technical Report

Credits

Our Team
Chris Carey
Maia Guo
Nuoyi Wang
Project Manager
Data Visualization Lead
Data Engineer
Data Modeling Lead
Research Analyst
Data Analytics Support
Designed project milestones, estimated work effort, tracked progress, and communication; led the implementation of client-facing data visualizations.
Led data extraction, integration, and processing; responsible for Machine Learning modeling; supported spatiotemporal data analysis.
Led theoretical support and direction guidance for research; led policy research; conducted time analysis; supported data integration, data processing and exploration analytics.
Our Sponsors & Mentor
CUNY Graduate School of Public Health & Health Policy
CUNY SPH // Sponsor

The CUNY Graduate School of Public Health & Health Policy (CUNY SPH) is committed to teaching, research and service that creates a healthier New York City and helps promote equitable, efficient and evidence-based solutions to pressing health problems facing cities around the world.

NYU Center for Urban Science + Progress
NYU CUSP // Sponsor

New York University's Center for Urban Science And Progress (CUSP) is an interdisciplinary research center dedicated to the application of science, technology, engineering, and mathematics in the service of urban communities across the globe.

Huy T. Vo
Assistant Profressor (CS@CUNY-CCNY)
Research Assistant Profressor (NYU-CUSP)
Capstone Mentor

Huy T. Vo is an assistant professor affiliated to both NYU CUSP and CUNY CCNY. His research interests involve big data analytics, large-scale visualization and high-performance computing.