Movement and disparities in COVID-19 in New York City

Like Comment
Read the paper

In the spring and summer of 2020, New York City suffered an intense wave of COVID-19. Testing capacity was limited, making it difficult to accurately measure how many people were infected across the city. Still, hospitalization and mortality data revealed that the outbreak affected some parts of the city far more severely than others. It was unclear to what extent the differences in clinical outcomes reflected variation in underlying risk factors and access to care leading to more severe illness as compared to geographic differences in the prevalence of infection across the city. To address this uncertainty and thereby help improve understanding of how and where to focus interventions, it was necessary to measure rates of infection across the city, not just in patients seeking care for COVID-19 but also in those with few or no symptoms.

Despite limited testing capacity, one population was consistently tested regardless of COVID-19 symptoms: pregnant women hospitalized for labor and delivery. We aggregated COVID-19 testing data from 1,746 pregnant women in six hospitals across New York City between March 22nd and May 3rd, 2020, during the height of the city’s epidemic. Most of the women had no symptoms upon hospitalization, yet 14% were positive for SARS-CoV-2. From a public health perspective, these women acted as sentinels providing insight into transmission in their communities. Using these data, we estimated the prevalence of SARS-CoV-2 across the New York City boroughs. The differences were substantial. The estimated prevalence in the Bronx and the south side of Queens was nearly twice the prevalence in Manhattan during the study period. These differences closely matched the previously observed differences in hospitalizations and mortality. They also align closely with recent findings from antibody tests, indicating that our approach can provide an accurate, timely, and efficient way to infer the prevalence of infection in the community in the midst of an outbreak.

What could have driven these geographic differences in prevalence? One likely possibility was that prevalence was associated with the ability to physically distance. Communities with many people who cannot isolate at home – such as essential workers – are likely to suffer a higher prevalence of infection. To explore this possibility, we measured declines in between-borough commuting-style movements during the study period compared to two months prior. The movement data came from millions of location recordings aggregated by Facebook’s Data for Good initiative. As expected, the boroughs with the lowest prevalence were also the ones where commuting-style movements declined the most. This suggests that it is imperative to invest in protecting essential workers and others who are unable to physically distance.

Our approach serves as an example of how diverse data sources can be integrated to identify drivers of disease transmission. We envision a future public health infrastructure that makes routine use of integrated microbiological, clinical, and behavioral data to provide more timely and accurate assessments of disease spread, revealing more effective ways to intervene.

Link to article:
Poster image credit: Sam Valadi / CC BY (

Stephen Kissler

Postdoctoral Fellow, Harvard T.H. Chan School of Public Health