Characterizing US Spatial Connectivity and Implications for Geographical Disease Dynamics and Metapopulation Modeling: Longitudinal Observational Study

Introduction

Human mobility plays a crucial role in the spread of respiratory diseases [Findlater A, Bogoch II. Human mobility and the global spread of infectious diseases: a focus on air travel. Trends Parasitol. Sep 2018;34(9):772-783. [CrossRef] [Medline]1]. The combination of regional travel and local commuting represents the spatial connectivity between locations, serving as the main driver in the geographic diffusion of infectious diseases. Characterizing the spatial dynamics of pathogen transmission is, therefore, intricately tied to unraveling human mobility patterns. Such a task has proven to be challenging due to the inherent complexity and privacy-related limitations on collecting mobility data [de Montjoye YA, Gambs S, Blondel V, et al. On the privacy-conscientious use of mobile phone data. Sci Data. Dec 11, 2018;5(1):180286. [CrossRef] [Medline]2]. Over the past few decades, researchers have extensively relied on mobility data obtained from census records, surveys, transportation statistics, commuting data, and international air traffic data. Such datasets have widely contributed to a better understanding of human mobility patterns and their impact on the epidemic spread [Colizza V, Barrat A, Barthélemy M, Vespignani A. Predictability and epidemic pathways in global outbreaks of infectious diseases: the SARS case study. BMC Med. Nov 21, 2007;5(1):34-7015. [CrossRef] [Medline]3-Viboud C, Bjørnstad ON, Smith DL, Simonsen L, Miller MA, Grenfell BT. Synchrony, waves, and spatial hierarchies in the spread of influenza. Science. Apr 21, 2006;312(5772):447-451. [CrossRef] [Medline]6], but can be limited in their resolution or scale. More recently, this gap has been filled by the use of mobile phone data [Blondel VD, Decuyper A, Krings G. A survey of results on mobile phone datasets analysis. EPJ Data Sci. Dec 2015;4(1):10-1127. [CrossRef]7,Maxmen A. Can tracking people through phone-call data improve lives? Nature New Biol. May 30, 2019;569(7758):614-617. [CrossRef]8], primarily based on phone records, but no such data have been available in the United States.

The global health crisis triggered by COVID-19 has underscored the critical need for swift access to mobility to help mitigate the spread of the virus. The urgency of the situation prompted an unprecedented sharing of data by private companies worldwide, through legally and ethically compliant agreements. These data were based on mobile location–based app usage and thus provided incomparable access to high-resolution, large-scale, and near-real-time mobility data that have expanded human mobility science [Pappalardo L, Manley E, Sekara V, Alessandretti L. Future directions in human mobility science. Nat Comput Sci. Jul 2023;3(7):588-600. [CrossRef] [Medline]9] and computational epidemiology [Oliver N, Lepri B, Sterly H, et al. Mobile phone data for informing public health actions across the COVID-19 pandemic life cycle. Sci Adv. Jun 2020;6(23):eabc0764. [CrossRef] [Medline]10,Grantz KH, Meredith HR, Cummings DAT, et al. The use of mobile phone data to inform analysis of COVID-19 pandemic epidemiology. Nat Commun. 2020;11(1):4961-1723. [CrossRef]11]. The availability of these data has especially represented a shift in the US public health and it has been used to inform epidemic models and reveal the impact of mitigation strategies on behavior [Garnier R, Benetka JR, Kraemer J, Bansal S. Socioeconomic disparities in social distancing during the COVID-19 pandemic in the United States: observational study. J Med Internet Res. Jan 22, 2021;23(1):e24591. [CrossRef] [Medline]12-Badr HS, Du H, Marshall M, Dong E, Squire MM, Gardner LM. Association between mobility patterns and COVID-19 transmission in the USA: a mathematical modelling study. Lancet Infect Dis. Nov 2020;20(11):1247-1254. [CrossRef] [Medline]16]. Although the association between mobility patterns and COVID-19 transmission in the United States has been extensively explored, for example, in studies by Pei et al [Pei S, Kandula S, Shaman J. Differential effects of intervention timing on COVID-19 spread in the United States. Sci Adv. Dec 4, 2020;6(49):2375-2548. [CrossRef]14], Badr et al [Badr HS, Du H, Marshall M, Dong E, Squire MM, Gardner LM. Association between mobility patterns and COVID-19 transmission in the USA: a mathematical modelling study. Lancet Infect Dis. Nov 2020;20(11):1247-1254. [CrossRef] [Medline]16], Xiong et al [Xiong C, Hu S, Yang M, Luo W, Zhang L. Mobile device data reveal the dynamics in a positive relationship between human mobility and COVID-19 infections. Proc Natl Acad Sci USA. Nov 3, 2020;117(44):27087-27089. [CrossRef]17], Tokey [Tokey AI. Spatial association of mobility and COVID-19 infection rate in the USA: a county-level study using mobile phone location data. J Transp Health. Sep 2021;22:101135. [CrossRef] [Medline]18], Zheng et al [Zheng Z, Xie Z, Qin Y, Wang K, Yu Y, Fu P. Exploring the influence of human mobility factors and spread prediction on early COVID-19 in the USA. BMC Public Health. Dec 2021;21(1):1-13. [CrossRef]19], and Gao et al [Gao S, Rao J, Kang Y, et al. Association of mobile phone location data indications of travel and stay-at-home mandates with COVID-19 infection rates in the US. JAMA Netw Open. Sep 1, 2020;3(9):e2020485. [CrossRef] [Medline]20], no studies have been devoted to assessing when the underlying mobility network needs to be embedded into models to characterize epidemic spread.

Furthermore, the effects of control measures such as social distancing and travel restrictions as well as the most suitable geographical and temporal granularity for implementing these measures still lack clarity. This gap in understanding the characteristic spatio-temporal scale of mobility not only limits target control policies but also our ability to model transmission dynamics effectively. To date, mobility data have been integrated into epidemic models without due consideration for the optimal geographical (eg, municipalities, regions, and states) and temporal resolution (eg, day, week, and month) required to accurately capture epidemic spread. The level of granularity used in these models has consistently been dictated by a priori assessments from data providers [Pullano G. Human mobility and epidemics. PhD thesis. Sorbonne Université; 2021. URL: https://theses.hal.science/tel-0409974221,Zhang M, Wang S, Hu T, et al. Human mobility and COVID-19 transmission: a systematic review and future directions. Ann GIS. Oct 2, 2022;28(4):501-514. [CrossRef]22].

To address these gaps, this study made 3 key contributions. First, for characterizing spatiotemporal scales, we systematically analyzed human mobility patterns across the United States using high-resolution mobile app–based location data to characterize intercounty connectivity at different spatial and temporal scales before and after the COVID-19 pandemic. Second, for quantifying mobility’s impact on the geographic spread of disease, we integrated these mobility data into spatially explicit transmission models to assess its role in the geographic diffusion of SARS-CoV-2 during the first wave. Finally, for evaluating the predictive power across scales, we analyzed how the predictive ability of epidemic models changes with varying resolutions of mobility data, identifying the spatial and temporal scales of intercounty connectivity that significantly influence disease dynamics. This analysis allowed us also to explore the trade-offs between using fine-grained mobility data and aggregated data at coarser scales.

While our approach yields valuable insights into human behavior and disease dynamics and makes a theoretical contribution to the field, we acknowledge its limitations—most notably the simplifications in the disease model, such as assuming homogeneous mixing within counties. However, our focus was on understanding heterogeneity at scales larger than counties, as most public health data are typically available at the county level. This approach represents a significant step toward understanding the role of mobility in geographical disease diffusion and optimizing the integration of mobility data into epidemic models to better inform public health policies.

Methods

Study Design and Population

This study investigated the role of human mobility at various temporal and spatial scales in the spread of COVID-19 across the US counties, using mobility and disease incidence data. First, we examined human displacements and the resulting intercounty connectivity patterns in the United States based on the daily number of visits between census blocks. We also analyzed disease incidence during the early phases of the COVID-19 pandemic, from January 2019 to July 2020, and used an inference framework to estimate underreporting based on the number of new deaths. In the second step, we developed a metapopulation model to simulate disease transmission, incorporating intercounty mobility patterns and epidemiological data, adjusted for underreporting. The study focused on US counties with populations greater than 10,000 to minimize biases in mobility and incidence data.

Data Collection

Characterizing Intercounty Connectivity With Mobility Data

We used data from SafeGraph [Safegraph. URL: https://www.safegraph.com [Accessed 2025-02-10] 23] (now Advan Patterns [Advan Inc]), a platform gathering mobility data from location-based mobile app usage. Specifically, we used the daily Social Distancing dataset provided by SafeGraph (refer to

Multimedia Appendix 1

Supplementary material.

We quantified monthly intercounty connectivity by normalizing visits from an origin county to destination counties and calculated average daily visits for all county pairs from January 2019 to March 2021. We then normalized the results so that the sum of probabilities for each county equals 1. The resulting time-evolving network reflected daily coupling probabilities. Figure 1A shows the connectivity network for March 2020. Figure S2 in

Multimedia Appendix 1

Supplementary material.

‎

Characterizing Early Phase of COVID-19 in the United States With Public Health Data

The first confirmed case of COVID-19 in the United States emerged in Washington state on January 21, 2020, quickly leading to local transmission. Guidelines promoting social distancing and discouraging gatherings were issued on March 16, 2020. While many European countries enacted nationwide restrictions, the United States implemented localized measures at varying times. Lockdowns peaked in the United States in April 2020, with over 40 states enforcing stay-at-home or shelter in-place orders [CDC. COVID Data Tracker. 2020. URL: https://covid.cdc.gov/covid-data-tracker [Accessed 2025-02-10] 24]. Despite these efforts, COVID-19 continued to spread, with most US counties reporting cases by June 2020. As cases surged again in October 2020, recommendations for social distancing were made to keep epidemic activity in check.

The COVID-19 disease incidence data were derived from the Centers for Disease Control and Prevention (CDC) data [CDC. COVID Data Tracker. 2020. URL: https://covid.cdc.gov/covid-data-tracker [Accessed 2025-02-10] 24]. We used daily reports of new cases and the timing of their arrival in each US county, defined as the day when at least 10 cases were recorded in that area. Daily new-reported cases and time of arrivals were adjusted for potential underreporting at the county level, estimated using global data on COVID-19 cases and fatalities, as outlined in a study by Russell et al [Russell TW, Golding N, Hellewell J, et al. Reconstructing the early global dynamics of under-ascertained COVID-19 cases and infections. BMC Med. Oct 22, 2020;18(1):332. [CrossRef] [Medline]25] (Figures 1B and 1C).

Statistical Analysis

Describing Temporal and Spatial Variability in the Mobility Network

We examined the monthly network structure to evaluate the temporal mobility patterns, quantifying each county’s degree (number of connections to other counties). We also defined link persistence as the probability links with nonzero mobility in a month of 2019 that remained in the same month of 2020 and 2021. We also fit a gravity model to the intercounty connectivity network for each month. Community detection analysis using the stochastic InfoMap algorithm developed by Rosvall et al identifies regions with more frequent internal movements [Rosvall M, Axelsson D, Bergstrom CT. The map equation. Eur Phys J Spec Top. Nov 2009;178(1):13-23. [CrossRef]26]. We used a bootstrap resampling method to account for stochasticity (refer to

Multimedia Appendix 1

Supplementary material.

Incorporating Human Mobility Into Infectious Disease Models

We used a stochastic non-Markovian transmission model with a metapopulation structure at the US county level [Pullano G. Human mobility and epidemics. PhD thesis. Sorbonne Université; 2021. URL: https://theses.hal.science/tel-0409974221]. The model accounts for disease transmission proportional to (1) infected residents not moving, (2) infected visitors coming from other counties, and (3) returning residents previously infected in other counties. The resulting force of infection in the county i is defined as follows:

${\mathrm{\lambda }}_{\mathrm{i}}={\mathrm{\lambda }}_{\mathrm{i}\mathrm{i}}+{\sum }_{\mathrm{i}\ne \mathrm{j}}{\mathrm{\lambda }}_{\mathrm{j}\mathrm{i}}^{\mathrm{v}}+{\sum }_{\mathrm{i}\ne \mathrm{j}}{\mathrm{\lambda }}_{\mathrm{i}\mathrm{j}}^{\mathrm{r}}$

${\displaystyle {\lambda }_{\mathrm{i}\mathrm{i}}=\beta {\mathrm{p}}_{\mathrm{i}\mathrm{i}}^2\frac{{\mathrm{I}}_{\mathrm{i}}}{\widehat{{\mathrm{N}}_{\mathrm{i}}}}}$

${\displaystyle {\lambda }_{\mathrm{j}\mathrm{i}}^{\mathrm{v}}=\beta {\mathrm{p}}_{\mathrm{i}\mathrm{i}}{\mathrm{p}}_{\mathrm{j}\mathrm{i}}\frac{{\mathrm{I}}_{\mathrm{j}}}{{\hat{\mathrm{N}}}_{\mathrm{i}}};{\lambda }_{\mathrm{i}\mathrm{j}}^{\mathrm{r}}=\beta {\mathrm{p}}_{\mathrm{i}\mathrm{j}}\frac{{\hat{\mathrm{I}}}_{\mathrm{j}}}{{\hat{\mathrm{N}}}_{\mathrm{j}}}}$

where p_ij is the coupling probability extracted from the intercounty connectivity network. The effective population, and effective number of infections are, respectively, defined as follows:

${\displaystyle {\displaystyle \widehat{N_i}=p_{ii}{N}_i+\sum _{i\ne j}{p}_{ji}{N}_j}}$

${\displaystyle {\displaystyle \widehat{I_i}=p_{ii}{I}_i+\sum _{i\ne j}{p}_{ji}{I}_j}}$

We considered the susceptible-exposed-infectious-recovered epidemic dynamics specific to COVID-19. The epidemics parameters were described by Pullano et al [Pullano G, Valdano E, Scarpa N, Rubrichi S, Colizza V. Evaluating the effect of demographic factors, socioeconomic factors, and risk aversion on mobility during the COVID-19 epidemic in France under lockdown: a population-based study. Lancet Digit Health. Dec 2020;2(12):e638-e649. [CrossRef] [Medline]29].

The detailed mathematical framework, model calibration, and implementation details can be found in

Multimedia Appendix 1

Supplementary material.

Inference Framework and Goodness of Fit

To calibrate the epidemic pathway, we used the cumulative number of infected US counties (Figure 1C). Calibration covered March 14 to July 15, 2020, when all counties reported infections. Parameter estimates β_pre−LD (March 15‐31) and β_post−LD (March 31-May 15) were derived using maximum likelihood, assuming a Poisson distribution for reported cases.

We assessed model performance by comparing the modeled invasion probability p_i,inv(t) with the observed early phase COVID-19 spatial invasion. p_i,inv(t) denotes the probability for a county i reporting at least 10 infected cases on day t in the simulation. The goodness of fit is defined as follows:

${\displaystyle {\displaystyle G(t)=\sum _i\left(I_i\log p_{i,\text{inv}}(t)+\left(1-I_i\right)\log \left(1-p_{i,\text{inv}}(t)\right)\right)}}$

I_i=1 if the county i reported at least 10 infected cases at the day t; I_i=0 otherwise.

Comparing Models Across Geographical Scales

In designing the metapopulation structure at different spatial scales, we aimed to better understand the role of mobility heterogeneity that matters in disease diffusion. Beyond this primary goal, this approach also highlighted the data requirements necessary for accurate modeling at different spatial scales. Specifically, it explored the implications of using mobility data at a coarser regional scale (eg, cluster or state), assuming higher-resolution flow data are unavailable.

For a given spatial scale R (eg, cluster and state), we randomized the mobility links among counties within a region R, preserving the number of links. Coupling probabilities for connected counties within each region R are equal to the average coupling probability of the links within that region, as follows:

${\displaystyle {\displaystyle p_{ij}^R=\frac{\sum _{l,k\in R}{p}_{lk}}{N}}}$

with i,j, ∈, and R, while coupling probabilities across regions are not changed.

Ethical Considerations

This study used publicly available, deidentified, and aggregated data from SafeGraph (now Advan Patterns) and the CDC. SafeGraph ensures user privacy by implementing differential privacy techniques, which involve adding Laplacian noise to anonymize data at the Census Block level. In addition, SafeGraph excluded connections associated with census block groups containing data from fewer than 2 devices. Ethical review for this study was sought from the Institutional Review Board at Georgetown University and the study was approved on October 14, 2020 (STUDY00003041).

Results

Analyzing the US county connectivity via mobility data revealed temporal and geographical stability, identifying the key scale driving COVID-19’s initial spread. Our findings addressed public health needs and determined the optimal scale for metapopulation model design.

Temporal Stability of the Intercounty Connectivity Network

From January 2019 to March 2021, mobility remained relatively stable, except for a significant drop in April 2020 coinciding with lockdown measures, reducing daily visits from 45 million to around 25 million nationwide (Figure 2A). The mobility shock extended throughout the month, encompassing a transitional period (Figures 2A and 2B). Analysis of the intercounty connectivity network’s temporal evolution revealed a consistent seasonal pattern in degree distribution and link persistence. Only April 2020 showed local variations, with a 23% reduction in degree (from 1144 to 877) compared with 2019 and a 20% reduction in link persistence (from 0.70 to 0.56) compared to previous months in 2020. Surprisingly, no variation occurred in November 2020, despite strong social distancing recommendations preceding the winter surge of SARS-CoV-2. The reduction in rural-urban connections was particularly pronounced, with a 22% decrease (from 0.75 to 0.58). The decrease stabilized in May and beyond. Notably, urban-urban connections exhibited greater resilience over time when compared with connections involving rural areas. We hypothesized that rural-rural connections were systematically less persistent due to the inherently less stable nature of these links. Factors such as lower population density, reduced economic activity, and less frequent interconnectivity in rural areas contributed to this instability. Furthermore, while coupling probabilities stayed consistent over the study period (Figure 1C), the probability of staying in the home location exhibited larger variability (

Multimedia Appendix 1

Supplementary material.

Despite occasional local fluctuations, the intercounty connectivity network demonstrated temporal stability and high predictability through a gravity fit model (Figures S7-S9 in

Multimedia Appendix 1

Supplementary material.

‎

Spatial Stability of the Intercounty Connectivity Network

To identify the geographic scale at which mobility is highly connected, we detected clusters of counties that were more connected via mobility within the cluster than outside the clusters, we used a network community detection algorithm. Our hypothesis was that this partitioning of the United States would be at a geographic scale larger than 3143 US counties but smaller than 50 US states or 10 Health and Human Services (HHS) regions. Indeed, we found that based on human mobility, the United States can be partitioned into around 100 regions that split most US states into multiple clusters (Figure 3). We also found that these clusters were highly and spatially contiguous and respected state boundaries (with a similarity measured by normalized mutual information as 0.82). Furthermore, these regions demonstrated stability over time (normalized mutual information=0.95) despite the perturbations of the early phases of the COVID-19 pandemic (Figure 3B and in Figures S9-S10 in

Multimedia Appendix 1

Supplementary material.

‎

Implications for Metapopulation Disease Models

After examining the stability of mobility patterns over time and space, we evaluated how spatiotemporal scales of mobility affect disease modeling in a metapopulation framework by integrating connectivity networks into a model simulating the initial SARS-CoV-2 spread in the United States. We investigated the influence of geographic scale by homogenizing networks at different spatial levels, maintaining the county resolution. In addition, to gauge the impact of temporal scale, we provided the model with either a time-evolving or static connectivity network representing mobility patterns since March 2020, considering the network’s temporal stability. Goodness of fit was assessed by comparing predicted and observed disease arrival times in counties.

Figure 4A shows that a county-level metapopulation model using empirical connectivity networks accurately predicts early COVID-19 diffusion, outperforming a Erdős–Rényi random network. Indeed, the empirical mobility network had a stronger goodness of fit throughout the early phase of the pandemic. This emphasized the crucial role of human mobility in the spatial spread of the initial SARS-CoV-2 wave and underscored the importance of mobility data for modeling SARS-CoV-2 spread. In addition, predictions based on static and time-varying mobility networks were comparable, suggesting that static data were sufficient to accurately capture epidemic spatial heterogeneity. Following the initial invasion phase in March, transmission became more widespread, reducing the importance of spatial connections. At this stage, random networks and county-level networks yielded similar performance, indicating that local epidemic growth outweighed the influence of importations and exportations. Figure 4B shows that county-level mobility data predicted spatial diffusion better than data at larger scales, such as US HHS regions, states, and clusters (as defined in Figure 3). All larger-scale data performed similarly to a random network, indicating they lack the granularity needed to capture diverse mobility patterns effectively. Sensitivity analyses on the definition of the time of arrivals and its impact on the goodness of fit are reported in

Multimedia Appendix 1

Supplementary material.

‎

Discussion

Principal Results

Our findings revealed significant insights into the dynamics of human mobility and their implications for infectious disease modeling. In contrast to findings from other countries (eg, France [Pullano G, Valdano E, Scarpa N, Rubrichi S, Colizza V. Evaluating the effect of demographic factors, socioeconomic factors, and risk aversion on mobility during the COVID-19 epidemic in France under lockdown: a population-based study. Lancet Digit Health. Dec 2020;2(12):e638-e649. [CrossRef] [Medline]29], India [Saha J, Barman B, Chouhan P. Lockdown for COVID-19 and its impact on community mobility in India: an analysis of the COVID-19 Community Mobility Reports, 2020. Child Youth Serv Rev. Sep 2020;116:105160. [CrossRef] [Medline]30], Germany [Sobolevsky S, Szell M, Campari R, Couronné T, Smoreda Z, Ratti C. Delineating geographical regions with networks of human interactions in an extensive set of countries. PLoS One. 2013;8(12):e81707. [CrossRef] [Medline]31], and Spain [Steinegger B, Granell C, Rapisardi G, et al. Joint analysis of the epidemic evolution and human mobility during the first wave of COVID-19 in Spain: retrospective study. JMIR Public Health Surveill. May 22, 2023;9:e40514. [CrossRef] [Medline]32]), we observed that despite the implementation of local social distancing measures and lockdowns, intercounty connectivity remained largely unperturbed, leading to rapid geographic diffusion of SARS-CoV-2. Mobility patterns experienced only marginal changes before and after the early-stage COVID-19 pandemic. The most notable disruption occurred during the first lockdown period in April 2020, when mobility sharply declined. Although a temporary reduction in mobility was observed, it proved to be short-lived as mobility patterns quickly returned to prerestriction levels. Importantly, this decline in mobility did not alter the underlying intercounty connectivity structure, potentially diminishing the overall effectiveness of the implemented travel restrictions. Even when mobility is limited—resulting in low-strength edges between locations—these connections can still facilitate the spread of disease, introducing pathogens into new areas [Colizza V, Vespignani A. Epidemic modeling in metapopulation systems with heterogeneous coupling pattern: theory and simulations. J Theor Biol. Apr 7, 2008;251(3):450-467. [CrossRef] [Medline]33]. This underscores the challenge of relying solely on travel restrictions as a mitigation strategy, as they may fail to disrupt the pathways that enable disease transmission, particularly when the connectivity network remains intact. Such insights emphasize the need for more nuanced approaches that address the persistence of network structures during public health interventions. Notably, even during periods of social distancing recommendations, the mobility network remains relatively stable. Assuming the lockdown represents the most extreme form of mobility disruption, the temporal stability findings suggest that global human mobility demonstrates resilience against short-term changes.

We also assessed the spatial stability of the intercounty connectivity network by detecting spatial communities based on mobility patterns. Our results indicate that mobility-driven clusters align closely with state boundaries, reflecting the influence of administrative and geographical factors on human movement, accordingly with Steinegger et al [Sobolevsky S, Szell M, Campari R, Couronné T, Smoreda Z, Ratti C. Delineating geographical regions with networks of human interactions in an extensive set of countries. PLoS One. 2013;8(12):e81707. [CrossRef] [Medline]31]. These clusters exhibited remarkable stability over time, reinforcing the idea that spatial mobility patterns are deeply ingrained and relatively resistant to abrupt changes. The fact that mobility patterns are highly correlated with state boundaries suggests that state-level structures could be effective for designing target public health interventions based on travel reductions. Our findings underscore the importance of considering mobility patterns when designing interventions, resource allocation, and disease control strategies.

As shown in the context of COVID-19 pandemic in France [Pullano G, Di Domenico L, Sabbatini CE, et al. Underdetection of cases of COVID-19 in France threatens epidemic control. Nature New Biol. Feb 4, 2021;590(7844):134-139. [CrossRef]34], we also demonstrated that incorporating high-resolution human mobility data are crucial for accurately capturing the spatial spread of infectious diseases. Our findings indicate that county-level, daily mobility data offer the most accurate representation of the spatial spread of disease in the United States. Notably, static county-level mobility data achieve similar model performance to real-time data, suggesting that an undisturbed representation of reality is adequate for reproducing spatial spread. More interestingly, our exploration of various spatial scales for metapopulation models underscores the significance of aligning the model’s structure with the inherent spatial scale of human mobility. While county-level mobility data yield the most accurate depiction, mobility databased clusters, US states, and HHS regions do not capture the heterogeneity of the COVID-19 geographical diffusion.

Limitations

While our study provides valuable insights, it is not without limitations. Our work focused on the early phase of the pandemic, during which response measures (eg, social distancing, closures, and lack of masking) were largely homogeneous in the United States, and pharmaceutical measures (eg, vaccination and antivirals) were not available; thus, these findings are not generalizable to later stages. Furthermore, we assume homogeneity within the US counties. In addition, Safegraph mobility data, like all mobile app–based location data, exhibit sampling biases. Ongoing efforts to comprehend these biases are crucial for developing better correction methods [Coston A, Guha N, Ouyang D, Lu L, Chouldechova A, Ho DE. Leveraging administrative data for bias audits: assessing disparate coverage with mobility data for COVID-19 policy. In: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency FAccT ’21. Association for Computing Machinery; 2021:173-184. [CrossRef] ISBN: 978-1-4503-8309-735]. An independent analysis by Safegraph revealed the underrepresentation of older and non-White individuals in point of interest-specific analyses, though the panel is representative of race, education, and income [Coston A, Guha N, Ouyang D, Lu L, Chouldechova A, Ho DE. Leveraging administrative data for bias audits: assessing disparate coverage with mobility data for COVID-19 policy. In: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency FAccT ’21. Association for Computing Machinery; 2021:173-184. [CrossRef] ISBN: 978-1-4503-8309-735,Fox R. “What about bias in your dataset?”: Quantifying sampling bias in SafeGraph Patterns. URL: https://medium.com/safegraph/what-about-bias-in-the-safegraph-dataset-b95b7208893c [Accessed 2025-02-10] 36].

Comparison With Previous Work

Since the onset of the COVID-19 pandemic, mobile phone data have played a crucial role in addressing the public health crisis [Oliver N, Lepri B, Sterly H, et al. Mobile phone data for informing public health actions across the COVID-19 pandemic life cycle. Sci Adv. Jun 2020;6(23):eabc0764. [CrossRef] [Medline]10,Grantz KH, Meredith HR, Cummings DAT, et al. The use of mobile phone data to inform analysis of COVID-19 pandemic epidemiology. Nat Commun. 2020;11(1):4961-1723. [CrossRef]11,Susswein Z, Rest EC, Bansal S. Disentangling the rhythms of human activity in the built environment for airborne transmission risk: An analysis of large-scale mobility data. elife. Apr 4, 2023;12:e80466. [CrossRef] [Medline]13-Chang S, Pierson E, Koh PW, et al. Mobility network models of COVID-19 explain inequities and inform reopening. Nature New Biol. Jan 2021;589(7840):82-87. [CrossRef] [Medline]15,Zhang M, Wang S, Hu T, et al. Human mobility and COVID-19 transmission: a systematic review and future directions. Ann GIS. Oct 2, 2022;28(4):501-514. [CrossRef]22,Pullano G, Valdano E, Scarpa N, Rubrichi S, Colizza V. Evaluating the effect of demographic factors, socioeconomic factors, and risk aversion on mobility during the COVID-19 epidemic in France under lockdown: a population-based study. Lancet Digit Health. Dec 2020;2(12):e638-e649. [CrossRef] [Medline]29,Saha J, Barman B, Chouhan P. Lockdown for COVID-19 and its impact on community mobility in India: an analysis of the COVID-19 Community Mobility Reports, 2020. Child Youth Serv Rev. Sep 2020;116:105160. [CrossRef] [Medline]30,Pullano G, Di Domenico L, Sabbatini CE, et al. Underdetection of cases of COVID-19 in France threatens epidemic control. Nature New Biol. Feb 4, 2021;590(7844):134-139. [CrossRef]34,Xia J, Yin K, Yue Y, et al. Impact of human mobility on COVID-19 transmission according to mobility distance, location, and demographic factors in the Greater Bay Area of China: population-based study. JMIR Public Health Surveill. Apr 26, 2023;9:e39588. [CrossRef] [Medline]37]. During this period, numerous network operators and private enterprises have made considerable efforts to swiftly share their data within the confines of legal regulations. Consequently, researchers worldwide have embarked on working with this data, monitoring human behavior caused by containment measures and adaptive responses to the epidemic, and using it to enhance epidemic models in order to increase their reliability [Pei S, Kandula S, Shaman J. Differential effects of intervention timing on COVID-19 spread in the United States. Sci Adv. Dec 4, 2020;6(49):2375-2548. [CrossRef]14,Chang S, Pierson E, Koh PW, et al. Mobility network models of COVID-19 explain inequities and inform reopening. Nature New Biol. Jan 2021;589(7840):82-87. [CrossRef] [Medline]15,Zhang M, Wang S, Hu T, et al. Human mobility and COVID-19 transmission: a systematic review and future directions. Ann GIS. Oct 2, 2022;28(4):501-514. [CrossRef]22,Pullano G, Valdano E, Scarpa N, Rubrichi S, Colizza V. Evaluating the effect of demographic factors, socioeconomic factors, and risk aversion on mobility during the COVID-19 epidemic in France under lockdown: a population-based study. Lancet Digit Health. Dec 2020;2(12):e638-e649. [CrossRef] [Medline]29,Pullano G, Di Domenico L, Sabbatini CE, et al. Underdetection of cases of COVID-19 in France threatens epidemic control. Nature New Biol. Feb 4, 2021;590(7844):134-139. [CrossRef]34].

While static mobility data have predominantly been analyzed and integrated into models over the past decades [Colizza V, Barrat A, Barthélemy M, Vespignani A. Predictability and epidemic pathways in global outbreaks of infectious diseases: the SARS case study. BMC Med. Nov 21, 2007;5(1):34-7015. [CrossRef] [Medline]3,Balcan D, Colizza V, Gonçalves B, Hu H, Ramasco JJ, Vespignani A. Multiscale mobility networks and the spatial spreading of infectious diseases. Proc Natl Acad Sci U S A. Dec 22, 2009;106(51):21484-21489. [CrossRef] [Medline]4], the current accessibility to real-time human behavior data prompts an essential investigation into the optimal scenarios for using this dynamic information versus relying solely on static representations of reality [Baker RE, Mahmud AS, Miller IF, et al. Infectious disease in an era of global change. Nat Rev Microbiol. Apr 2022;20(4):193-205. [CrossRef] [Medline]38]. Equally important is the exploration of the characteristic mobility scale to comprehensively capture the intricate coupling between different locations, a consideration with potential implications for target control policies to reduce epidemic activity, and for improving epidemic model forecasting. Furthermore, numerous researchers have emphasized the pressing necessity to implement standardized strategies that facilitate rapid data access while upholding stringent data privacy measures [de Montjoye YA, Gambs S, Blondel V, et al. On the privacy-conscientious use of mobile phone data. Sci Data. Dec 11, 2018;5(1):180286. [CrossRef] [Medline]2]. To address this gap in the literature, in this study, we investigated the spatial connectivity of the US counties during the early phase of the COVID-19 pandemic using high-resolution real-time human mobility data obtained from mobile phone usage.

Conclusions

While characterizing the key role of mobility in the spatial invasion of the COVID-19 pandemic in the United States, our study sheds light on the global stability of human mobility patterns, and the relevant information needed to design a reliable predictive model. This result may be specific to countries, such as the United States, in which mobility restrictions were not stringent, specified for intercounty mobility, nor enforced. Metapopulation models that incorporate accurate mobility data can provide valuable insights into disease dynamics and enhance our ability to predict and control the spread of future infectious disease outbreaks. Furthermore, standardized data extraction and sharing that we introduced might help facilitate the timelines associated with legal agreements for data sharing, which do not always align with the rapid spread of epidemics, thus diminishing the feasibility of timely responses to such outbreaks.