 Research
 Open access
 Published:
Impact of network centrality and income on slowing infection spread after outbreaks
Applied Network Science volumeĀ 8, ArticleĀ number:Ā 16 (2023)
Abstract
The COVID19 pandemic has shed light on how the spread of infectious diseases worldwide are importantly shaped by both human mobility networks and socioeconomic factors. However, few studies look at how both socioeconomic conditions and the complex network properties of human mobility patterns interact, and how they influence outbreaks together. We introduce a novel methodology, called the Infection Delay Model, to calculate how the arrival time of an infection varies geographically, considering both effective distancebased metrics and differences in regionsā capacity to isolateāa feature associated with socioeconomic inequalities. To illustrate an application of the Infection Delay Model, this paper integrates household travel survey data with cell phone mobility data from the SĆ£o Paulo metropolitan region to assess the effectiveness of lockdowns to slow the spread of COVID19. Rather than operating under the assumption that the next pandemic will begin in the same region as the last, the model estimates infection delays under every possible outbreak scenario, allowing for generalizable insights into the effectiveness of interventions to delay a regionās first case. The model sheds light on how the effectiveness of lockdowns to slow the spread of disease is influenced by the interaction of mobility networks and socioeconomic levels. We find that a negative relationship emerges between network centrality and the infection delay after aĀ lockdown, irrespective of income. Furthermore, for regions across all income and centrality levels, outbreaks starting in less central locations were more effectively slowed by a lockdown. Using the Infection Delay Model, this paper identifies and quantifies a new dimension of disease risk faced by those most central in a mobility network.
Introduction
Since the start of the COVID19 pandemic, an active literature has evolved to study the spread and dynamics of the disease from mobility networks (Coelho etĀ al. 2020; Peixoto etĀ al. 2020; Chang etĀ al. 2020; Levin etĀ al. 2021) or sociospatial perspectives (Lee etĀ al. 2021; Li etĀ al. 2021; Cordes and Castro 2020). However, very few studies look at how both socioeconomic conditions and network properties interact, and how they influence outbreaks together (Chang etĀ al. 2020; Nande etĀ al. 2021). Chang etĀ al. (2020) create a mobility network from cell phone mobility data and model the spread of COVID19, identifying the importance of āsuperspreaderā points and higher infection rates among disadvantaged racial and sociodemographic groups. Nande etĀ al. (2021) explore how evictions influence the spread of COVID19 in various simulation scenarios, considering different policy responses and highly local contact networks. Further, while extensive work has been done to model the spread of the virus and nonpharmaceutical intervention effectiveness in terms of cases, hospitalizations, and deaths (Li etĀ al. 2021; Cordes and Castro 2020; Flaxman etĀ al. 2020; Oraby etĀ al. 2021; Oka etĀ al. 2021; Meo etĀ al. 2020) there is a lack of emphasis on the timing of case spread, and how interventions can delay a regionās first infection. Given the spatiotemporal granularity of cell phone mobility data capturing responses to lockdown policies, it is now possible to develop generalized, preventative methodologies which seek to further our understanding of disease vulnerability, and better prepare for novel outbreaks or variants.
This paper develops the Infection Delay Model (IDM), a novel effective distancebased methodology that can be used for assessing how lockdowns can delay a regionās first case and their intersection with socioeconomic inequalities. The IDM captures the difference between disease arrival times with and without a lockdown, using a novel application of cellĀ phone mobility data for effective distance research. To develop a forwardlooking understanding of the impacts of interventions on the timing of disease spread, a usecase of the IDM is presented which considers the potential variability of future outbreak scenarios. Drawing from recent studies of networkdriven contagion phenomena (Brockmann and Helbing 2014; Iannelli etĀ al. 2017; Balcan etĀ al. 2009), we simulate epidemics from every node in the transport network. By connecting those simulations with socioeconomic data, generalizable insights are uncovered which can be applicable beyond the specific spreading patterns observed during COVID19.
This paper uses the Metropolitan Region of SĆ£o Paulo (MRSP) as a case study to apply the IDM. Given its unique position as an area of early disease introduction and high intrastate transmission, COVID19 studies in the MRSP can help with preparation for future variants of COVID19 or other pandemics (Coelho etĀ al. 2020; Candido etĀ al. 2020).
Background
Networkbased analyses of COVID19
One branch of literature on COVID19 has focused on mobility networks to model the spread of the disease and assess the risks of cases and deaths. The data sources used to generate such networks range from domestic and international flight records (Coelho etĀ al. 2020; Kuo and Chiu 2021), to cell phone mobility records and geolocated visits to places of interest (Peixoto etĀ al. 2020; Chang etĀ al. 2020; Nande etĀ al. 2021; Ferreira etĀ al. 2021). The varying spatiotemporal granularity of the data sources used in these analyses have led to diverse outputs to identify regions at risk and explore how nonpharmaceutical interventions (NPIs) such as lockdowns impact mobility and vulnerability, but also how that impact might be different across wealthier or poorer regions (Gozzi etĀ al. 2021).
This area of literature uses transport flows to construct aggregated networks of population movement. Various methods have been implemented to study COVID19 risks on these mobility networks. Effective distancebased studies calculate the ādistanceā between two regions based on the degree of mobility flows between themāmore connected regions are effectively ācloserā (Coelho etĀ al. 2020). The effective distance of a region from an outbreak location has been shown to be predictive of infection arrival times (Brockmann and Helbing 2014; Iannelli etĀ al. 2017). Other studies build compartmental models on top of the mobility networks, calibrated to regional epidemic trajectories, and use epidemiological parameters and outbreak locations to simulate the course of an epidemic (Chang etĀ al. 2020; Balcan etĀ al. 2009; Peixoto etĀ al. 2020; Nande etĀ al. 2021). As greater mobility and persontoperson contact is associated with transmission, epidemic simulations can be run on mobility networks with adjusted levels of mobility or contact patterns to explore the impacts of real or hypothetical interventions on health outcomes (Levin etĀ al. 2021; Nande etĀ al. 2021).
Sociospatial analyses of COVID19
A separate branch of literature on COVID19 has focussed on disease vulnerability and its intersection with existing sociospatial inequalities. The range of analyses includes studies on how socioeconomic levels are associated with differences in terms of cases, hospitalizations, and deaths (Li etĀ al. 2021; Cordes and Castro 2020), health care facility access (Pereira etĀ al. 2021; Tao etĀ al. 2020), and inequalities in NPI adherence (Li etĀ al. 2021; Lee etĀ al. 2021; Jay etĀ al. 2020; Heroy etĀ al. 2022). It is worth noting that some of these analyses also study the impact of mobility restrictions, by either incorporating them as a proxy for the intensity of the economic downturn associated with the lockdown (Bonaccorsi etĀ al. 2020), or identifying likely determinants of spatial variations of reductions in mobility (Gauvin etĀ al. 2021). Overall, these spatial analyses often seek to uncover how variables such as race and income relate to COVID19 risks, to identify how existing inequalities are being compounded by the ongoing pandemic.
In an analysis of hospitalization and deaths in SĆ£o Paulo, it was found that black and pardo Brazilians were more likely to be hospitalized and die of COVID19 (Li etĀ al. 2021). Similarly, an analysis of clusters and contextual factors of COVID19 in New York City found that regions with larger black populations without health insurance had higher positive testing rates (Cordes and Castro 2020). Cell phone mobility data has also been used to study the interaction of lockdown adherence and socioeconomic inequalities. Conceptualizing mobility restrictions as a luxury not everyone can afford, it has been found that more vulnerable individuals were less able to reduce their mobilityāpotentially due to a lower probability of furlough or teleworking opportunities (Lee etĀ al. 2021; Li etĀ al. 2021).
Contributions
The first contribution of this paper is an integrated analysis of how the complex network properties of mobility patterns interact with socioeconomic characteristics to produce disease risk. While the sociospatial branch of literature has consistently identified intersections between socioeconomic vulnerability and disease burdens (Li etĀ al. 2021), current networkbased studies either use socioeconomic data to contextualize networkbased results (Coelho etĀ al. 2020; Gauvin etĀ al. 2021), or include it to identify relationships between mobility reductions and socioeconomic vulnerability (Pullano etĀ al. 2020; Valdano etĀ al. 2021; Gozzi etĀ al. 2021). There is a lack of investigation into the interaction between network properties and socioeconomic factors, and how they jointly drive the distribution of disease risk. It cannot be assumed that features such as network centrality and income are proxies for each other, justifying an investigation which explicitly examines both.
The second contribution of this study is the introduction of a new method that estimates the extent to which lockdown measures can slow down the spread of diseases while taking into account the spatial and temporal heterogeneity of disease spread in a network science approach. Existing networkbased and sociospatial research on cases, hospitalizations, and deaths fail to measure a crucial goal of early lockdowns, namely delaying the time until a regionās first case. Delaying disease onset with early interventions can buy time for health systems to increase hospital and intensive care capacity, and establish rapid testing sites (Rocha etĀ al. 2021). To investigate this dimension of disease risk, we introduce the Infection Delay Model, an effective distancebased method of calculating disease arrival times under baseline and lockdown mobility scenarios. Current literature which explores rankings of disease arrivals using effective distances does so while assuming a single known outbreak location (Brockmann and Helbing 2014; Iannelli etĀ al. 2017), or including a small subset of potential outbreak locations (Coelho etĀ al. 2020). These studies also overlook how rankings of disease arrivals are shaped by socioeconomic inequalities. Given recent literature on the outsized influence of the outbreak region on the trajectory of a communicable disease (Schlosser and Brockmann 2021), this study simulates outbreaks beginning in every region of the MRSP, to allow for generalizable findings that do not assume that the next outbreak will begin in the same region as the last.
Methodology
Data
CellĀ phone mobility data
Through an agreement with InLoco (Incognia 2020), a Brazilian cellĀ phone analytics company now known as Incognia, this paper had access to daily isolation levels for MRSP from March 1, 2020 to April 19, 2020. These data come spatially aggregated on a hexagonal grid using the H3 index at resolution 8 (Brodsky 2018). The data set contains 2893 hexagonal cells of roughly 740Ā m\(^2\) across the MRSP, of which 2599 had suitable time frames and auxiliary data after interpolation to be used in the analysis. The hexagonal isolation data is openly available in a data repository (see Availability of Data and Materials section). InLoco/Incognia gathers data by partnering with mobile phone applications, and uses software development kits to harvest location data while individuals are using partnered apps (Peixoto etĀ al. 2020). This form of location gathering provides precise geocoordinates, which are anonymized and aggregated to develop the social isolation indices. For a given hexagon cell, the proportion of individuals who reside in the cell and stay within it on a given day is recorded. This proportional value is used as a proxy for social isolation (Li etĀ al. 2021), recording the extent to which individuals travel outside their residence area. Higher or lower social isolation values indicate that fewer or more individuals are leaving their residence area, respectively (Ferreira etĀ al. 2021). The distribution of social isolation hexagon cells is presented in Fig.Ā 1. The same data set was used in Li etĀ al. (2021), showing that lower income individuals were less able to reduce their mobility after SĆ£o Pauloās lockdown, justifying our use of the cell phone data as capturing incomerelated mobility inequalities. The uneven coverage of the MRSP hexagon cells is a feature of the data set provided by InLoco/Incognia, discussed in the limitations section.
Travel survey data
The travel survey data for the MRSP were gathered from the 2017 MRSP household travel survey, conducted by the SĆ£o Paulo Metropolitan Transportation Department between June 2017 and October 2018 (METRĆSP 2018). The original data set is a table of survey responses regarding the total daily trips of 86,318 individuals who reside in the MRSP. On average, each individual reports 2.12 daily journeys, leading to a total of 182,994 trip reports (METRĆSP 2018). Key information for the reports are the journey origin and destination, along with the travel time. The interviews were conducted across 39 municipalities within the MRSP, divided into 510 research zones for the purposes of the survey. Of all the research zones, 66% lie within the main municipality in the MRSP, SĆ£o Paulo. The survey was designed to be statistically representative across the MRSP, and includes journey and population weights to scale responses by their frequency in the true population. From these weights, the total 2017 mobility flows between travel survey zones and 2017 estimates of populations were calculated. Population levels in 2020 were estimated by determining the geometric growth rate from 2010 and 2019 population totals, and scaling the 2017 populations to obtain population estimates per zone (Instituto Brasileiro de Geografia e EstatĆstica 2010, 2019). Population counts for the commuting areas are necessary as inputs to the networkbased compartmental epidemiological model used to simulate epidemics.
Census data
This study uses socioeconomic data from the official 2010 Brazilian Census, focused on the census tracts within the MRSP (Instituto Brasileiro de Geografia e EstatĆstica 2010; Pereira and GonĆ§alves 2019). Within the state of SĆ£o Paulo, there are 68,296 tracts, with data included on the total population, racial aggregates, average income per capita (Brazilian Real per calendar month), functioning water networks, and other relevant socioeconomic features. The census tracts within SĆ£o Paulo state cover a larger area than both the cellĀ phone mobility hexagon cells and travel survey zones, which are primarily focused on the MRSP. The population data from the the MRSP travel survey is more up to date than the 2010 census, therefore it is used in favour of the census data population totals. The census data remains useful for calculating regional income per capita averages, which are interpolated from census tracts into the hexagon cells.
Interpolating data to hexagonlevel
While the social isolation hexagon cells provide spatially and temporally granular information on the daily proportion of residents leaving a given area, information on which population subgroups are included in each hexagon cell remain unknown. This problem is shared across the growing body of literature using cellĀ phone mobility data for public health purposes, where anonymity measures by cellĀ phone data providers obscure information on the sample (Grantz etĀ al. 2020). While fundamental selection biases in the mobile phone data are a persistent issue, discussed in the limitations section, traditional data sources can be leveraged to generate population estimates within the hexagons (Aleta etĀ al. 2020).
The census tracts and travel survey zones are constructed of varying spatial structures which must be mapped to the social isolation hexagon cells. This process, known as spatial interpolation, is used in geospatial studies to estimate values in unknown area units using values in known geographic units (Comber and Zeng 2019). The spatial interpolation method used in this analysis is known as aerial weighting, which integrates socioeconomic estimates based on proportional overlap (Comber and Zeng 2019). This method depends on the assumption of homogeneously distributed characteristics within census tracts and travel survey zones, but benefits from transparency and simplicity relative to interpolation methods which rely on auxiliary information (Comber and Zeng 2019). Each hexagon cellās overlap with census tracts and travel survey zones was determined relative to their total areas. This proportional overlap area was used to generate a weighted allocation for income and population levels. For example, if a hexagon cell covered 50% of a travel survey zone with a population of 20, the hexagon cell would be assigned 10 individuals. FigureĀ 1 geographically displays the interpolated populations across the hexagon cells in the MRSP, and Fig.Ā 2 and TableĀ 1 display the distribution of incomes.
To interpolate the 2017 travel survey network to the hexagonal cells, the homogeneity assumptions of aerial weighting are extended to mobility flows between travel survey zones (Jang and Yao 2011). It is assumed that a hexagon cell overlapping with a given origin zone has a proportional quantity of outflow to all its targets. Similarly, hexagon cells overlapping with a given destination zone receive inflow from all relevant origin zones proportional to their intersection with that destination zone. An illustration of the travel flow interpolation to hexagon cells is provided in Fig.Ā 3.
Based on the interpolated mobility network, the indegree centrality of each hexagon cell is calculated. Indegree centrality is the number of edges that directly flow into a cell, representing the diversity of inflow connectionsāassociated with a regionās time to infection (Hunter etĀ al. 2020; Christley etĀ al. 2005). The weighted indegree (total travellers in) and weighted outdegree (total travellers out) are also highly correlated with the indegree, and have been shown to influence the spread of disease (Francetic and Munford 2021). Other centrality measures to determine a nodeās level of influence in a network include betweenness centrality, which measures the number of shortest paths that pass through a given node, and closeness centrality, which is the inverse of the geodesic distance from a given node to all others (LĆ¼ etĀ al. 2016). Other, more recent measures focus on community structures and distinguish inter versus intracommunity links when considering centrality (Rajeh etĀ al. 2022). This includes the neighbourhoodbased bridge node centrality, which measures how a nodeās neighbours will belong to other network components if it is removed (Meghanathan 2021). Indegree is chosen in this analysis because the infection delay model focuses solely on initial disease arrivals, and indegree centrality isolates these inward flows in a simple and interpretable measurement. The distribution of indegree centrality in the hexagoninterpolated mobility networkĀ is presented in Fig.Ā 4 and Table 2.
Infection delay model
This section will discuss in detail the methodology of the proposed Infection Delay Model. To provide an overview, the inputs to the Infection Delay Model are the effective distances from all pairs of hexagon cells, calculated under baseline and intervention mobility scenariosāthe baseline being the scenario with no mobility restrictions, i.e. no form of lockdown. These effective distances are translated into two sets of infection arrival times, whose differences represent the āinfection delayā of an intervention. For a given outbreak location, the infection delay algorithm first determines how much time would be āaddedā until every regionās first case if the outbreak location implemented a lockdown on the first day. Assuming a lockdown is not immediately implemented at the outbreak location, we simulate the unmitigated spread of the disease using a SIR model and calculate the infection delay of a lockdown intervention at every subsequent day based on the currently infected regions. This produces a time series plot for every region, known as its infection delay curve, showing the infection delay values over time for an outbreak beginning at a known region. While this provides estimates of the ātime addedā to a regionās first case from a known location, we calculate and characterize infection delay curves for every region under every possible outbreak scenario to understand general trends.
Calculating effective distance
To calculate the effective distances for the hexagonscaled mobility network, this analysis uses the ādominant pathā effective distance, a metric used in numerous disease arrival time analyses (Iannelli etĀ al. 2017; Brockmann and Helbing 2014; Coelho etĀ al. 2020; Gautreau etĀ al. 2008), translated into Python by Iannelli etĀ al. (2017). Measures of the dominant path effective distance focus solely on the most probable path of transmission from hexagon i to j. To calculate this value, for every connected origin i and destination j in the network, we calculate the transition rate as the proportion of total travellers beginning in hexagon i who arrive in hexagon j, denoted as \(0 \le P_{ij} \le 1\) (Brockmann and Helbing 2014). The effective distance between hexagons i and j is calculated as:
which is used as an edge weight for every pair of i,Ā j hexagons, or nodes in the mobility network (Brockmann and Helbing 2014). These edge weights, greater than or equal to one, are used in a weighted shortest path analysis to determine the dominant path effective distance between every pair of hexagon cells. With \(d_{ij}\) calculated for all edges in the network, the dominant path between i and j is chosen as the path which minimizes the sum of effective distance edge weights between them. Finally, the dominant path effective distance between two hexagons (\(D_{ij}\)) is calculated as the sum of the effective distances along the determined shortest path. This basic dominant path effective distance can be used to detect rankings of arrival times for a given outbreak location (Brockmann and Helbing 2014).
The traditional dominant path effective distance model is solely based on the mobility network, captured by \(P_{ij}\), and does not have parameters which can incorporate changing epidemiological parameters or rates of mobility reduction in the network. To add epidemiological and mobilitybased parameters, useful for a comparative analysis, the effective distance formula is altered to
shown in Iannelli etĀ al. (2017), where \(\beta\) and \(\mu\) are the infection and recovery rate. The mobility compound parameter \(\kappa\), representing the proportion of the circulating population, is altered to incorporate mobility reductions, given by \((1.0\textit{mobility reduction}/100.0)\times \kappa _0\), where the mobility reduction goes from 0 to \(100\%\). In this compound parameter, \(\kappa _0\) is the mobility rate, chosen to be 10%, which also ensures the logarithm is positive after the subtraction of \(\lambda\), the EulerMascheroni constant (Iannelli etĀ al. 2017). As \(\kappa _0\) is constant between the baseline and intervention scenarios, its value does not impact the infection delay value when the differences in arrival times are calculated between the two. The reproductive number \(R_0\) is chosen to be 2.9, based on an epidemiological characterisation of the MRSP early in the pandemic (de Souza etĀ al. 2020). The infectious period is chosen to be 9.2 from a mathematical analysis of COVID19 in Brazil (Pinto Neto etĀ al. 2021). The infection rate is thus \(R_0/\textit{infectious period}\) = 2.9/9.2, and the recovery rate is given by \(1/\textit{infectious period}\)=1/9.2. It is important to note that the transition rate \(P_{ij}\) calculation is unaltered from the traditional model. As the mobility compound parameter \(\kappa\) rises, \(d_{ij}\) decreases, indicating that i and j are effectively closer. Similarly to the traditional model, for every potential outbreak and target hexagon cell in the network, the dominant path effective distance is generated from the weighted shortest path analysis, generating a \(2599 \times 2599\) matrix of effective distances. This method is able to calculate effective distances between hexagons irrespective of whether they are directly or indirectly connected.
Two \(2599 \times 2599\) matrices of effective distances are calculated for every potential origin i and destination j, under the following mobility flow scenarios:

1.
No mobility reduction (baseline scenario).

2.
Reduction in mobility based on hexagonal isolation changes.
The first scenario assumes no interventions, where arrival times are calculated using the baseline travel pattern information (\(\textit{mobility reduction}=0\)). The second scenario assumes that hexagon cells reduce their mobility by the same amount as observed during the first wave of the pandemic, through leveraging the cellĀ phone social isolation information. To determine the extent of the mobility reduction for each region, the marginal change in social isolation from prelockdown to postlockdown is calculated. The initial isolation value for each hexagon is calculated as the mean across March 1 to March 15, the two weeks leading up to the MRSPās lockdown (Siciliano etĀ al. 2020). The lockdown isolation value for each hexagon is calculated as mean from March 16 to March 30 2020, capturing the initial regional responses to lockdown measures. After determining the marginal change in real isolation for each origin hexagon, the effective distance calculation becomes:
then used to calculate the dominant path effective distance between all i,Ā j nodes.
This representation of effective distance is used to approximate how rapidly a disease would spread from hexagon i to j given the observed change in pandemic isolation for region i. The adjustment of the compound \(\kappa\) term to \(\kappa ^{\textit{mobility reduction}}_i\) is a novel contribution of the study, allowing the analysis to capture heterogeneous changes in mobility based on cellĀ phone mobility data, known to intersect with socioeconomic vulnerability in the MRSP (Li etĀ al. 2021).
Infection delay of intervention
To generate an estimation of arrival times based on the effective distances, this paper employs the methods used in Iannelli etĀ al. (2017), dividing the effective distance by the effective velocity, defined as \(V^{EF} \approx \beta \mu\), where \(\beta\) is the infection rate and \(\mu\) is the recovery rate. The arrival time for a disease to arrive from location i to location j, including both the dominant path effective distance \(D_{ij}\) (sum of shortest effective distance path from i to j) and velocity is thus:
Having generated the arrival times under both scenarios for every i,Ā j combination, the infection delay by an intervention for an introductory case arriving from origin i to destination j is calculated as:
The infection delay (\(ID_{ij}\)) values are calculated for every pair of hexagon cells, generating a \(2599 \times 2599\) matrix where each i,Ā j value represents the additional time to a case arriving from i to j given a mobilityĀ reduction proportional to iās real mobility change.
Using known changes in mobility to understand intervention effectiveness takes into account the inequality in regional responses, and allows intervention scenarios to mimic the real capacities of hexagon cells to isolate and adhere to policy guidelines. Having the arrival times in \(T^{\textit{intervention}}_{ij}\) reflecting the real mobility changes allows for an infection delay analysis which better captures the lived experience of each of the 2599 hexagon cells in determining the relative benefits from early interventions.
From the MRSPās first case of COVID19 to its widespread presence, this analysis determines the time āaddedā until a regionās first case (infection delay) by an intervention at every hypothetical time t, assuming no intervention before t. At time \(t=0\), only the initial outbreak location \(i_0\) has the disease, and each hexagonās infection delay by an intervention is \(ID^0_{ij_0}\), representing the change in intervention arrival time relative to the baseline arrival time from \(i_0\) to j. For every \(t \ge 1\), each hexagon cellās infection delay value is determined based on the currently infected regions. To calculate this value, for every hexagon cell j and discrete time step t, the following algorithm is developed:

1.
Determine all infected hexagon cells at time t.

2.
Determine the infection delay of an intervention across all currently infected hexagon cells relative to destination j.

3.
Select the minimum infection delay value.
Following this algorithm, the IDM generates a timeseries infection delay curve. An example plot is presented in Fig.Ā 5, for a given hexagon A and outbreak location B. There are two primary factors that interact to create the structure of the infection delay curve: (1) the effective distance of infected hexagon cells to the hexagon cell of interest; (2) the degree of mobility reduction of infected hexagon cells. The outbreaks used in this analysis will be simulations calculated from a compartmental epidemiological model.
The example in Fig. 5 shows how the IDM can be used to estimate the time āaddedā to all regionsā first cases in a scenario with a specific outbreak location B, known a priori. To generalize the findings of the infection delay analysis to outbreak scenarios other than those observed during COVID19, epidemic outbreaks are simulated beginning in each of the 2599 hexagons in the MRSP. This paper uses a commuter susceptibleinfectedremoved (SIR) model to simulate the spread of the disease, where members of the population progress from susceptible, to infected, to removed ācompartmentsā (Salimipour etĀ al. 2023). These models have been used in numerous studies with mobility networks to explore disease risk in relation to COVID19 (Chang etĀ al. 2020; Goel etĀ al. 2021; Ajbar etĀ al. 2021; Salimipour etĀ al. 2023). This paper focuses on the initial outbreak of the disease in a short time interval, where SIR models have been shown as an effective predictor despite difficulty forecasting epidemic spread in the longer term (Moein etĀ al. 2021). This paper employs the commuteroriented susceptibleinfectedremoved (SIR) model used in Schlosser etĀ al. (2021), on GitHub as EpiCommute. While the original model is used to simulate the spread of COVID19 in 401 German counties, this analysis uses the 2599 social isolation hexagons, providing their interpolated populations and mobility flows.
For each outbreak scenario, the calculated arrival times are used in conjunction with the IDM to generate infection delay curves for every hexagon cell. The end result is 2598 infection delay curves for every hexagon (excluding its own outbreak), each one encapsulating the infection delay to the first case by an intervention at every time t.
Median infection delay values
To extract key information from each hexagon cellās 2598 infection delay curves, the median value taken over the first 10 days is used to summarize the curve describing infection delay from an intervention. The first 10 days are chosen as they best exemplify the differences in infection delays across early outbreak scenarios, after which the curves begin to converge. FigureĀ 6 displays the pipeline for calculating median infection delay curves for each hexagon cell. Rather than assigning every infection delay curve an equal weight and assuming that each scenario is equally likely, each curve is weighted by the indegree centrality of its outbreak location, thus resulting in a weighted median value of the infection delay. In the first set of results, each hexagon cell is divided into centrality and income quartiles, and their relationships to infection delays are explored. A oneway ANOVA test is performed on the infection delay values to determine whether the differences are statistically significant. In the second set of results, each hexagon cellās 2598 infection delay curves are divided into two groups based on the indegree centrality of the outbreak location. A studentās ttest is performed on the two groups of infection delay values to test whether the differences are statistically significant.
Results
Weighted median infection delay curve
The relationship between greater centrality and lower infection delay values is displayed in Table 3 and Fig.Ā 7. Within every income quartile, greater centrality is associated with a lower median infection delay value. These differences between infection delay values across centrality quartiles, controlling for income quartile, are statistically significant at the \(p<0.01\) level based on the oneway ANOVA test. FigureĀ 8 shows the geographic distribution of weighted median infection delay values.
FigureĀ 9 displays the distribution of infection delay curves across income groups, controlling for their levels of centrality. Observing the hexagon cellsā infection delays from Fig.Ā 9, this analysis finds no discernable trend across income groups. The median infection delay values of hexagon cells in the bottom 25% of centrality are between 7.5 and 8 days. Hexagon cells in the highest centrality quartile all have median infection delay values between 6 and 6.5 days.
Division by outbreak location centrality
Each hexagon cellās infection delay value is subsequently calculated and shown when the outbreak location is in the bottom versus top 50% of centrality. For every hexagon cell, this creates two infection delay values, shown sidebyside in Figs.Ā 10 and 11. We see that greater centrality is associated with lower infection delays, irrespective of income, and no clear pattern across income groups is observed when controlling for centralityāsimilarly to Figs.Ā 7 and 9. The results also show that irrespective of the income and centrality grouping, outbreaks beginning in hexagon cells of lower centrality lead to greater infection delays of lockdowns. The studentās ttest indicates a statistically significant (\(p<0.01\)) difference between infection delay values depending on whether the outbreak locationās centrality is below of above the median.
Discussion
This analysis has sought to uncover how the socioeconomic and network characteristics of a region relate to the delay of its first case from an early intervention. The results of the Infection Delay Model indicate that the centrality of a region, independent of its income level, plays the largest role in determining how an early intervention will delay their first infection. There is no discernable relationship between income levels and the ability of a lockdown to slow the arrival of disease when controlling for centrality. This is surprising, considering that previous research using the same mobility dataset has shown that lower income individuals were less able to reduce their mobility after SĆ£o Pauloās lockdown (Li etĀ al. 2021). Although previous studies have shown that vulnerable communities with lower isolation levels have higher infection rates of COVID19 (Lee etĀ al. 2021; Li etĀ al. 2021; Cordes and Castro 2020), our results suggest that the influence of socioeconomic and isolation inequalities in determining disease arrival is overridden by the outsized influence of centrality in the network. As an effective distancebased analysis, more central regions tend, on average, to be ācloserā to infected regions. This proximity reduces the potential infection delay of a lockdown, with an opposite mechanism in play for less central regions.
A potential reason why income does not have a clear impact on infection delay values, when controlling for centrality, is that socioeconomic dynamics can be already embedded in the network topology. In Brazil, these dynamics have been shown to be at play, as lowerincome regions face larger average commuting times (Pereira and Schwanen 2015)āa factor which would already be embedded in the commuter travel network used in this analysis. This study is not concluding that income does not have an effect on disease spreadāpriorly shown to do so in our region of studyābut that in the IDM, income effects that are not already entangled with the network topology do not influence the delay to regionsā first cases caused by an intervention.
The literature produced during the COVID19 pandemic has thoroughly highlighted the importance of socioeconomic factors and their relationship to disease risk, rationalizing their use as more than a passive addon to networkbased results. The growing prioritization of socioeconomic inequalities as a driving force of disease risk is exemplified in studies such as Nande etĀ al. (2021), who study how eviction rates in Philadelphia have a measurable impact on the spread of COVID19. The Infection Delay Model reflects socioeconomic inequalities in the MRSP by incorporating reallife mobility reductionsāknown to be weaker in vulnerable areas (Li etĀ al. 2021)āas a core component in the effective distance network analysis. Income is then used as a key axis to explore infection delays, found to be overpowered by a regionās centrality.
Rather than contradicting existing literature on the health burden inequalities associated with socioeconomic status, this paper uncovers an unexplored perspective on pandemic preparedness. The emphasis of previous literature on case, death, and hospitalization counts illuminate how vulnerable groups are most at risk during the course of an outbreak (Li etĀ al. 2021; Rocha etĀ al. 2021; Jay etĀ al. 2020; Lee etĀ al. 2021; Cordes and Castro 2020; Pereira etĀ al. 2021; Coelho etĀ al. 2020). This paper targets a different, interventionfocused question: How much time can be gained to a regionās first case from an early lockdown? It cannot be assumed that the same mechanisms leading to greater disease risk during an outbreak lead to reduced intervention effectiveness prior to an outbreak. Our results, in conjunction with the established literature on socioeconomic vulnerability and COVID19, illuminate an additional burden faced by lowincome, centrally located regions.
A major contribution of this study is its generalized, forwardlooking characterisation of intervention effectiveness. Rather than relying on a single set of initial conditions when modelling a disease, or using a subset of transport hubs as outbreak locations, this analysis incorporates all possible outbreak locations when assessing how early interventions lead to infection delays. This allows for broad understandings of intervention effectiveness whose validity is not reliant on the next epidemic beginning in the same location as the last. This addresses the recently explored importance of outbreak locations on disease trajectories, providing generalizable insights for future disease preparedness (Schlosser etĀ al. 2021). We are able to use the abundance of scenarios to generate weighted median infection delay values (Figs.Ā 7, 9), emphasizing the dominant role of centrality. Further, we can divide outbreak locations into low and high centrality groups (Figs.Ā 10, 11), and show that the infection delays of interventions vary based on the centrality of the outbreak location. We see that irrespective of the income or centrality quartile of recipient regions, outbreaks beginning in less central regions tend to lead to greater slowdowns.
Conclusion
Research into the effectiveness of government interventions to slow disease spread is essential, as the disaster resulting from the COVID19 pandemic and its emerging new variants continues globally. The novel Infection Delay Model proposed in this study provides a method of capturing how mobility reductions can slow the spread of an outbreak while considering the network patterns of mobility flows, an important element of intervention effectiveness. The datalinkage approach, interpolating travel behaviour and socioeconomic data, allowed for insights into the social context of regions and how interventions can delay a regionās first case. The unique integration of cellĀ phone mobility data into the effective distance metrics has captured heterogeneous changes in isolation, found in prior literature to intersect with socioeconomic inequalities (Li etĀ al. 2021; Lee etĀ al. 2021). While this analysis is focused on Brazil, a region where income, health, and transport inequalities are stark (Malta etĀ al. 2020), the presented approach can be applied in other regions to observe the intersection of intervention effectiveness, centrality, and socioeconomic vulnerability. Similarly, the epidemiological parameters in this analysis are chosen to mimic COVID19, but a novel variant or diseaseās reproduction rate and infectious period could be used as substitutes. Adopting interdisciplinary methodologies to investigate the effectiveness of interventions, with a focus on exploring inequalities, may provide novel insights into the factors driving the unequal playing field exposed during the COVID19 pandemic.
Limitations and future directions
Based on the Infection Delay Model algorithm, the delays calculated for a given region are dependent on the reductions of mobility flows that arrive to it, rather than its own mobility reduction. This operates well under a regime where first cases arrive from individuals travelling from other locations. Advancements of the Infection Delay Model which capture how a first disease introduction to a region can originate from one of its residents travelling elsewhere would capture an important dimension of disease transmission. This may lead socioeconomic and isolation inequalities to play a stronger role in shaping infection delay curves. Further, rather than calculating the time to a regionās first case, a case threshold such as 5% infectionrate of the population could be implemented, in which case a regionās own social isolation capabilities would more directly impact its infection delay value. These adaptations of the Infection Delay Model can expand its scope in capturing the concept of intervention effectiveness, as its current focus on the delay to a regionās first case is only one important element.
When considering cellĀ phone data sources, originally collected for commercial purposes, coverage bias should be noted. As a cellĀ phone analytics company, the sample of users in the Inloco/Incognia data set is determined by their market share, rather than an emphasis on representative samples (Tizzoni etĀ al. 2014). The nearglobal ubiquity of cellĀ phones does not preclude biases, as possession and use rates vary across demographic and income groups (Kraemer etĀ al. 2020). The elderly are often underrepresented in such samples, while educated urban males are overrepresented relative to lowerincome individuals (Kraemer etĀ al. 2020).
In a preliminary analysis, a modified radiation model was used to determine if the results using real commuting data could be replicated with a generalized model. When observing the outbreak locations which led to above and below average infection delays for the rest of the MRSP, the radiation network overstated the influence of incomerelated mobility reductions relative to centrality. This may have occurred because the radiation model failed to replicate regional hubs with disproportionately large connectivity throughout the commuting network. This caveat should be considered for future research using effective distancebased metrics on artificially generated commuting data.
The suitability of integrating traditional household travel survey data with the aggregated social isolation cellĀ phone data deserves exploration by future research. This study recommends comparing granular cellĀ phone mobility location pairs, and observing how daily travel patterns and their changes after lockdown resemble those found in this paperās analysis. If providing similar results, the greater anonymity of the aggregated social isolation data may render it a more readily accessible and minimally invasive method for granular mobilityrelated studies.
Availability of data and materials
The datasets and code supporting the conclusions of this article are both openly available. Data sets are stored as a Zenodo repository here: https://doi.org/10.5281/zenodo.5947174. The code is stored as a GitHub repository (Python/R) here: https://github.com/shivyucel/infectiondelayproject, and archived at time of submission as a Zenodo repository here:Ā https://doi.org/10.5281/zenodo.6008499.
Abbreviations
 NPI:

Nonpharmaceutical intervention
 MRSP:

Metropolitan Region of SĆ£o Paulo
 IDM:

Infection Delay Model
References
Ajbar A, Alqahtani RT, Boumaza M (2021) Dynamics of an SIRbased COVID19 model with linear incidence rate, nonlinear removal rate, and public awareness. Front Phys 9:634251. https://doi.org/10.3389/fphy.2021.634251
Aleta A, MartĆnCorral D, Bakker MA, Pastore y Piontti A, Ajelli M, Litvinova M, Chinazzi M, Dean NE, Halloran ME, Longini Jr IM, Pentland A (2022) Quantifying the importance and location of SARSCoV2 transmission events in large metropolitan areas. Proceed Nat Acad Sci 119(26):e2112182119
Balcan D, Colizza V, GonĆ§alves B, Hud H, Ramasco JJ, Vespignani A (2009) Multiscale mobility networks and the spatial spreading of infectious diseases. Proc Natl Acad Sci USA 106(51):21484ā21489. https://doi.org/10.1073/pnas.0906910106
Bonaccorsi G, Pierri F, Cinelli M, Flori A, Galeazzi A, Porcelli F, Schmidt AL, Valensise CM, Scala A, Quattrociocchi W et al (2020) Economic and social consequences of human mobility restrictions under Covid19. Proc Natl Acad Sci 117(27):15530ā15535
Brockmann D, Helbing D (2014) The hidden geometry of complex, networkdriven contagion phenomena (Science (1337)). Science 343(6172):730. https://doi.org/10.1126/science.343.6172.730c
Brodsky I (2018) H3: Uberās Hexagonal Hierarchical Spatial Index. https://eng.uber.com/h3/
Candido DS, Claro IM, de Jesus JG, Souza WM, Moreira FRR, Dellicour S, Mellan TA, du Plessis L, Pereira RHM, Sales FCS, Manuli ER, ThĆ©zĆ© J, Almeida L, Menezes MT, Voloch CM, Fumagalli MJ, Coletti TM, da Silva CAM, Ramundo MS, Amorim MR, Hoeltgebaum HH, Mishra S, Gill MS, Carvalho LM, Buss LF, Prete CA, Ashworth J, Nakaya HI, Peixoto PS, Brady OJ, Nicholls SM, Tanuri A, Rossi ĆD, Braga CKV, Gerber AL, de GuimarĆ£es APC, Gaburo N, Alencar CS, Ferreira ACS, Lima CX, Levi JE, Granato C, Ferreira GM, Francisco RS, Granja F, Garcia MT, Moretti ML, Perroud MW, CastiĆ±eiras TMPP, Lazari CS, Hill SC, de Souza Santos AA, Simeoni CL, Forato J, Sposito AC, Schreiber AZ, Santos MNN, de SĆ” CZ, Souza RP, ResendeMoreira LC, Teixeira MM, Hubner J, Leme PAF, Moreira RG, Nogueira ML, Ferguson NM, Costa SF, ProencaModena JL, Vasconcelos ATR, Bhatt S, Lemey P, Wu CH, Rambaut A, Loman NJ, Aguiar RS, Pybus OG, Sabino EC, Faria NR (2020) Evolution and epidemic spread of SARSCoV2 in Brazil. Science 369(6508):1255ā1260. https://doi.org/10.1126/SCIENCE.ABD2161
Chang S, Pierson E, Koh PW, Gerardin J, Redbird B, Grusky D, Leskovec J (2020) Mobility network models of COVID19 explain inequities and inform reopening. Nature. https://doi.org/10.1038/s4158602029233
Christley RM, Pinchbeck GL, Bowers RG, Clancy D, French NP, Bennett R, Turner J (2005) Infection in social networks: using network analysis to identify highrisk individuals. Am J Epidemiol 162(10):1024ā1031. https://doi.org/10.1093/aje/kwi308
Coelho FC, Lana RM, Cruz OG, Villela DAM, Bastos LS, Pastore Y, Piontti A, Davis JT, Vespignani A, CodeĆ§o CT (2020) Gomes MFC Assessing the spread of COVID19 in Brazil: mobility, morbidity and social vulnerability. PLoS ONE 15(9 September):1ā11. https://doi.org/10.1371/journal.pone.0238214
Comber A, Zeng W (2019) Spatial interpolation using areal features: a review of methods and opportunities using new forms of data with coded illustrations. Geogr Compass 13(10):12465. https://doi.org/10.1111/gec3.12465
Cordes J, Castro MC (2020) Spatial analysis of COVID19 clusters and contextual factors in New York City. Spat Spatiotemporal Epidemiol 34:100355. https://doi.org/10.1016/j.sste.2020.100355
de Souza WM, Buss LF, Candido DS, Carrera JP, Li S, Zarebski AE, Pereira RHM, Prete CA, de SouzaSantos AA, Parag KV, Belotti MCTD, VincentiGonzalez MF, Messina J, da Silva Sales FC, Andrade PS, Nascimento VH, Ghilardi F, Abade L, Gutierrez B, Kraemer MUG, Braga CKV, Aguiar RS, Alexander N, Mayaud P, Brady OJ, Marcilio I, Gouveia N, Li G, Tami A, de Oliveira SB, Porto VBG, Ganem F, de Almeida WAF, Fantinato FFST, MacĆ”rio EM, de Oliveira WK, Nogueira ML, Pybus OG, Wu CH, Croda J, Sabino EC, Faria NR (2020) Epidemiological and clinical characteristics of the COVID19 epidemic in Brazil. Nat Hum Behav 4(8):856ā865. https://doi.org/10.1038/s4156202009284
Ferreira CP, Marcondes D, Melo MP, Oliva SM, Peixoto CM, Peixoto PS (2021) A snapshot of a pandemic: the interplay between social isolation and COVID19 dynamics in Brazil. Patterns. https://doi.org/10.1016/j.patter.2021.100349
Flaxman S, Mishra S, Gandy A, Unwin HJT, Mellan TA, Coupland H, Whittaker C, Zhu H, Berah T, Eaton JW, Monod M, PerezGuzman PN, Schmit N, Cilloni L, Ainslie KEC, Baguelin M, Boonyasiri A, Boyd O, Cattarino L, Cooper LV, CucunubĆ” Z, CuomoDannenburg G, Dighe A, Djaafara B, Dorigatti I, van Elsland SL, FitzJohn RG, Gaythorpe KAM, Geidelberg L, Grassly NC, Green WD, Hallett T, Hamlet A, Hinsley W, Jeffrey B, Knock E, Laydon DJ, NedjatiGilani G, Nouvellet P, Parag KV, Siveroni I, Thompson HA, Verity R, Volz E, Walters CE, Wang H, Wang Y, Watson OJ, Winskill P, Xi X, Walker PGT, Ghani AC, Donnelly CA, Riley S, Vollmer MAC, Ferguson NM, Okell LC, Bhatt S, Team ICCR (2020) Estimating the effects of nonpharmaceutical interventions on COVID19 in Europe. Nature 584(7820):257ā261. https://doi.org/10.1038/s4158602024057
Francetic I, Munford L (2021) Corona and coffee on your commute: a spatial analysis of COVID19 mortality and commuting flows in England in 2020. Eur J Pub Health. https://doi.org/10.1093/eurpub/ckab072
Gautreau A, Barrat A, BarthĆ©lemy M (2008) Global disease spread: statistics and estimation of arrival times. J Theor Biol 251(3):509ā522. https://doi.org/10.1016/j.jtbi.2007.12.001. arXiv:0801.1846
Gauvin L, Bajardi P, Pepe E, Lake B, Privitera F, Tizzoni M (2021) Socioeconomic determinants of mobility responses during the first wave of Covid19 in Italy: from provinces to neighbourhoods. J R Soc Interface 18(181):20210092
Goel R, Bonnetain L, Sharma R, Furno A (2021) Mobilitybased SIR model for complex networks: with case study Of COVID19. Soc Netw Anal Min 11(1):105. https://doi.org/10.1007/s13278021008143
Gozzi N, Tizzoni M, Chinazzi M, Ferres L, Vespignani A, Perra N (2021) Estimating the effect of social inequalities on the mitigation of Covid19 across communities in Santiago de Chile. Nat Commun 12(1):1ā9
Grantz KH, Meredith HR, Cummings DAT, Metcalf CJE, Grenfell BT, Giles JR, Mehta S, Solomon S, Labrique A, Kishore N, Buckee CO, Wesolowski A (2020) The use of mobile phone data to inform analysis of COVID19 pandemic epidemiology. Nat Commun 11(1):1ā8. https://doi.org/10.1038/s41467020181905
Heroy S, Loaiza I, Pentland A, OāClery N (2022) COVID19 policy analysis: labour structure dictates lockdown mobility behaviour. J R Soc Interface 18(176):20201035. https://doi.org/10.1098/rsif.2020.1035
Hunter E, Namee BM, Kelleher JD (2020) A model for the spread of infectious diseases in a region. Int J Environ Res Public Health 17(9):3119. https://doi.org/10.3390/ijerph17093119
Iannelli F, Koher A, Brockmann D, HĆ¶vel P, Sokolov IM (2017) Effective distances for epidemics spreading on complex networks. Phys Rev E 95(1ā1):12313. https://doi.org/10.1103/PhysRevE.95.012313
Incognia (2020) PolĆtica de Privacidade COVID19. https://www.incognia.com/pt/politicas/covid
Instituto Brasileiro de Geografia e EstatĆstica: Demographic Census 2010 (2010). https://www.ibge.gov.br/estatisticas/sociais/populacao/9662censodemografico2010.html
Instituto Brasileiro de Geografia e EstatĆstica: DIĆRIO OFICIAL DA UNIĆO, RESOLUĆĆO \({\text{N}}^{\underline{{\rm o}}}\) 3, DE 26 DE AGOSTO DE 2019 (2019). https://www.in.gov.br/web/dou//resolucaon3de26deagostode2019212912572
Instituto Brasileiro de Geografia e EstatĆstica: Sinopse de Censo DemogrĆ”fico 2010 (2010). https://censo2010.ibge.gov.br/sinopse/index.php
Jang W, Yao X (2011) Interpolating spatial interaction data 1: interpolating spatial interaction data. Trans GIS 15(4):541ā555. https://doi.org/10.1111/j.14679671.2011.01273.x
Jay J, Bor J, Nsoesie EO, Lipson SK, Jones DK, Galea S, Raifman J (2020) Neighbourhood income and physical distancing during the COVID19 pandemic in the United States. Nat Hum Behav 4(12):1294ā1302. https://doi.org/10.1038/s41562020009982
Kraemer MUG, Sadilek A, Zhang Q, Marchal NA, Tuli G, Cohn EL, Hswen Y, Perkins TA, Smith DL, Reiner RC, Brownstein JS (2020) Mapping global variation in human mobility. Nat Hum Behav 4(8):800ā810. https://doi.org/10.1038/s4156202008750
Kuo PF, Chiu CS (2021) Airline transportation and arrival time of international disease spread: a case study of Covid19. PLoS ONE 16(8):0256398
Lee WD, Qian M, Schwanen T (2021) The association between socioeconomic status and mobility reductions in the early stage of Englandās COVID19 epidemic. Health Place 69:102563. https://doi.org/10.1016/j.healthplace.2021.102563
Levin MW, Shang M, Stern R (2021) Effects of shortterm travel on COVID19 spread: a novel SEIR model and case study in Minnesota. PLoS ONE 16(1):1ā16. https://doi.org/10.1371/journal.pone.0245919
Li SL, Pereira RHM, Prete CA Jr, Zarebski AE, Emanuel L, Alves PJH, Peixoto PS, Braga CKV, de Souza Santos AA, de Souza WM, Barbosa RJ, Buss LF, Mendrone A, de AlmeidaNeto C, Ferreira SC, Salles NA, Marcilio I, Wu CH, Gouveia N, Nascimento VH, Sabino EC, Faria NR, Messina JP (2021) Higher risk of death from COVID19 in lowincome and nonWhite populations of SĆ£o Paulo, Brazil. BMJ Glob Health. https://doi.org/10.1136/bmjgh2021004959
LĆ¼ L, Chen D, Ren XL, Zhang QM, Zhang YC, Zhou T (2016) Vital nodes identification in complex networks. Phys Rep 650:1ā63. https://doi.org/10.1016/j.physrep.2016.06.007
Malta M, Murray L, da Silva CMFP, Strathdee SA (2020) Coronavirus in Brazil: the heavy weight of inequality and unsound leadership. EClinicalMedicine 25:100472. https://doi.org/10.1016/j.eclinm.2020.100472
Meghanathan N (2021) Neighborhoodbased bridge node centrality tuple for complex network analysis. Appl Netw Sci 6(1):47. https://doi.org/10.1007/s41109021003881
Meo SA, Abukhalaf AA, Alomar AA, AlMutairi FJ, Usmani AM, Klonoff DC (2020) Impact of lockdown on COVID19 prevalence and mortality during 2020 pandemic: observational analysis of 27 countries. Eur J Med Res 25(1):56. https://doi.org/10.1186/s40001020004569
METRĆSP: Pesquisa Origem Destino 2017: a mobilidade urbana da RegiĆ£o Metopolitana de SĆ£o Paulo em detalhes (2018). http://www.metro.sp.gov.br/pesquisaod/arquivos/Ebook%20Pesquisa%20OD%202017_final_240719_versao_4.pdf
Moein S, Nickaeen N, Roointan A, Borhani N, Heidary Z, Javanmard SH, Ghaisari J, Gheisari Y (2021) Inefficiency of SIR models in forecasting COVID19 epidemic: a case study of Isfahan. Sci Rep 11(1):4725. https://doi.org/10.1038/s41598021840556
Nande A, Sheen J, Walters EL, Klein B, Chinazzi M, Gheorghe AH, Adlam B, Shinnick J, Tejeda MF, Scarpino SV, Vespignani A, Greenlee AJ, Schneider D, Levy MZ, Hill AL (2021) The effect of eviction moratoria on the transmission of SARSCoV2. Nat Commun 12(1):2274. https://doi.org/10.1038/s41467021225215
Oka T, Wei W, Zhu D (2021) The effect of human mobility restrictions on the COVID19 transmission network in China. PLoS ONE 16(7):1ā16. https://doi.org/10.1371/journal.pone.0254403
Oraby T, Tyshenko MG, Maldonado JC, Vatcheva K, Elsaadany S, Alali WQ, Longenecker JC, AlZoughool M (2021) Modeling the effect of lockdown timing as a COVID19 control measure in countries with differing social contacts. Sci Rep 11(1):3354. https://doi.org/10.1038/s41598021828732
Peixoto PS, Marcondes D, Peixoto C, Oliva SM (2020) Modeling future spread of infections via mobile geolocation data and population dynamics. An application to COVID19 in Brazil. PLoS ONE 15(7 July):1ā23. https://doi.org/10.1371/journal.pone.0235732
Pereira R, GonĆ§alves C (2019) geobr: loads shapefiles of official spatial data sets of brazil. GitHub repository
Pereira RHM, Schwanen T (2015) Commute time in brazil (1992ā2009): differences between metropolitan areas, by income levels and gender. Technical report, discussion paper
Pereira RHM, Vieira Braga CK, Servo LM, Serra B, Amaral P, Gouveia N, Paez A (2021) Geographic access to COVID19 healthcare in Brazil using a balanced float catchment area approach. Soc Sci Med 273(February):113773. https://doi.org/10.1016/j.socscimed.2021.113773
Pinto Neto O, Kennedy DM, Reis JC, Wang Y, Brizzi ACB, Zambrano GJ, de Souza JM, Pedroso W, de Mello Pedreiro RC, de Matos Brizzi B, Abinader EO, ZĆ¢ngaro RA (2021) Mathematical model of COVID19 intervention scenarios for SĆ£o PauloBrazil. Nat Commun 12(1):418. https://doi.org/10.1038/s4146702020687y
Pullano G, Valdano E, Scarpa N, Rubrichi S, Colizza V (2020) Evaluating the effect of demographic factors, socioeconomic factors, and risk aversion on mobility during the COVID19 epidemic in France under lockdown: a populationbased study. Lancet Digit Health 2(12):638ā649. https://doi.org/10.1016/S25897500(20)302430
Rajeh S, Savonnet M, Leclercq E, Cherifi H (2022) Comparative evaluation of communityaware centrality measures. Qual Quant. https://doi.org/10.1007/s11135022014167
Rocha R, Atun R, Massuda A, Rache B, Spinola P, Nunes L, Lago M, Castro MC (2021) Effect of socioeconomic inequalities and vulnerabilities on healthsystem preparedness and response to COVID19 in Brazil: a comprehensive analysis. Lancet Glob Health 9(6):782ā792. https://doi.org/10.1016/S2214109X(21)000814
Salimipour A, Mehraban T, Ghafour HS, Arshad NI, Ebadi MJ (2023) SIR model for the spread of COVID19: a case study. Oper Res Perspect 10:100265. https://doi.org/10.1016/j.orp.2022.100265
Schlosser F, Brockmann D (2021) Finding disease outbreak locations from human mobility data. EPJ Data Sci 10(1):52. https://doi.org/10.1140/epjds/s13688021003066
Schlosser F, Maier BF, Jack O, Hinrichs D, Zachariae A, Brockmann D (2021) COVID19 lockdown induces diseasemitigating structural changes in mobility networks. Proc Natl Acad Sci USA 117(52):32883ā32890. https://doi.org/10.1073/PNAS.2012326117
Siciliano B, Carvalho G, da Silva CM, Arbilla G (2020) The impact of COVID19 partial lockdown on primary pollutant concentrations in the atmosphere of Rio de Janeiro and SĆ£o Paulo Megacities (Brazil). Bull Environ Contam Toxicol 105(1):2ā8. https://doi.org/10.1007/s00128020029079
Tao R, Downs J, Beckie TM, Chen Y, McNelley W (2020) Examining spatial accessibility to COVID19 testing sites in Florida. Ann GIS 26(4):319ā327. https://doi.org/10.1080/19475683.2020.1833365
Tizzoni M, Bajardi P, Decuyper A, Kon Kam King G, Schneider CM, Blondel V, Smoreda Z, GonzĆ”lez MC, Colizza V (2014) On the use of human mobility proxies for modeling epidemics. PLoS Comput Biol. https://doi.org/10.1371/journal.pcbi.1003716
Valdano E, Lee J, Bansal S, Rubrichi S, Colizza V (2021) Highlighting socioeconomic constraints on mobility reductions during COVID19 restrictions in France can inform effective and equitable pandemic response. J Travel Med 28(4):045. https://doi.org/10.1093/jtm/taab045
Acknowledgements
We would like to thank Inloco/Incognia, for providing access to valuable privately owned mobility data.
Funding
PSP was supported by the SĆ£o Paulo Research Foundation (FAPESP) under grant number 2021/061760.
Author information
Authors and Affiliations
Contributions
All authors conceived and designed the study. SGY implemented the models, carried out the analysis and wrote the first draft. RHMP and PSP performed data curation and contributed with the formal analysis. CQC performed the initial analyses and coordinated the project. All authors discussed, edited, and reviewed the manuscript, and gave final approval for publication. All authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare that they have no competing interests.
Additional information
Publisherās Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
YĆ¼cel, S.G., Pereira, R.H.M., Peixoto, P.S. et al. Impact of network centrality and income on slowing infection spread after outbreaks. Appl Netw Sci 8, 16 (2023). https://doi.org/10.1007/s4110902300540z
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s4110902300540z