Estimating the effect of social inequalities on the mitigation of COVID-19 across communities in Santiago de Chile

The COVID-19 pandemic disproportionately affected Santiago, Chile, becoming one of the largest urban outbreaks globally. While initial NPIs (mid-March 2020, including school closures and gathering bans) failed to contain the spread, a full lockdown was implemented on May 15th. This study aimed to provide a data-driven analysis of the epidemic's unfolding and assess the impact of NPIs. The research uses a unique dataset of anonymized mobile phone data from 1.4 million users (approximately 22% of Santiago's population), provided by Telefónica Movistar. This data, combined with official surveillance data, allowed for the creation of a spatially and age-structured epidemic model to characterize changes in mobility and contacts over time, and to model the spread of the virus across 37 comunas. Understanding the impact of interventions across different socioeconomic groups within Santiago is particularly crucial due to significant social and economic inequalities in the city. This investigation into a specific urban context contrasts with broader global analyses and offers insights into the effectiveness and equity of NPIs. Existing research on epidemic modeling at similar geographical scales has been conducted for other cities, including Boston, Wanzhou, New York City, and London; however, the heterogeneous nature of COVID-19 mitigation strategies across countries and regions emphasizes the importance of context-specific modeling to isolate the effects of individual NPIs and their impact when modified or reintroduced.

The authors cite numerous studies examining the impact of COVID-19 and NPIs in various locations, including work on similar geographical scales in Boston, Wanzhou, New York City and London. These studies provided a basis for comparison and highlighted the heterogeneity of responses and their effectiveness in diverse contexts. The literature review also touches upon studies examining the relationship between socioeconomic disparities and the impact of the COVID-19 pandemic. Research highlighted the unequal effects of NPIs, showing that higher income populations have a greater capacity to afford social distancing. Studies from the United States, France, Italy, and other countries informed the research design by illustrating how socioeconomic factors influence both the implementation and impact of government-mandated mobility restrictions.

The study used anonymized mobile phone data (XDRs) from 1.4 million users to track mobility patterns across 37 comunas in Santiago, Chile. The data, spanning February 27th to June 1st, 2020, tracked device interactions with antennas, enabling the researchers to estimate the location of the devices and their movements between comunas. A home antenna was assigned to each device based on nighttime activity. Three phases were identified: before interventions (baseline), partial lockdown (March 16th – May 15th), and full lockdown (after May 15th). Mobility was measured by calculating the fraction of devices traveling between comunas daily, averaged for each phase. While direct contact data was unavailable, the researchers estimated contacts reduction by analyzing the variation in co-located users within each antenna, assuming homogeneous mixing within each comuna. A stochastic mechanistic metapopulation epidemic model, similar to GLEAM, was developed. This model considered comunas as distinct subpopulations with individuals divided into 16 age brackets based on Chilean demographics and a contact matrix reflecting the mixing rates between different age groups. An SLIR (Susceptible, Latent, Infectious, Removed) compartmentalization was used. The model incorporated mobility data to define the effective coupling between comunas using a time-scale separation technique, approximating the impact of mobility on transmission without explicit modeling of individual movements. Government interventions (partial and full lockdowns) were integrated by adjusting the mobility network and contacts reduction parameters in the model. Approximate Bayesian Computation (ABC) rejection sampling was used to calibrate the model to official weekly death data, estimating parameters such as the basic reproduction number (R0) and the delay in deaths. The model's performance was evaluated based on the median absolute percentage error in predicting weekly deaths. Socioeconomic differences were assessed using the Human Development Index (HDI), calculated using census data from 2013-2015, and considering alternative socioeconomic indicators as a robustness check.

The study's key findings center around the impact of NPIs, the underreporting of infections, and the heterogeneous effect of the pandemic across socioeconomic groups. The model estimated a basic reproduction number (R0) of 2.66 (95% CI: [2.58, 2.72]). The full lockdown, while inducing a modest additional decrease in mobility (17%) compared to the initial NPIs, proved crucial in bringing R below 1 and curbing the epidemic. This result underscored the critical threshold effects that characterize epidemic dynamics. The full lockdown prevented an estimated 34.7% (95% CI: [27.2%, 44.1%]) increase in total deaths. Delaying the lockdown by a week could have led to an 18.1% (95% CI: [6.0%, 34.0%]) more intense peak in incidence. The model suggested a significant underreporting of infections, estimating an infected fraction of 38.7% (95% CI: [35.1%, 41.6%]) on August 1st, 2020, ten times higher than official figures. A strong negative correlation was found between the HDI of comunas and both the attack rate and deaths per 1000 inhabitants, supporting the observation that wealthier comunas experienced significantly smaller outbreaks and lower mortality rates. The study compared the main model to simpler alternatives (neglecting inter-comuna mobility or treating Santiago as a single population), demonstrating the superiority of the spatially explicit model in capturing the heterogeneous spread. A counterfactual simulation, applying the average mobility reduction of the top 25% of comunas (in terms of HDI) to all comunas, revealed that a uniform reduction in mobility could have resulted in substantially fewer cases and deaths, illustrating the significant role of inequality in the pandemic's impact. The evolution of the effective reproduction number (Rt) demonstrated that the partial lockdown only slowed the epidemic (Rt remained above 1), while the full lockdown was critical in pushing Rt below 1 and effectively containing the spread.

The study's findings strongly suggest a correlation between socioeconomic status and COVID-19 outcomes in Santiago. While lockdowns are effective, they can exacerbate existing inequalities by disproportionately affecting disadvantaged communities. The higher mobility reduction in wealthier comunas, correlated with HDI, reflects the capacity of these communities to better implement social distancing. The study highlights the limitations of relying solely on official case counts and underscores the need for more comprehensive strategies that address social determinants of health to ensure equitable pandemic responses. The findings are not only significant for understanding the Chilean context but also hold implications for policymakers in other regions facing similar disparities. The observed effectiveness of the full lockdown in Santiago, despite its modest impact on overall mobility, emphasizes the importance of timely, decisive, and potentially stronger interventions when necessary to control outbreaks effectively. The analysis also reinforces the need to consider the socioeconomic context when designing and implementing public health interventions during infectious disease outbreaks.

This study contributes to our understanding of COVID-19 spread and the effectiveness of NPIs in a context of significant socioeconomic inequality. The full lockdown was key to curbing the epidemic in Santiago, highlighting the importance of timely, strong interventions. However, the findings highlight that the benefits of such interventions are not equally shared among communities. Future research should explore targeted strategies that address health disparities and ensure equitable access to resources and information during pandemic situations. Further research could also analyze the long-term effects of the pandemic on health disparities in Santiago and explore more nuanced models capturing varying transmission risks across different locations within comunas.

The study acknowledges several limitations. The focus on the Santiago Metropolitan Area limits generalizability, as national and international importations after March 1, 2020, were not explicitly considered. The use of a relatively simple disease model and reliance on aggregate mobility measures, without specific information about points of interest, might affect accuracy. The mobile phone user sample may not perfectly represent the whole population, although it is suggested to adequately cover the socio-demographic diversity of Santiago. The dependence of mobile phone data on antenna distribution is also a potential factor affecting accuracy; however, this is mitigated by aggregating the data at the comuna level. Finally, the study acknowledges that not all mobility equally contributes to transmission risk.