Análise de estratégias de vacinação para COVID-19 baseada em redes complexas

• Ano de defesa: 2024
• Autor(a) principal: Gonçalves, Carlos Miguel Moreira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Redes (Matemática); Epidemiologia; COVID-19 (doença) - vacinação; Medidas de centralidade; Networks (Mathematics); Epidemiology; COVID-19 (disease) - Vaccination.; Centrality measures; CNPQ::CIENCIAS EXATAS E DA TERRA::PROBABILIDADE E ESTATISTICA
• Link de acesso: http://repositorio.ufc.br/handle/riufc/79918

he rapid spread, high mortality rate, and overloading of hospital systems turned the SARS CoV-2 virus into a significant challenge for humanity. The urgency of the situation required substantial investments in vaccine development as a crucial measure to mitigate this issue. However, the financial costs and time needed for vaccine development represent obstacles to achieving comprehensive vaccination coverage. Therefore, understanding the most effective strategy for population immunization is of utmost importance to contain the virus’s spread and optimize resource allocation in public policies. It is relevant to highlight that a large part of these policies was based on social distancing, which had significant impacts on economic, educational, and mental health spheres. The purpose of this project is to propose a modeling-based approach using the SEIHARDS contagion model on a complex contact network derived from a questionnaire-based survey. The objective is to identify optimal vaccination strategies to reduce the virus’s spread, minimize hospitalizations, and mitigate the number of deaths. For this, data were collected from a survey called POLYMOD, conducted in 8 European countries, aiming to map connections between people. This survey is the most aligned with the project because it holds data on the age groups of the respondents, something quite relevant in COVID-19 hospitalization and mortality. However, the respondents of this survey had no connections among themselves, making it necessary to develop an algorithm called the Stratified Configuration Model (SCM) to form a network with this database limitation that takes into account ages and adds the duration of contact time between individuals. Additionally, the proposed model manages to increase the network’s clustering because the traditional Configuration Model achieves low clustering. Moreover, by collecting data from articles and OpenDataSUS, it was possible to find the constants for the proposed epidemiological model, where some parameters vary with age group. This project analyzed the behavior of the proposed algorithm with changes in clustering, proposed centrality metrics for this study, and analyzed the behavior and sensitivity of the infection model concerning variations in clustering and network construction. It showed that the model is not very sensitive to increases in clustering and that with the insertion of weighting on the edges, there is no gain in the metrics and an increase of up to 5 times in computational time. The centrality measures that considered weights on nodes and the network structure proved to be highly effective for vaccination against COVID-19, with strategies like PageRank reducing mortality by more than 60%, total hospitalization time by 66%, and requiring vaccinating 31% of the network to inhibit the virus’s proliferation compared to the absence of vaccination. Some centralities that use weights on edges presented two calculation approaches: altruistic and individualistic. In the first, the calculation was done assuming that the individual will prevent the death of others in the network, while in the second, the individual wants to prevent their own death. In this sense, altruistic approaches proved to be better strategies in reducing mortality and hospitalization time, but the individualistic ones were better against the disease’s spread. Finally, the best-performing metrics were identified—the most effective metrics on average were PageRank and PageRank with edge weighting—as well as the Pareto frontier for the analyzed data for different metrics, networks, and clustering values.