Full text
| |
| Author(s): |
Carlos Augusto Prete Junior
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | São Paulo. |
| Institution: | Universidade de São Paulo (USP). Escola Politécnica (EP/BC) |
| Defense date: | 2024-07-31 |
| Examining board members: |
Vitor Heloiz Nascimento;
Marcelo Gomes da Silva Bruno;
Rodrigo Malavazi Corder;
Wesley Francis Costa Cota;
Sergio Muniz Oliva Filho
|
| Advisor: | Vitor Heloiz Nascimento; Ester Cerdeira Sabino |
| Abstract | |
Learning algorithms for non-linear stochastic models have garnered significant attention in the past years, driven by the success of deep neural networks in several pattern recognition applications and the extensive use of Bayesian models for SARS-CoV-2 epidemic modelling. Real time inference of key epidemiological parameters such as seroprevalence and effective reproduction number is essential for tracking emerging epidemics and predicting future outcomes. In this work, we employ signal processing techniques to model and identify non-linear stochastic systems that represent mechanisms of infectious disease spread. We present methods to estimate attack rate and metrics of disease severity using imperfect serological tests, accounting for antibody waning. These methods were applied to analyse the SARS-CoV-2 epidemic in eight Brazilian capitals using blood donation samples. We also developed methods to estimate the reinfection rate, protection against reinfection conferred by previous infections, and bounds for the attack rate. When applied to the second SARS-CoV-2 wave in Manaus, dominated by the Gamma Variant of Concern (VOC), these methods revealed increased disease severity and reinfection risk of the Gamma VOC when compared to previously circulating lineages in Manaus. Additionally, we derived a framework to estimate local effective reproduction numbers (e.g. disaggregated by lineage) using local generation intervals, thereby improving the global estimate. We also developed a Bayesian model to jointly infer effective reproduction number and generation interval, removing the need for prior knowledge of generation intervals. Finally, we introduced a model-based approach to estimate the serial interval distribution from limited number of samples, inferring it for SARS-CoV-2 in Brazil. Additionally, we showed that the distributions of generation and incubation interval can be recovered from the exact serial interval distribution, and proposed Bayesian models to infer these quantities from serial interval samples. (AU) | |
| FAPESP's process: | 19/21858-0 - Bayesian models for estimating the attack rate of epidemics |
| Grantee: | Carlos Augusto Prete Junior |
| Support Opportunities: | Scholarships in Brazil - Doctorate |