Full text
| |
| Author(s): |
Verena Hokino Yamaguti
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | Ribeirão Preto. |
| Institution: | Universidade de São Paulo (USP). Faculdade de Medicina de Ribeirão Preto (PCARP/BC) |
| Defense date: | 2023-05-17 |
| Examining board members: |
Antonio Ruffino Netto;
Domingos Alves;
Valdes Roberto Bollela;
José Alberto da Silva Freitas
|
| Advisor: | Antonio Ruffino Netto |
| Abstract | |
Tuberculosis is among the leading causes of death from infectious diseases. Tuberculosis was the leading cause of deaths by infectious disease until the beginning of the coronavirus pandemics. For effective control of the disease in Brazil, the Directly Observed Treatment Strategy is used. However, there is a significant number of new cases reported. The use of information system aims to improve the quality of planning, program implementation, treatment control and information management. It is therefore crucial to provide such systems with capabilities that help health professionals to manage the resources available and focus their efforts on cases that require their attention. Thus, the general objective of this project is to carry out a study for a prediction model to the tuberculosis treatment abandonment. This would provide a way to predict treatment loss to follow-up and redirect resources in advance to improve the adherence of these cases, reducing the treatment loss to follow-up rate and the number of infections by resistant bacilli. Initially (Article 1), we analyzed different databases comparing them for the completeness and reliability of the data from 50 treatments in 2 ambulatories in the state of São Paulo. The databases used to collect the information were: 1) SISTb; 2) Hygia; 3) Local ambulatory medical records; 4) Clinical Hospital electronic medical records; and 5) TBWEB. The database which presented the best completeness and reliability scores was TBWEB. Subsequently (Article 2), the selection of the most related attributes to the tuberculosis treatment outcome was performed. Later (Article 3 and 4), we developed two different models for tuberculosis prediction. First (Article 3), we developed a classifier using CART model. Later (Article 4), we developed a prediction classifier from the clusters generated by a hierarchical clustering model. Both developed models showed similar performance when compared to other models found in the literature. Besides, our models are highlighted by generating human comprehensible models that can be used by healthcare professionals through tuberculosis treatment to identify undesired outcomes patterns. (AU) | |
| FAPESP's process: | 18/23963-2 - Study of a prediction model for Tuberculosis treatment abandonment |
| Grantee: | Verena Hokino Yamaguti |
| Support Opportunities: | Scholarships in Brazil - Doctorate (Direct) |