Mineração de texto aplicada às análises de intervenção de Políticas Públicas de Saúde: o caso da epidemia de sífilis no Brasil

Abstract

Syphilis is a chronic, curable infectious disease that has been known for centuries, caused by the Treponema Pallidum bacterium. Even with easy treatment and diagnosis, syphilis remains a serious public health problem in much of the world. Only in Brazil, data from the 2017 Epidemiological Report revealed an increase in the number of cases of syphilis in pregnant women, acquired and congenital. Considering the magnitude of the problem to be faced, in 2018 the “Applied research for intelligent integration aimed at strengthening care networks for a rapid response to syphilis” appears in Brazil, the “Syphilis No!” project, which aims to reduce cases of acquired syphilis and in pregnant women and eliminate congenital syphilis in the country. This project has several strategies to combat syphilis and, among them, the creation of a group of field researchers who worked in priority municipalities and produced thousands of text reports and added them to a platform. The objective of the thesis is the development of computational methods using text mining that help to understand the impact of syphilis in the territory using the textual productions of the LUES platform of the field researchers of the “Syphilis No!” project. The database extracted from the “LUES Platform” with 4,874 documents in text file and 3,071 documents in spreadsheets between the years 2018 and 2020 was used. This was followed by the pre-processing of these texts, with the choice for analysis of the texts referring to the field researchers’ reports. Finally, for this analysis, extraction of the N-grams (N=2,3,4) was performed using the combination of the TF-IDF metric with the BoW algorithm to verify the importance and frequency of the terms, for the grouping of the texts that were then analyzed using content analysis techniques and interpretation of the terms, thus, associations of the data extracted from the reports with indicators of syphilis and its epidemic impact on the territory were tested. Text mining, when used in conjunction with the traditional content analysis method, is able to meet public health research objects. The computational method extracted intervention actions from the field researchers, as well as subsidized inferences about how the strategies of the “Syphilis No!” project impacted the reduction of congenital syphilis cases in the territory.