Predição do risco de adoecimento por hanseníase em contatos de casos da doença de uma região endêmica brasileira

Abstract

This study aimed to investigate the determinants of Mycobacterium leprae infection and illness due to leprosy in contacts of cases of the disease in the Microregion of Almenara, Minas Gerais, Brazil, to support the construction of a leprosy prediction model in contacts of patients from endemic regions. We conducted a retrospective cohort of household contacts of leprosy patients with a follow-up period from 1999 to 2018. The researchers performed interviews using a semi-structured questionnaire containing sociodemographic and health questions, biological samples collection, and dermatological examination. Biological samples supported the evaluation of genetic polymorphisms and reactivity to serological tests. The choice of the explanatory variables included in the analysis was based on the theoretical model of the determinants of leprosy in contacts, developed from a systematic review. The association analysis used a logistic regression model using the Generalized Estimating Equations estimation method. The construction of the prediction model involved exploratory data analysis and applied machine learning algorithms. The determinants of M. leprae infection were: black and mixed skin color, homozygous and heterozygous genotypes containing the rs8057341 polymorphism in the NOD2 gene (Nucleotide-binding Oligomerization Domain Containing 2), and living with a leprosy patient with disabilities at diagnosis. The determinants of illness due to leprosy in contacts were: consanguinity with the index case and living in the same household or yard after the diagnosis of the leprosy patient. Age, living in the same household or lot, with more than one leprosy patient that presented disabilities at the diagnosis were determinants for the risk of infection and illness in contacts. The heterozygous genotype that carried the rs2430561 polymorphism in the IFNG gene (Interferon-gamma) was a protective factor for M. leprae infection and leprosy in contacts of patients. The supervised learning algorithms Naive Bayes with the discretization of numerical variables, J48 and Random Forest had the best performances in the evaluated datasets. The determinants of the process of infection and illness due to leprosy were able to compose prediction models with accuracy and sensitivity higher than 90%. These results indicate that using these technologies in Primary Health Care services can improve contact surveillance, especially in highly endemic areas.