Evaluating the Surveillance System for Spotted Fever in Brazil Using Machine-Learning Techniques

Area of Science:

• Epidemiology
• Public Health
• Machine Learning Applications in Medicine

Background:

• The Brazilian spotted fever (SF) surveillance system in Rio de Janeiro (RJ) requires performance analysis.
• Accurate diagnosis and confirmation of suspected SF cases are crucial for effective public health interventions.
• Previous studies highlight challenges in differentiating SF from other febrile illnesses based on clinical presentation.

Purpose of the Study:

• To analyze the diagnostic performance of the SF surveillance system in Rio de Janeiro from 2007 to 2016.
• To identify key risk factors and clinical indicators associated with SF diagnosis, confirmation, and patient outcomes.
• To propose improvements for the Disease Notification Information System (SINAN) for better SF surveillance.

Main Methods:

• Retrospective analysis of 890 suspected SF cases reported to SINAN in Rio de Janeiro.
• Application of machine learning techniques, specifically decision trees, to analyze diagnostic classifications and clinical data.
• Utilized cartographic techniques to map patient movement patterns.

Main Results:

• Only 11.7% of reported cases were confirmed as SF, with significant misclassification rates for other diseases (dengue, leptospirosis) and unspecified categories (72.7%).
• Man-tick contact (71.2%) was identified as a significant risk indicator for SF, unlike man-capybara contact (1.7%).
• Clinical symptoms such as respiratory distress, convulsion, shock, petechiae, coma, icterus, and diarrhea were associated with SF patient mortality or cure.

Conclusions:

• Obstacles exist in the clinical and symptomatic diagnosis of suspected SF cases within the current surveillance system.
• The findings underscore the importance of tick exposure as a primary risk factor and specific clinical signs for prognosis.
• Recommendations are made for SINAN modifications to enhance understanding of SF dynamics and serve as a model for other endemic regions.