Febrile disease modeling and diagnosis system for optimizing medical decisions in resource-scarce settings

Abstract

Febrile diseases are highly prevalent in tropical regions due to elevated humidity and high temperatures. These regions, mainly comprising low- and middle-income countries, often face challenges related to inadequate medical infrastructure and a lack of skilled personnel for accurately diagnosing febrile diseases. Distinguishing one febrile illness from another posed a significant challenge, adding to the complexity of accurate diagnoses. This study developed a multi-symptom multi-disease model to address this challenge, leveraging exploratory data analysis of patient datasets from field studies and the expertise of medical practitioners specializing in tropical diseases. The research investigated the most effective modeling approach for differentiating among 11 febrile illnesses that are prevalent in Nigeria using three intelligent techniques: Extreme Gradient Boost (XGBoost), Fuzzy Cognitive Map (FCM), and Analytic Hierarchy Process (AHP). Comparative analysis demonstrates that AHP surpassed the others, achieving a precision of 84%, recall of 83%, and an F1-score of 84%. Consequently, the AHP technique was integrated into the development of “Febra Diagnostica,” an app aimed at enhancing febrile disease diagnosis in resource-constrained settings. The app was then deployed and utilized in select Nigerian states, offering scalability and empowering frontline health workers in primary health facilities. Febra Diagnostica featured user-friendly interfaces, automated diagnosis and treatment suggestions, streamlined referrals, and provisions for further investigations. Encryption, access control, and multi-factor authentication were some of the security and privacy considerations in the app which gained acceptance from medical experts and adapted to regulatory and ethical policies for smart healthcare systems.