Machine Learning for Improved Dengue Diagnosis in Puerto Rico

Area of Science:

• Infectious Disease Epidemiology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Accurate dengue diagnosis is challenging in resource-limited settings due to symptom overlap and diagnostic method limitations.
• Existing diagnostic tools for dengue fever may not be readily accessible or rapid enough for widespread clinical use.

Purpose of the Study:

• To develop and evaluate machine learning models for improved dengue diagnosis using accessible clinical data.
• To enhance diagnostic accuracy for dengue, providing a potential rapid and accessible tool for healthcare providers.

Main Methods:

• Utilized data from the Sentinel Enhanced Dengue Surveillance System (SEDSS) in Puerto Rico (May 2012-June 2024).
• Evaluated various machine learning models (XGBoost, LightGBM, logistic regression, random forest, SVM, ANN, adaptive boosting) using demographic, clinical, laboratory, and epidemiological variables.
• Assessed model performance via the area under the receiver operating characteristic curve (AUC).

Main Results:

• XGBoost and LightGBM models achieved the highest diagnostic accuracy, with AUCs over 90%.
• Key predictors for improved dengue diagnosis included monthly incidence, leukopenia, thrombocytopenia, rash, age, and absence of nasal discharge.
• Incorporating comprehensive clinical and epidemiological features consistently enhanced model sensitivity and specificity.

Conclusions:

• Machine learning models, especially XGBoost and LightGBM, demonstrate significant potential for accurate dengue diagnosis using readily available clinical data.
• These models can be particularly valuable in resource-limited settings for improving dengue detection.
• Future work should focus on developing user-friendly tools (e.g., mobile apps, web platforms) for clinical implementation and exploring predictive applications.