Estimating measles reintroduction risk in São Paulo using machine learning

Area of Science:

• Epidemiology and Public Health
• Computational Biology and Bioinformatics
• Infectious Disease Modeling

Background:

• Measles remains a global health concern, necessitating robust surveillance and risk prediction systems.
• Brazil, particularly São Paulo State, requires effective tools to monitor and mitigate measles transmission risks.
• Predictive modeling offers a promising approach to identify high-risk areas for targeted public health interventions.

Purpose of the Study:

• To develop and validate a machine learning-based tool for estimating measles risk at the municipal level in São Paulo State.
• To identify key epidemiological and demographic factors influencing measles occurrence in the region.
• To provide a data-driven framework for proactive measles risk management.

Main Methods:

• Utilized public data from 2007-2023, encompassing demographic, mobility, socioeconomic, and vaccination information.
• Trained five machine learning models using K-fold stratified cross-validation to predict annual measles occurrence.
• Employed Receiver Operating Characteristic (ROC) curves for threshold determination and Shapley Additive Explanations (SHAP) for feature importance analysis.

Main Results:

• All models demonstrated comparable performance based on AUC-ROC, F1-Score, and Brier Score metrics.
• The São Paulo metropolitan region and Santos were identified as areas with the highest measles risk.
• Key predictive features included international passenger arrivals, WHO-reported cases, municipal population size, and population density.

Conclusions:

• A novel measles risk estimation tool was successfully developed using publicly available data.
• The tool can aid in identifying high-risk municipalities for targeted measles prevention strategies in São Paulo State.
• The proposed framework offers a scalable model for similar risk assessment initiatives in other Brazilian states or regions.