Development of Machine Learning-Based Mpox Surveillance Models in a Learning Health System

Area of Science:

• Computational epidemiology
• Medical informatics
• Machine learning in healthcare

Background:

• The accurate and timely identification of mpox cases is crucial for public health surveillance and patient management.
• Clinical notes contain valuable information for disease detection but require sophisticated methods for analysis.
• Learning health systems aim to integrate data-driven insights into clinical practice for continuous improvement.

Purpose of the Study:

• To develop and evaluate machine learning and deep learning models for identifying mpox cases from unstructured clinical notes.
• To compare the performance of different modeling approaches, specifically Lasso regression and deep learning, in detecting mpox.
• To assess the utility of these models in a learning health system for quality improvement initiatives.

Main Methods:

• Development of machine learning (Lasso regression) and deep learning models.
• Training and validation of models using a dataset of clinical notes.
• Evaluation of model performance metrics, with a focus on accuracy and false positive rates.

Main Results:

• Lasso regression demonstrated superior performance compared to deep learning models in identifying mpox cases.
• The Lasso regression model was particularly effective in minimizing false positive identifications.
• The developed models show potential for integration into a learning health system.

Conclusions:

• Machine learning, specifically Lasso regression, offers a viable and effective approach for identifying mpox from clinical notes.
• The ability of Lasso regression to minimize false positives makes it a valuable tool for quality improvement, potentially aiding in the detection of missed or delayed diagnoses.
• These findings support the use of computational methods in learning health systems for enhanced disease surveillance and clinical quality improvement.