Machine learning-based risk prediction model for pertussis in children: a multicenter retrospective study.

Area of Science:

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

Background:

• Pertussis (whooping cough) remains a public health concern due to waning vaccine immunity and immune escape.
• Current diagnostic methods (PCR, culture) are slow and costly, hindering timely intervention.
• There is a need for efficient tools to identify high-risk pertussis patients for prompt management.

Purpose of the Study:

• To develop and validate an efficient machine learning model for pertussis risk prediction.
• To create a generalizable model applicable across different datasets and clinical settings.
• To provide a rapid, accurate auxiliary diagnostic tool for clinical practice and public health management.

Main Methods:

• Collected data from 1085 suspected pertussis patients across 7 centers.
• Utilized lasso regression and Boruta algorithm to identify 10 key predictive features.
• Trained and validated eight machine learning models, including a random forest model, using multicenter data.
• Developed an online platform for real-time clinical application of the validated model.

Main Results:

• The random forest model achieved high discrimination with an AUC of 0.98 (validation) and 0.97 (external validation).
• Calibration and decision curve analyses confirmed the model's accuracy, particularly for low-to-medium risk patients.
• The model assists clinicians in avoiding unnecessary interventions, especially in resource-limited environments.

Conclusions:

• A multicenter-validated pertussis prediction model demonstrates high performance and online applicability.
• The model facilitates optimized early identification and management of high-risk pertussis cases.
• Future work should expand data sources and integrate dynamic data for enhanced accuracy and broader use.