Machine Learning-Based Predictive Modeling for the Development of Chronic Rhinosinusitis Using Longitudinal Health...

Area of Science:

• Computational biology
• Medical informatics
• Epidemiology

Background:

• Chronic Rhinosinusitis (CRS) diagnosis relies on clinical presentation and history.
• Identifying early disease trajectories can improve patient outcomes and resource allocation.
• Electronic Health Records (EHRs) offer vast data for studying disease progression.

Purpose of the Study:

• To apply machine learning to identify reproducible pre-diagnostic trajectories of Chronic Rhinosinusitis (CRS).
• To evaluate the generalizability and performance of these predictive models in large EHR cohorts.
• To identify key predictors of CRS development and progression for clinical insight.

Main Methods:

• Utilized machine learning algorithms on large-scale Electronic Health Record (EHR) data.
• Developed models to predict pre-diagnostic Chronic Rhinosinusitis (CRS) trajectories.
• Validated model performance using external datasets and assessed generalizability (AUC ~ 0.81).

Main Results:

• Identified reproducible pre-diagnostic CRS trajectories within EHR data.
• Achieved stable external model performance with an Area Under the Curve (AUC) of approximately 0.81.
• Highlighted key predictors reflecting evolving CRS diagnostic and management patterns.

Conclusions:

• Machine learning effectively identifies distinct, reproducible pre-diagnostic CRS pathways.
• The models exhibit robust generalizability, indicating potential for real-world application.
• Findings support the use of these models as future clinical decision support tools for CRS.