Prediction of Intracranial Aneurysm Risk using Machine Learning.

Area of Science:

• Medical Informatics
• Epidemiology
• Machine Learning

Background:

• Intracranial aneurysms (IA) pose significant health risks, necessitating efficient identification of high-risk individuals for timely screening and resource allocation.
• Current screening strategies may benefit from advanced predictive modeling to improve accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the pre-diagnosis risk of intracranial aneurysms (IA).
• To identify the most effective algorithm for stratifying individuals into distinct risk groups for targeted screening.

Main Methods:

• Utilized a national claims database and health examination records from Korea's National Health Screening Program.
• Applied and compared four machine learning algorithms: logistic regression (LR), random forest (RF), XGBoost (XGB), and deep neural networks (DNN).
• Evaluated model performance using the area under the receiver operating characteristic curve (AUROC) and compared incidence rate ratios between risk groups.

Main Results:

• The XGBoost model demonstrated the highest predictive performance with an AUROC of 0.765.
• The XGBoost model predicted the lowest IA incidence in the lowest-risk group (3.20), while the RF model predicted the highest incidence in the highest-risk group (161.34).
• Significant incidence rate ratios between the lowest- and highest-risk groups were observed across all models, with XGBoost showing the largest ratio (49.85).

Conclusions:

• A machine learning-based prediction model, particularly XGBoost, can effectively identify individuals at high risk for intracranial aneurysms.
• The developed model holds potential for enhancing current IA screening strategies and optimizing medical resource allocation.
• This approach facilitates personalized risk assessment, paving the way for more efficient and effective public health interventions.