Machine Learning Informed Diagnosis for Congenital Heart Disease in Large Claims Data Source.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Health Data Science

Background:

• Large claims databases are increasingly vital for medical research.
• Efficient patient identification is crucial for disease research and clinical practice.
• Congenital heart disease (CHD) patient identification in large datasets presents challenges.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) approach for identifying patients with congenital heart disease (CHD) in large claims databases.
• To compare the performance of various ML algorithms against traditional methods for automated patient classification.

Main Methods:

• Utilized Quebec claims and hospitalization data (1983-2000) for 19,187 patients.
• Compared Gradient Boosting Decision Tree, Support Vector Machine, Decision Tree, and logistic regression.
• Evaluated models using Area Under the Precision Recall Curve (AUC), with external validation on data up to 2010.

Main Results:

• Gradient Boosting Decision Tree achieved the highest performance with 99.3% AUC.
• Sensitivity and specificity for the best model were 98.0% and 99.7%, respectively.
• External validation confirmed the model's robust performance on updated datasets.

Conclusions:

• ML algorithms can replace time-consuming manual clinical inspections for patient identification in large databases.
• Automated ML approaches are effective for identifying patients with complex diseases like CHD.
• This study demonstrates the efficiency and accuracy of ML in health claims data analysis.