Machine-learning-based identification of patients with IgA nephropathy using a computerized medical billing database.

Area of Science:

• Nephrology
• Medical Informatics
• Machine Learning

Background:

• Billing databases offer large patient cohorts but risk misdiagnosis for research.
• Accurate identification of immunoglobulin A nephropathy (IgAN) patients from billing data is challenging.
• Existing diagnostic codes are insufficient for reliable IgAN patient selection in research.

Purpose of the Study:

• To develop a machine learning method for identifying IgAN patients from Japanese healthcare billing data.
• To improve the accuracy of IgAN patient cohort construction for clinical research.

Main Methods:

• Extracted medical records and billing data from 3,743 patients consulting nephrologists.
• Manually labeled IgAN diagnoses by reviewing medical records.
• Applied XGBoost machine learning with five-fold cross-validation after clinical viewpoint preprocessing.

Main Results:

• Manual criteria had specificity and sensitivity < 0.8.
• Machine learning achieved an Area Under the Curve (AUC) > 0.9.
• Machine learning demonstrated high performance in detecting IgAN patients.

Conclusions:

• Machine learning applied to clinically preprocessed billing data accurately identifies IgAN patients.
• This methodology facilitates the creation of IgAN-specific cohorts from large-scale billing datasets.
• The approach enhances the utility of healthcare big data for clinical research.