Classification of angioedema types using decision tree modeling

Area of Science:

• Immunology
• Computational Biology
• Medical Diagnostics

Background:

• Angioedema (AE) presents with localized swelling, but diverse causes, prognoses, and treatments complicate diagnosis.
• Accurate identification of specific AE types is crucial for effective management.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for improving the diagnostic accuracy of various angioedema types.
• To create an AI-driven tool to assist clinicians in differentiating between six types of AE.

Main Methods:

• A Random Forest (RF) ML model was developed using clinical data from 342 patients across 12 European AE centers.
• The model was trained on established AE characteristics, validated, and optimized through literature review and expert collaboration.

Main Results:

• The optimized RF model achieved up to 94% true positive rate for hereditary angioedema (HAE) due to C1 inhibitor deficiency (C1INH).
• Overall accuracy across six AE types was 89.2%, with a Kappa value of 81.8%, demonstrating high agreement with expert diagnoses.

Conclusions:

• This study presents the first ML algorithm designed for pre-assessment to aid in angioedema diagnosis.
• The developed ML model shows significant potential in accurately identifying AE types, supporting clinical decision-making.