Classification of angioedema types using decision tree modeling

Abstract

Introduction: All angioedema (AE) presents with transient, localized swelling; however, the underlying causes, prognosis, and treatments vary significantly. Consequently, identifying a specific AE type is challenging. Methods: We aimed to apply a machine learning (ML) model to improve AE diagnosis. Random forest (RF) ML was used to create a prediction model for diagnosing correct AE types. Development comprised a literature search to establish AE's clinical characteristics, developing and translating questions in collaboration with 12 European AE centers, and selecting, testing, validating and optimizing the established ML model. Analysis included 342 specialist-diagnosed patients with one of six AE types. Results: The final optimized RF model correctly identified AE types with true positive rates of up to 94% in hereditary AE due to C1 inhibitor deficiency (C1INH), with a Percentage Accuracy of 89·2% and a Kappa value of 81·8% across the six AE types, with a high agreement with the diagnoses made by experts. Discussion: This is the first ever reported ML algorithm designed to pre-assess to aid AE diagnosis.