Predicting first-time anaphylaxis in the elderly using stacked machine learning and population registers.

Area of Science:

• Medical Informatics
• Clinical Decision Support
• Machine Learning in Healthcare

Background:

• Anaphylaxis is a severe, life-threatening allergic reaction requiring prompt recognition and treatment.
• Predicting anaphylaxis risk is challenging due to its complex causes and varied symptoms.

Purpose of the Study:

• To create and validate explainable machine learning (ML) models for predicting anaphylaxis risk.
• Utilize routinely collected clinical data for ML model development.

Main Methods:

• A matched case-control study using anonymized electronic health records.
• Employed chi-squared feature selection and trained various classification algorithms (logistic regression, decision trees, random forests, XGBoost, stacking ensemble).
• Evaluated model performance with AUC, sensitivity, specificity, precision, F1-score, and utilized SHAP values for explainability.

Main Results:

• The top-performing model achieved an Area Under the Curve (AUC) of 0.79, indicating strong predictive capability.
• Key predictors identified include healthcare utilization, age, socioeconomic status proxy (copayment level), and specific allergy-related diagnostic codes.
• The model demonstrated balanced sensitivity and specificity.

Conclusions:

• Interpretable machine learning models show promise for early identification of individuals at high risk of anaphylaxis.
• These ML tools can improve clinical risk stratification and guide preventive measures in healthcare settings.