Machine Learning Model Development and Validation in Individuals With Suggestive Hypersensitivity to Nonsteroidal...

Area of Science:

• Allergy and Immunology
• Computational Medicine
• Pharmacovigilance

Background:

• Drug hypersensitivity reactions (HSRs) are frequently triggered by nonsteroidal anti-inflammatory drugs (NSAIDs).
• Accurate diagnosis of NSAID-induced HSRs is challenging, with many patients incorrectly identified.
• Misdiagnosis leads to unnecessary NSAID avoidance, impacting patient care and healthcare resources.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for discriminating NSAID-hypersensitive patients.
• To differentiate individuals with true NSAID hypersensitivity from those with suspected but unconfirmed reactions.

Main Methods:

• Retrospective and prospective cohorts of patients with suspected NSAID-induced HSRs were analyzed.
• Multiple ML models, including logistic regression and gradient-boosting machines, were developed and compared.
• The most effective model was validated on independent prospective populations.

Main Results:

• All developed models demonstrated high classification accuracy (≥85%).
• The Light Gradient-Boosting Machine (LGBM) model achieved superior performance with 99% sensitivity and 97% accuracy in initial testing.
• The validated LGBM model showed 91.76% accuracy and an area under the curve of 95.19%, with >95% accuracy in external validation cohorts.

Conclusions:

• The developed LGBM model effectively distinguishes NSAID-hypersensitive patients from those who can safely tolerate NSAIDs.
• Implementation in clinical settings can improve diagnostic accuracy for NSAID-induced HSRs.
• This ML approach can help reduce waiting lists and optimize healthcare resource allocation.