The following is a summary of “Predicting Penicillin Allergy: A United States Multicenter Retrospective Study,” published in the May 2024 issue of Allergy & Immunology by Gonzalez-Estrada, et al.

Previous studies have explored using logistic regression and machine learning (ML) models to predict penicillin allergy, primarily based on non-US data. However, ML models need to be tailored explicitly to US-based data. For a study, researchers sought to develop ML models for predicting positive penicillin allergy testing using multisite US data.

They conducted a retrospective analysis using data from four hospitals in the US, which were divided into four datasets: enriched training (with a 1:3 case-control matched cohort), enriched testing, nonenriched internal testing, and nonenriched external testing. ML algorithms were employed for model development. We assessed the performance of the models using the area under the curve (AUC) and utilized the Shapley Additive exPlanations (SHAP) framework to interpret the risk drivers.

The study included 4,777 patients (mean age 60 [standard deviation: 17] years; 68% women, 91% White, and 86% non-Hispanic), among whom 513 (11%) tested positive for penicillin allergy. Notably, model input variables frequently had missing data, including immediate or delayed onset (71%), signs or symptoms (13%), and treatment (31%). The gradient-boosted model demonstrated the strongest performance, with an AUC of 0.67 (95% CI: 0.57-0.77), which improved to 0.87 (95% CI: 0.73-1) when only cases with complete data were included. The top SHAP drivers for positive testing included recent reactions, reactions requiring medical attention, female sex, and the presence of hives/urticaria.

Despite efforts to develop ML prediction models for positive penicillin allergy testing using US-based retrospective data, the performance did not meet the required threshold for acceptance and adoption. Nevertheless, key drivers identified in the optimal ML prediction model included time since reaction, seeking medical attention, female sex, and the presence of hives/urticaria.

Reference: sciencedirect.com/science/article/abs/pii/S221321982400062X