Artificial intelligence and delabeling of penicillin allergy
Published online: July 20, 2020
Artificial Intelligence implies the modeling of intelligent behavior using computers. It can be applied in different medical aspects, such as robotics, medical diagnosis, medical statistics, and "omics." In a recent study published in The Journal of Allergy and Clinical Immunology: In Practice, Moreno et al. evaluated the usefulness of an artificial neural network (ANN) in the prediction of beta-lactam (BL) allergy. Penicillins are the drugs most frequently involved in allergic reactions. Nevertheless, a significant percentage of patients that claim to be allergic to BL are not, after a proper evaluation. The consequences of being labeled with penicillin or BL allergy for both patients and public health systems are significant. An accurate diagnosis of BL allergy improves antibiotic use, increases patients' safety and reduces health systems costs. Nevertheless, this requires skin and drug provocation tests, which are time-consuming and are not exempt from risk. Furthermore, allergy testing may not be available in all circumstances such as an urgent need for antibiotic therapy.
In recent years, there has been a growing interest in developing risk stratification classifications to identify low-risk patients labeled as allergic to penicillin, in whom other BLs can be safely used. There has also been growing interest in the design of predictive models based on medical history. Logistic regression has been the most commonly used method. Artificial intelligence and machine learning have been developing to assist doctors in decision-making situations. An artificial neural network (ANN) is a data-processing method designed to emulate the human brain neuronal architecture and function. A neural network can learn from data, so it can be trained to recognize patterns, classify data, and predict future events.
In a single-center study performed in Spain, Moreno et al. retrospectively analyzed 656 patients evaluated for BL allergy that had a final diagnosis of allergy or tolerance to BL. These data were used to construct an ANN able to predict whether the patients were allergic or not. The ANN predictive capabilities were compared to logistic regression and then prospectively evaluated in 615 patients that underwent BL evaluation. The ANN performance was far superior to LR. The ANN classified correctly 89.5% and 81.1% of allergic patients (LR only rated 31.9%) and 86.1% and 97.9% of non-allergic patients in the training and assess groups. Notably, an exciting advantage was that no patient with a history of severe reactions was misclassified.
Moreno et al suggest that the use of ANNs with high performance in emergencies could facilitate the decision of whether or not to give BL to patients claiming to be allergic. The ANN could also be a helpful tool to classify the reaction risk, particularly in identifying low-risk patients to whom an open challenge without previous skin testing could be done to delabel patients. The present study is the first to explore the usefulness of an artificial neural network to assist doctors in decision-making for patients with a label of penicillin or BL allergy.
The Journal of Allergy and Clinical Immunology: In Practice is an official journal of the AAAAI, focusing on practical information for the practicing clinician.