An Artificial Intelligence Approach to Support Detection of Neonatal Adverse Drug Reactions Based on Severity and Probability Scores: A New Risk Score as Web-Tool

Abstract

Background: Critically ill neonates are at greater risk for adverse drug reactions (ADRs). The differentiation of ADRs from reactions associated with organ dysfunction/immaturity or genetic variability is difficult. Methods: In this prospective cohort study, each ADR was assessed using newborn-specific severity and probability scales by the clinical pharmacist. Subsequently, a machine learning-based risk score was designed to predict ADR presence in neonates. Results: In 98/412 (23.8%) of (56.3%; male) neonates included, 187 ADRs (0.42 ADR/patient) were determined related to 49 different drugs (37.12%). Drugs identified as high risk were enoxaparin, dexmedetomidine, vinblastine, dornase alfa, etoposide/carboplatin and prednisolone. The independent variables included in the risk score to predict ADR presence, according to the random forest importance criterion, were: systemic hormones (2 points), cardiovascular drugs (3 points), diseases of the circulatory system (1 point), nervous system drugs (1 point), and parenteral nutrition treatment (1 point), (cut-off value: 3 points). This risk score correctly classified 91.1% of the observations in the test set (c-index: 0.914). Conclusions: Using the high-performing risk score specific to neonates, it is expected that high-risk neonatal ADRs can be determined and prevented before they occur. Moreover, the awareness of clinicians of these drugs can be improved with this web-tool, and mitigation strategies (change of drug, dose, treatment duration, etc.) can be considered, based on a benefit-harm relationship for suspected drugs with a newborn-centered approach.