Machine learning para predição de reação adversa ao medicamento: aplicação a neonatos em terapia intensiva

Abstract

Introduction: The intensive care of newborns is associated with a large volume of data in their medical records. The processing of these data can be done through Machine Learning: the ability to improve the performance of some task through experience and, thus, assist in the detection and decision-making a tool capable of assisting in the detection and decision-making of a wide range of medical conditions, including adverse drug reactions (ADR). Purpose: Train prediction model to help detect adverse drug reactions in neonates admitted to an intensive care unit (ICU). Methods: observational study developed in the Neonatal Intensive Care Unit of a teaching hospital in Brazil. Clinical data were collected from the daily pharmacotherapeutic follow-up, processed and analyzed by machine learning through libraries written in Python language. Results: Eight hundred and three newborns were included in the study, with a mean gestational age of 32.2 ± 4.2 weeks and a mean birth weight of 1807.2 ± 936.6g. The incidence of ADR was 10.8%. Antimicrobials, especially aminoglycosides, were the most prescribed drugs in this population. An algorithm was trained and tested in the prediction of ADR in NICU, whose metrics were precision (0.35) and recall (0.823), with specificity (80%) and sensitivity (67%). Conclusion: There is a high potential in the machine learning method for predicting ADR in newborns admitted to an ICU.