20 janvier 2023
Guy Lacroix vient de publier un article dans la revue Health & Technology intitulé "Using machine learning to predict nosocomial infections and medical accidents in a NICU », Lien vers l’article
Résumé en anglais
Frail newborns admitted to a neonatal intensive care unit (NICU) are at high risk of contracting a nosocomial infection and are vulnerable to medical incidents due to the intensity of care they require. In this paper, we use a generalized mixed-effects regression tree model with random effects (GMERT-RI) to elaborate predictions trees for the two outcomes. GMERT-RI is an extension of the standard tree-based methods which is suitable for binary unbalanced panel data with random effects. These account for neonate-specific unobserved variables that may impact their outcomes. The sample includes 4,356 neonates admitted in the CHU de Québec Level-III NICU between 10 April 2008 and 28 March 2013. The algorithm is fed numerous neonates- and NICU-specific daily predictors and manages to unearth non-linear relationships between them as well as identify critical predictor variables thresholds. Interestingly, in addition to usual clinical predictors, important ones are within management's realm. Hence, machine learning algorithms such as GMERT-RI may prove to be a useful decision tool for management and pediatricians alike.