Machine Learning Models to Analyze the Effect of Drugs on Neonatal-ICU Length of Stay

Abstract

The Neonatal intensive care unit (NICU) is a specialized section for newborn babies. The neonates are in vulnerable conditions in the ICU, so the predictive models will help to indicate the seriousness of the patients and assist the doctors in taking immediate actions. The Medical Information Mart for Intensive Care III (MIMIC-III) dataset is used in this research. The medicines medicated in critical newborn children were detected, and how the drugs and the doses of drugs can affect the Length of Stay (LOS) in NICU is analyzed. The predictive result of ICU Length of Stay (LOS) for the patients admitted to NICU for seven days is analyzed. Different Machine Learning algorithms were implemented for developing the classification model, and Logistic Regression Algorithm performed well and showed an F1 score of about 85%, which was better than the F1 score of the deep learning model long Short-Term Memory (LSTM). The automated Machine Learning (AutoML) tool, AutoNLP was also implemented for classifying LOS. But traditional methods demonstrated better performance in comparison to AutoML.