Length of Stay Analysis at Neonatal Care Units with Data Science - Preliminary Results

Abstract

The paper presents preliminary results to the length-of-stay (LOS) analysis on neonatal intensive care units, taking as base a 10 year dataset that encompasses data from all neonatal Portuguese units. Unlike conventional studies based on statistical analysis, this investigation uses data science techniques, including machine learning hyperparameters optimization automatization and explainable artificial intelligence (XAI) tools, to understand the relevance of factors and their impact on LOS. In this work was designed and implemented an accurate LOS prediction model which results have been included in the paper. Through XAI techniques application, it was already possible to confirm insights claimed by state-of-the-art studies, namely about the importance of birth weight and gestational age for LOS as well as the importance of hospital-acquired infections. The work presents as novelty the weights (quantification) of these factors as well as the introduction of two new factors, which were not considered until now. The study makes part of ongoing work and will continue to provide better insights for better investment LOS in neonatal care units.