Preprocessing and Visualisation of Dataset for Neonatal Sepsis Prediction Using MIMIC-IV

This paper aims at exploring MIMIC-IV dataset for extraction for neonatal patients diagnosed with “Septicemia [sepsis] of newborn”. The term neonate is used for newborn babies less than four weeks old. Medical Information Mart for Intensive Care version 4 (MIMIC-IV) is a large, open access electronic health record database having more than 3 lakhs patients' data. The main objective of this paper is to access and filter this database using Google Big query. The Neonatal Sepsis is one of the reasons affecting neonatal mortality. Neonatal mortality reduction is one of the goal for sustainable development as mentioned by World Health Organization. This paper serves as the first step in identifying trends from the neonatal records from MIMIC dataset thereby paving away for creation of Sepsis prediction dataset and feature extraction. This paper deals with assessing the combined effect of various modules available in MIMIC database such as patient's demographic information, admission and diagnosis related details. It was found that there were total 96564 entries with sepsis diagnosis out of which 42.3 % were females and 57.7% were male neonates. The highest number of neonates diagnosed with Sepsis were of “White” ethnicity. The highest number of neonates were prescribed to perform “Blood Culture Neonate” test. The majority of neonates were found to have micro-organism named “Staph Aureus Coag+” in their body. This paper successfully concludes that the exploratory analysis of neonates with sepsis would be a foundation step in developing the neonatal sepsis prediction model using machine learning techniques.