Neonatal Sepsis Prediction Model for Resource-Poor Developing Countries

摘要

Out of 5 million global neonatal deaths annually 96% of them occur in developing countries. Majority of these deaths are due to sepsis. Blood culture which is a gold standard to diagnose sepsis is time consuming and nearly takes 24–72 hours. To save time, prediction models/screening tests are used to start antimicrobial therapy. Most of the prediction models require invasive parameters to predict sepsis. But rural areas of developing countries lack laboratory facilities. Therefore, these prediction models cannot be implemented in these areas. In this retrospective study, binary logistic regression was used to develop and compare two prediction models using invasive and non-invasive parameters. The data for this study was taken from Medical Information Mart for Intensive care (MIMIC) III database. An Android application was developed to calculate the probability of sepsis after manually entering the independent parameter values. Prediction model developed from non-invasive parameters performed equally well as compared to prediction model made from invasive parameters. The AUROC in derivation dataset were 0.777 and 0.824 in case of invasive and non-invasive prediction models respectively whereas they were 0.830 and 0.824 in case of validation dataset. Both models were significant with p<0.001.