NON-INVASIVE PREDICTION MODEL FOR DEVELOPING COUNTRIES TO PREDICT SEPSIS IN NEONATES

摘要

Majority of global neonatal deaths is due to sepsis. A vast portion of these deaths occurs in developing countries due to inaccessibility of hospitals or lack of resources. Blood culture is the test to confirm sepsis, but it requires the presence of laboratory and is time-consuming. Therefore, we require simple, easy to use methods to predict sepsis in homes. Majority of the available prediction models need invasive parameters and hence become useless in the rural areas of developing countries where laboratory facilities do not exist. Non-invasive prediction models overcome these challenges to predict neonatal sepsis in places where there is a scarcity of laboratories. The aim and objective of this study are as follows: (i) to develop a practical, non-invasive prediction-model for neonatal sepsis which can be used in the rural areas of developing countries and to validate its performance. (ii) To compare the prognostic performance of the non-invasive prediction model with invasive prediction model and (iii) to create a prototype of the hardware which calculates the probability of the sepsis in neonates and sends the real-time data to the cloud. For this retrospective analysis, we extracted the data of 1446 neonates from Medical Information Mart for Intensive care III (MIMIC) database. Using stepwise logistic regression analysis, we developed and validated two prediction models. These two models were named as model NI and model O. Model O contains invasive as well as non-invasive parameters whereas model NI contains only non-invasive parameters. Model NI performed equally well in comparison to Model O despite using different predictors. The area under ROC curves for model NI and model O were 0.879 (95% CI: 0.857 to 0.899) and 0.861 (95% CI: 0.838 to 0.881) respectively. Both models were statistically significant with p-value<0.001.