Conference Program

Variable selection in the development of ICU scoring systems (310219)

Sisa Pazi, Nelson Mandela University 
*Gary David Sharp, Nelson Mandela University 
Elizabeth Van der Merwe, Nelson Mandela University 

Keywords: variable selection, logistic regression, stepwise regression, ICU scoring systems

Multivariable regression models are often used in health research to develop Intensive Care Unit (ICU) scoring systems. Therefore, variable selection remains an important task in the development of ICU scoring systems. This study explored five variable selection techniques in developing an ICU scoring system for assessing severity of illness in critically ill patients and prediction of in-hospital mortality in a tertiary hospital in South Africa. The database used was prospectively collected over a period of one year from 3 January 2017 to 3 January 2018. The results showed that stepwise regression is not always the best variable selection technique.