Prediction of Length of ICU Stay Using Data-mining Techniques: an Example of Old Critically Ill Postoperative Gastric Cancer Patients

Abstract

<br/><b>Objective</b>: With the background of aging population in China and advances in clinical medicine, the amountof operations on old patients increases correspondingly, which imposes increasing challenges to critical caremedicine and geriatrics. The study was designed to describe information on the length of ICU stay from a singleinstitution experience of old critically ill gastric cancer patients after surgery and the framework of incorporatingdata-mining techniques into the prediction. <br/><b>Methods</b>: A retrospective design was adopted to collect the consecutivedata about patients aged 60 or over with a gastric cancer diagnosis after surgery in an adult intensive care unitin a medical university hospital in Shenyang, China, from January 2010 to March 2011. Characteristics ofpatients and the length their ICU stay were gathered for analysis by univariate and multivariate Cox regressionto examine the relationship with potential candidate factors. A regression tree was constructed to predict thelength of ICU stay and explore the important indicators. <br/><b>Results</b>: Multivariate Cox analysis found that shockand nutrition support need were statistically significant risk factors for prolonged length of ICU stay. Altogether,eight variables entered the regression model, including age, APACHE II score, SOFA score, shock, respiratorysystem dysfunction, circulation system dysfunction, diabetes and nutrition support need. The regression treeindicated comorbidity of two or more kinds of shock as the most important factor for prolonged length of ICUstay in the studied sample. <br/><b>Conclusions</b>: Comorbidity of two or more kinds of shock is the most important factorof length of ICU stay in the studied sample. Since there are differences of ICU patient characteristics betweenwards and hospitals, consideration of the data-mining technique should be given by the intensivists as a lengthof ICU stay prediction tool.