Prediction Model of Intensive Care Unit Length of Stay for Patients with Cardiac Surgery

The analysis and prediction of influencing factors of length of stay in intensive care unit (ICU) of cardiac surgery patients is conducive to the early intervention and cost control of inpatients, and is of great significance to the treatment and nursing of cardiac surgery patients. This paper uses the intensive care database medical information mart for intensive care IV (MIMIC-IV) as the experimental data set, 7567 patients were included. 41 important predictors were selected from 126 influencing factors by least absolute shrinkage and selection operator (Lasso). This paper constructs a prediction model of length of stay in cardiac surgery intensive care unit based on gradient enhanced decision tree (GBDT) algorithm. The experimental results show that under the condition of training all predictors, the average accuracy of GBDT model is 0.688 higher than that of traditional logistic regression algorithm, which is 0.603. The GBDT algorithm with the selected important predictors has the same effect on the final average accuracy as that with all factors, which shows that this method can optimize data collection, accurately predict length of stay in ICU, and provide algorithm support for clinical decision support system.