Development of model predicting the risk of readmission to the intensive care unit after cardiac surgery based on MIMIC-III database

  AIM   To determine the independent predictors of readmission to intensive care unit (ICU) after cardiac surgery.

  METHODS   Patient data needed for research is extracted from Medical Information Mark for Intensive Care III (MIMIC-III); They were assigned into control group and case group according to the definition of “readmission to ICU”. The least absolute shrinkage and selection operator (LASSO) regression model was used to select the feature for multivariable logistic regression analysis that applied to build a predicting model incorporating the feature selected in the LASSO regression model; and based on the selected patients, a nomogram model predicting the risk of readmission to ICU was developed.

  RESULTS   A total of 4420 patients after cardiac surgery were included for retrospective analysis, divided into control group (n=4183, accounting for 94.6%) and case group (n=237, accounting for 5.4%); Multivariate logistic regression analysis showed that various perioperative parameters were implicated as independent predictors for ICU readmission after cardiac surgery, such as history of cardiac arrhythmias (OR=0.723,95%CI:0.546-0.958,P<0.05), peripheral vascular disease (OR=0.658, 95%CI: 0.465-0.930, P<0.05), renal failure (OR=0.649,95%CI:0.428-0.983, P<0.05), fluid electrolyte disorder (OR= 0.549, 95%CI: 0.380-0.792, P<0.01), congestive heart failure (OR = 0.476, 95%CI: 0.358-0.633, P<0.01), drug abuse (OR = 0.453, 95%CI:0.206-0.992, P<0.05), alcohol abuse (OR=0.402, 95%CI:0.206-0.786, P<0.01) and blood loss anemia (OR=0.260, 95%CI:0.085-0.796,P<0.05); and underwent non-elective surgery (OR= 2.906, 95%CI: 2.118-3.986, P<0.01) and complicated with severe sepsis after surgery (OR = 0.304, 95%CI: 0.095-0.974, P<0.05).Moreover, the model display good discrimination, with a AUC of 0.737.

  CONCLUSION   The nomogram model developed in our study to predict the risk of readmission to ICU after cardiac surgery may assist ICU physicians to identify high-risk patients. However, the model requires external validation before being recommended for clinical practice.