Sequentially updated discharge model for optimizing hospital resource use and surgical patients' satisfaction

Background The ability to estimate cardiac surgical patients' length of stay (LOS) and discharge to a continuing care facility (nonhome discharge) may allow earlier discharge planning and optimal use of limited hospital resources. We developed a sequentially updated tool for postoperative discharge planning. Methods Using preoperative, intraoperative, and postoperative day (POD) 2 and POD 4 variables, we created and validated a model to predict early discharge (less than 4 days), standard discharge (5 to 8 days), delayed discharge (9 to 14 days), late discharge (more than 15 days), and nonhome discharge. Results When predicting LOS, model accuracy using preoperative variables alone had a C-statistic of 0.80, but improved with sequential addition of intraoperative and POD 2 (0.87) and POD 4 variables (0.89). At 48 hours, the strongest predictors of longer LOS were higher preoperative creatinine, elevated blood urea nitrogen, lower postoperative albumin, atrial fibrillation, and longer intensive care unit stay. On POD 4, the strongest predictors were red blood cell transfusion, lower postoperative albumin, white blood cell transfusion, longer intensive care unit stay, and readmission to the intensive care unit. For nonhome discharge, however, preoperative variables alone produced a highly predictive model (C-statistic 0.88), and sequential addition of intraoperative and POD 2 (C-statistic 0.91) and POD 4 data (C-statistic 0.90) did not significantly improve it. Conclusions This sequentially updated model of postoperative LOS can be used by the discharge planning team to identify both patients imminently ready for discharge and patients with a high likelihood of nonhome discharge, with the goals of decreasing unnecessary hospital days, managing patients' expectations, and engaging patients early in the discharge process.