Unraveling the impact of time-dependent perioperative variables on 30-day readmission after coronary artery bypass surgery

Objectives: Readmission within 30 days of discharge after coronary artery bypass grafting is a measure of quality and a driver of cost in health care. Traditional predictive models use time-independent variables. We developed a new model to predict time to readmission after coronary artery bypass grafting using both time-independent and time-dependent preoperative and perioperative data.

Methods: Adults surviving to discharge after isolated coronary artery bypass grafting at a multi-hospital academic health system from January 2017 to September 2018 were included in this study. Two distinct data sources were used: the institutional cardiac surgical database and the clinical data warehouse, which provided more granular data points for each patient. Patients were divided into training and validation sets in an 80:20 ratio. We evaluated 82 potential risk factors using Cox survival regression and machine learning techniques. The area under the receiver operating characteristic curve was used to estimate model predictive accuracy.

Results: We trained the model with 21 variables that scored a P value of less than .05 in the univariable analysis. The multivariable model determined 16 significant risk factors, and 6 of them were time-dependent. These included preoperative hemoglobin a1c level, preoperative creatinine, preoperative hematocrit, intraoperative hemoglobin, postoperative creatinine, and postoperative hemoglobin. Area under the receiver operating characteristic values were 0.906 and 0.868 for training and validation sets, respectively.

Conclusions: Time-dependent perioperative variables in an isolated coronary artery bypass grafting cohort provided better predictive ability to a readmission model. This study was unique in the inclusion of time-dependent covariates in the predictive model for readmission after discharge after coronary artery bypass grafting.