A. B. Singh1, D. R. Gibula3, M. R. Bronsert1, W. G. Henderson1, K. E. Hammermeister1, N. O. Glebova4, R. A. Meguid1 1University Of Colorado Denver,Aurora, CO, USA 3University Of Utah,Salt Lake City, UT, USA 4Mid-Atlantic Permanente Medical Group,Vascular Department,Rockville, MD, USA
Introduction: Unplanned postoperative readmissions may indicate inferior healthcare quality which adversely impact patient recovery and quality of life. Patients desire to know their risk of unplanned readmissions and the surgeons need to know the risk to adequately counsel their patients. The existing Surgical Risk Preoperative Assessment System (SURPAS) shared decision making tool is a parsimonious model including eight predictor variables: Current Procedural Terminology-related risk, operative complexity, age, functional health status, American Society of Anesthesiologists physical status classification, in- or outpatient status, surgeon specialty, and emergency or elective operation. Developed from the American College of Surgeons’ National Surgical Quality Improvement Program (ACS NSQIP) dataset, SURPAS applies to >3000 operations in nine surgical specialties and predicts mortality, overall morbidity and eight clusters of common complications, and is incorporated into our health system’s electronic health record (EHR). We aim to develop an accurate preoperative prediction model for identifying the risk of unplanned postoperative readmission related to the primary procedure for integration into the EHR using all ACS NSQIP preoperative non-laboratory predictor variables and compare it to a model limited to the eight SURPAS predictor variables.
Methods: The full model was developed using logistic regression from all twenty-eight non-laboratory variables from the ACS NSQIP 2012-2016 dataset. It was compared to the model of the eight SURPAS predictor variables using the c-index as a measure of discrimination, the Hosmer-Lemeshow observed-to-expected plots testing calibration, and the Brier score, a combined metric of discrimination and calibration
Results: Of 3,715,921 patients,149,648 (4.03%) experienced an unplanned readmission related to the initial operation. The SURPAS model’s c-index, 0.727, was of 99.2% of that of the full model, 0.733, and Brier score of 0.0375 equal to the full model. Hosmer-Lemeshow analyses indicated similar calibration between the two models (see Figure).
Conclusions: The eight variable SURPAS model detects patients at risk for postoperative unplanned, related readmission as accurately as the full model developed from all available non-laboratory preoperative variables in the ACSNSQIP dataset. Therefore, unplanned readmission can be integrated into the existing SURPAS tool providing accurate prediction of postoperative readmission without necessitating the collection of additional predictor variables.
We aim to develop an accurate preoperative prediction model for identifying the risk of unplanned postoperative readmission related to the primary procedure for integration into the EHR using all ACS NSQIP preoperative non-laboratory predictor variables and compare it to a model limited to the eight SURPAS predictor variables.
