The Prognostic Multivariate Models For Severe Complications After Heart Valve Surgery

Abstract

\n Background: To provide prognostic multivariate models for severe complications prediction after heart valvular surgery, such as low cardiac output syndrome (LOCS), acute kidney injury requiring hemodialysis (AKI-rH) and multiple organ dysfunction syndrome (MODS).Methods: We developed multivariate logistic regression models to predict the severe complications after heart valvular surgery using 930 patients collected retrospectively from the first affiliated hospital of Sun Yat-Sen University from January 2014 to December 2015. The validation was conducted using a retrospective dataset of 713 patients from the same hospital from January 2016 to March 2017. We considered two kinds of prognostic models: the PRF models which were built by using the preoperative risk factors only, and the PIRF models which were built by using both of the preoperative and intraoperative risk factors. The Least absolute shrinkage selector operator was used for developing the models. We assessed and compared the discriminative abilities for both of the PRF and PIRF models via the receiver operating characteristic curve. Results: Compared with the PRF models, the PIRF modes selected additional intraoperative factors, such as auxiliary cardiopulmonary bypass time and combined tricuspid valve replacement. The area under the curves (AUCs) of the PRF models for predicting LOCS, AKI-rH and MODS are 0.565 (0.466, 0.664), 0.688 (0.62, 0.757) and 0.657 (0.563, 0.751), respectively. As a comparison, the AUCs of the PIRF models for predicting LOCS, AKI-rH and MODS are 0.821 (0.747, 0.896), 0.78 (0.717, 0.843) and 0.774 (0.7, 0.847), respectively. Conclusions: Adding the intraoperative factors can increase the predictive power of the prognostic models for severe complications prediction after heart valvular surgery.