A local risk prediction model for prolonged ventilation after adult heart valve surgery in a Chinese single center

Abstract

Background and Aim of the StudyThe aim of this study was to develop a logistic risk prediction model for prolonged ventilation after adult heart valve surgery. Materials and MethodsThis is a retrospective observational study of collected data on 3965 consecutive patients older than 18 years, who had undergone heart valve surgery between January 2000 and December 2010. Data were randomly split into a development dataset (n = 2400) and a validation dataset (n = 1565). A multivariate logistic regression analysis was undertaken using the development dataset to identify independent risk factors for prolonged ventilation (defined as ventilation greater than 72 h). Performance of the model was then assessed by observed and expected rates of prolonged ventilation on the development and validation dataset. Model calibration and discriminatory ability were analyzed by the Hosmer–Lemeshow goodness-of-fit statistic and the area under the receiver operating characteristic (ROC) curve, respectively. ResultsThere were 303 patients that required prolonged ventilation (7.6%). Preoperative independent predictors of prolonged ventilation are shown with odds ratio and P value as follows: (1) age, 1.9, P < .0001; (2) hypercholesterolemia, 5.3, P = .001; (3) renal failure, 18.2, P = .004; (4) previous cardiac surgery, 2.4, P = .0002; (5) left bundle branch block, 4.2, P = .011; (6) ejection fraction, 1.4, P = .003; (7) left ventricle weight, 1.5, P = .007; (8) New York Heart Association class III-IV, 1.8, P = .021; (9) critical preoperative state, 4.5, P < .0001; (10) tricuspid insufficiency, 1.2, P = .031; (11) concurrent CABG, 2.2, P = .019; and (12) concurrent other cardiac surgery, 2.1, P = .001. The Hosmer–Lemeshow goodness-of-fit statistic was not statistically significant in both development and validation dataset (P = .202 vs P = .291). The ROC curve for the prediction of prolonged ventilation in development and validation dataset was .789 and .710, respectively. ConclusionsWe developed and validated a local risk prediction model for prolonged ventilation after adult heart valve surgery. This model can be used to calculate patient-specific risk by the logistic equation with an equivalent predicted risk at our center in future clinical practice.