Abstract
BACKGROUND: Heart failure (HF) is one of the most important indications of the severity of valvular heart disease (VHD). VHD with HF is frequently associated with a higher surgical risk. Our study sought to develop a risk score model to predict the postoperative mortality of suspected HF patients after valvular surgery. METHODS: Between January 2016 and December 2018, all consecutive adult patients suspected of HF and undergoing valvular surgery in the Chinese Cardiac Surgery Registry (CCSR) database were included. Finally, 14,645 patients (55.39 ± 11.6 years, 43.5% female) were identified for analysis. As a training group for model derivation, we used patients who had surgery between January 2016 and May 2018 (11,292 in total). To validate the model, patients who underwent surgery between June 2018 and December 2018 (a total of 3353 patients) were included as a testing group. In training group, we constructed and validated a scoring system to predict postoperative mortality using multivariable logistic regression and bootstrapping method (1000 re-samples). We validated the scoring model in the testing group. Brier score and calibration curves using bootstrapping with 1000 re-samples were used to evaluate the calibration. The area under the receiver operating characteristic curve (AUROC) was used to evaluate the discrimination. The results were also compared to EuroSCORE II. RESULTS: The final score ranged from 0 to 19 points and involved 9 predictors: age ≥ 60 years; New York Heart Association Class (NYHA) IV; left ventricular ejection fraction (LVEF) < 35%; estimated glomerular filtration rate (eGFR) < 50 mL/min/1.73 m2 ; preoperative dialysis; Left main artery stenosis; non-elective surgery; cardiopulmonary bypass (CPB) time > 200 minutes and perioperative transfusion. In training group, observed and predicted postoperative mortality rates increased from 0% to 45.5% and from 0.8% to 50.3%, respectively, as the score increased from 0 up to ≥ 10 points. The scoring model's Brier scores in the training and testing groups were 0.0279 and 0.0318, respectively. The area under the curve (AUC) values of the scoring model in both the training and testing groups were 0.776, which was significantly higher than EuroSCORE II in both the training (AUC = 0.721, Delong test, p < 0.001) and testing (AUC = 0.669, Delong test, p < 0.001) groups. CONCLUSIONS: The new risk score is an effective and concise tool that could accurately predict postoperative mortality rates in suspected HF patients after valve surgery.