A total of 57,180 patients (45,744 patients in the derivation cohort and 11,436 in the validation cohort) were retrospectively analyzed during the 7-year period. The prevalence of AD after cardiac surgery in the whole cohort was 3.6% (2085/57180), whereas in the derivation and validation cohorts, it was 3.3% (1504/45744) and 5.1% (581/11436), respectively. The mortality rate in the whole cohort was 0.9% (504/57180), Whereas in the AD cohort, it was 1.4% (29/2085). The baseline clinical characteristics of patients in the derivation and validation cohorts are illustrated in eTable 1 in the Supplement. The perioperative information of the patients in the derivation group is shown in Table 1.
Analysis of the Risk Variables
The details of the preoperative prediction model for AD are shown in eTable 2, whereas the pre- and intraoperative prediction model is shown in eTable 3, and the pre-, intra- and postoperative prediction model is shown in eTable 4. The risk variables contributing to AD were age, male sex, obesity, previous chronic obstructive pulmonary disease (COPD), hypertension, type 2 diabetes, New York Heart Association (NYHA) classification=4, low preoperative left ventricular ejection fraction (LVEF), elevated serum creatinine, emergency surgery, alcohol use, carotid artery stenosis, history of stroke, coronary artery disease, low total protein, type of surgery, intraoperative hemorrhage volume > 600 mL, intraoperative red blood cell (RBC) count, platelet or plasma use and low postoperative LVEF.
Diagnostic Utility of the Prediction Score
1. AD prediction model based on preoperative variables
After using only the preoperative variables in the derivation cohort to construct the AD prediction model, the AUC for AD was 0.68 (95% CI, 0.67, 0.70, Figure 1), and in the validation cohort, the AUC was 0.67 (95% CI, 0.64, 0.69, Figure 1). Nevertheless, the calibration according to the Hosmer-Lemeshow test was poor for this model (P = 0.01). The sensitivity, specificity, positive predictive value, and negative predicted value for predicting the medium- and high-risk groups were 37.5%, 83.1%, 10.6%, and 96.1%, respectively.
2. AD prediction model based on pre- and intraoperative variables
In the derivation data set, the performance of the AD risk prediction model, which was based on pre- and intraoperative variables, was as follows: AUC = 0.74 (95% CI, 0.72, 0.75, Figure 1), indicating good discrimination ability. It was very similar to the validation cohort (AUC = 0.74, 95% CI, 0.72, 0.76). The Hosmer-Lemeshow test showed that the calibration of the prediction model was good (P = 0.49). The sensitivity, specificity, positive predictive value, and negative predicted value for predicting the medium- and high-risk groups were 49.7%, 82.2%, 13.00%, and 96.80%, respectively.
3. AD prediction model based on pre-, intra-, and postoperative variables
The AUC for the AD prediction model with the derivation cohort was 0.75 (95% CI, 0.73, 0.76, Figure 1). The Hosmer-Lemeshow test demonstrated good calibration for the derivation cohort (P = 0.35). The performance with the validation cohort also indicated good discriminability (AUC=0.74, 95% CI, 0.71, 0.76). The sensitivity, specificity, positive predictive value, and negative predicted value for predicting the medium- and high-risk groups were 52.7%, 79.5%, 12.10%, and 96.90%, respectively.
Development of the Prediction Score
Three scoring systems for predicting postoperative AD after cardiac surgery are presented in Table 2. The prediction scoring systems were developed from regression coefficients of data from the derivative cohort patients. Depending on the scoring system, the predicted risks of AD could be grouped into three classifications: low, medium, and high, in accordance with the practically observed incidence of AD (Table 3). In the validation cohorts, the incidences of AD onset predicted by the model were similar to those observed clinically (Figure 2). The risk scores and their associated predictive risks are presented in eTable 5 in the Supplement.
Negative Outcomes
AD was associated with greater medical expenditures and a prolonged length of hospital stay (P < 0.001, eTable 6). AD was also associated with mortality (P =0.011). However, Postoperative AD was not correlated with the risk of pulmonary complications or the requirement of reintubation (P = 0.572, P = 0.496, eTable 7).