A scoring model based on clinical factors to predict postoperative moderate to severe acute respiratory distress syndrome in Stanford type A aortic dissection

Abstract

Abstract Background Postoperative acute respiratory distress syndrome (ARDS) after type A aortic dissection is common and has high mortality. However, it is not clear which patients are at high risk of ARDS and an early prediction model is deficient. Methods From May 2015 to December 2017, 594 acute Stanford type A aortic dissection (ATAAD) patients who underwent aortic surgery in Anzhen Hospital were enrolled in our study. We compared the early survival of MS-ARDS within 24 h by Kaplan–Meier curves and log-rank tests. The data were divided into a training set and a test set at a ratio of 7:3. We established two prediction models and tested their efficiency. Results The oxygenation index decreased significantly immediately and 24 h after TAAD surgery. A total of 363 patients (61.1%) suffered from moderate and severe hypoxemia within 4 h, and 243 patients (40.9%) suffered from MS-ARDS within 24 h after surgery. Patients with MS-ARDS had higher 30-day mortality than others (log-rank test: p-value <0.001). There were 30 variables associated with MS-ARDS after surgery. The XGboost model consisted of 30 variables. The logistic regression model (LRM) consisted of 11 variables. The mean accuracy of the XGBoost model was 70.7%, and that of the LRM was 80.0%. The AUCs of XGBoost and LRM were 0.764 and 0.797, respectively. Conclusion Postoperative MS-ARDS significantly increased early mortality after TAAD surgery. The LRM model has higher accuracy, and the XGBoost model has higher specificity.

Keywords

• Machine learning
• Prediction model
• Acute respiratory distress syndrome
• Stanford type A aortic dissection