Inflammatory signature-based theranostics for acute lung injury in acute type A aortic dissection

Abstract

Abstract Acute lung injury (ALI) is a serious adverse event in the management of acute type A aortic dissection (ATAAD). Using a large-scale cohort, we applied artificial intelligence-driven approach to stratify patients with different outcomes and treatment responses. A total of 2,499 patients from China 5A study database (2016–2022) from 10 cardiovascular centers were divided into 70% for derivation cohort and 30% for validation cohort, in which extreme gradient boosting algorithm was used to develop ALI risk model. Logistic regression was used to assess the risk under anti-inflammatory strategies in different risk probability. Eight top features of importance (leukocyte, platelet, hemoglobin, base excess, age, creatinine, glucose, and left ventricular end-diastolic dimension) were used to develop and validate an ALI risk model, with adequate discrimination ability regarding area under the receiver operating characteristic curve of 0.844 and 0.799 in the derivation and validation cohort, respectively. By the individualized treatment effect prediction, ulinastatin use was significantly associated with significantly lower risk of developing ALI (odds ratio [OR] 0.623 [95% CI 0.456, 0.851]; P = 0.003) in patients with a predicted ALI risk of 32.5–73.0%, rather than in pooled patients with a risk of <32.5 and >73.0% (OR 0.929 [0.682, 1.267], P = 0.642) (Pinteraction = 0.075). An artificial intelligence-driven risk stratification of ALI following ATAAD surgery were developed and validated, and subgroup analysis showed the heterogeneity of anti-inflammatory pharmacotherapy, which suggested individualized anti-inflammatory strategies in different risk probability of ALI.