Abstract
This study aimed to identify highly valuable blood indicators for predicting the clinical outcomes of patients with aortic aneurysms (AA). Baseline data of 1180 patients and 16 blood indicators were obtained from the public Medical Information Mart for Intensive Care-IV (MIMIC-IV) database. The association of blood indicators with 4 types of clinical outcomes was analyzed, and the prediction performance of core indicators on different outcomes was next evaluated. Then, we explored the detailed association between core indicators and key outcomes among subgroups. Finally, a machine learning model was established to improve the prediction performance. Generalized linear regression analysis indicated that only red cell volume distribution width (RDW) was commonly associated with 4 end-points including surgery requirement, ICU stay requirement, length of hospital stay, and in-hospital death (all P < .05). Further, RDW showed the best performance for predicting in-hospital death by receiver operating characteristic (ROC) analysis. The significant association between RDW and in-hospital death was then determined by 3 logistic regression models adjusting for different variables (all P < .05). Stratification analysis showed that their association was mainly observed in unruptured AA and abdominal AA (AAA, all P < .05). We subsequently established an RDW-based model for predicting the in-hospital death only in patients with unruptured AAA. The favorable prediction performance of the RDW-based model was verified in training, validation, and test sets. RDW was found to make the greatest contribution to in-hospital death within the model. RDW had favorable clinical value for predicting the in-hospital death of patients, especially in unruptured AAA.