Preoperative serum Raman spectroscopy and machine learning for predicting postoperative acute kidney injury after acute type A aortic dissection

Acute type A aortic dissection (ATAAD) represents an aggressive cardiovascular condition with substantial morbidity and mortality rates. Postoperative acute kidney injury (AKI) is a complication that contributes significantly to worse prognosis following total aortic arch replacement (TAAR) surgery. Currently, there are neither effective predictive methods nor data regarding the application of preoperative serum Raman spectroscopy for early identification of AKI after TAAR surgery. Our study aimed to develop a predictive model and assess its effectiveness in prediction of AKI after TAAR surgery. Preoperative blood biochemical indicators were investigated while Raman spectra from preoperative serum samples were analyzed by quantifying the peak intensities. Following appropriate feature selection procedures, multiple machine learning classification models were integrated to identify the optimal model for predicting postoperative AKI. To facilitate personalized risk assessment, Shapley Additive exPlanations (SHAP) interpretation was developed. In summary, 396 patients were enrolled in this study and incidence of postoperative AKI was 56.1% (222/396). Out of the 15 biochemical parameters and 5 intensities of specific Raman peaks possessing statistical differences, 6 indictors (creatinine, cystatin C, neutrophil, platelet, glucose, carbon dioxide combining power [CO2cp]) and 3 Raman peak intensities (505, 947 and 1208 cm−1) were identified as predictors of postoperative AKI. The logistic regression model emerged as the optimal model, with a test set area under the curve (AUC) of 0.848 (95% confidence interval [CI]: 0.777–0.918) To our knowledge, this is the first study to integrate preoperative serum Raman spectroscopy with machine learning for predicting AKI risk in patients undergoing ATAAD surgery.