Abstract
OBJECTIVES: To analyze the differential expression of lipid spectrum between ST-segment elevated myocardial infarction (STEMI) and patients with emergency chest pain and excluded coronary artery disease (CAD), and establish the predictive model which could predict STEMI in the early stage. METHODS: We conducted a single-center, nested case-control study using the emergency chest pain cohort of Peking University Third Hospital. Untargeted lipidomics were conducted while LASSO regression as well as XGBoost combined with greedy algorithm were used to select lipid molecules. RESULTS: Fifty-two STEMI patients along with 52 controls were enrolled. A total of 1925 lipid molecules were detected. There were 93 lipid molecules in the positive ion mode which were differentially expressed between the STEMI and the control group, while in the negative ion mode, there were 73 differentially expressed lipid molecules. In the positive ion mode, the differentially expressed lipid subclasses were mainly diacylglycerol (DG), lysophophatidylcholine (LPC), acylcarnitine (CAR), lysophosphatidyl ethanolamine (LPE), and phosphatidylcholine (PC), while in the negative ion mode, significantly expressed lipid subclasses were mainly free fatty acid (FA), LPE, PC, phosphatidylethanolamine (PE), and phosphatidylinositol (PI). LASSO regression selected 22 lipids while XGBoost combined with greedy algorithm selected 10 lipids. PC (15: 0/18: 2), PI (19: 4), and LPI (20: 3) were the overlapping lipid molecules selected by the two feature screening methods. Logistic model established using the three lipids had excellent performance in discrimination and calibration both in the derivation set (AUC: 0.972) and an internal validation set (AUC: 0.967). In 19 STEMI patients with normal cardiac troponin, 18 patients were correctly diagnosed using lipid model. CONCLUSIONS: The differentially expressed lipids were mainly DG, CAR, LPC, LPE, PC, PI, PE, and FA. Using lipid molecules selected by XGBoost combined with greedy algorithm and LASSO regression to establish model could accurately predict STEMI even in the more earlier stage.