To construct a reliable prognostic signature for EC patients, EC patients were categorized into a high-risk group and a low-risk groups based on the median risk score. Heat map analysis (Fig.
4a) shown seven genes were differentially expressed in the high- and low-risk groups, which were ALDOB, NAXE, NDUFA6, ACAT2, CYP27A1, CARS2, and TIMM50. Patients were categorized into two groups based on the optimal cutoff value for the risk of LMRGs for diagnostic accuracy (Fig.
4b). More mortality events were observed in the high-risk group, suggesting that the increased risk of LMRGs reflected an unfavorable prognosis for EC patients (Fig.
4c). Principal component analysis (PCA) at different levels was employed to verify whether the risk states among different influencing factors were relatively independent. PCA demonstrated that samples with two risk scores were divided into two independent groups (Fig.
4d). Kaplan–Meier survival curves analyses showed that the high-risk subgroup had a shorter OS and PFS than the low-risk groups (Fig.
4e-f). To further investigate whether the prognostic signature could be an independent predictor of survival in EC, univariate and multivariate Cox regression analyses were performed. Univariate Cox regression analysis revealed that risk score was an independent predictor of poor OS in EC patients (HR = 4.778, 95% CI: 2.870–7.955). The results of multivariate COX regression analysis were the same (Fig.
4g-h). Remarkably, the predictive capability of these traditional clinical parameters (age, grade, and stage) was significantly lower than that of the LMGRs risk score. ROC curves of clinically relevant factors were conducted to assess the performance of the risk prediction model, which described excellent predictive capability. The AUCs for risk, age, grading and staging were 0.774, 0.613, 0.684 and 0.732, respectively (Fig.
4i). To prove the stability of the model, 503 EC patients (entire set) were randomly allocated to the train set (n = 352) and test set (n = 151) by seven-to-three ratio. The distribution of risk score (Supplement Fig.
1a), the survival status (Supplement Fig.
1b), the survival outcome(Supplement Fig.
1c-d), univariate and multivariate Cox regression analyses and the ROC curve (Supplement Fig.
1e-f) of EC patients between two groups of the train set and test set were constructed, respectively. All results in the two sets show no difference. The risk curves and scatter plots for the train and test sets implied mortality was positively related to the risk score in two sets. Kaplan–Meier survival analysis indicated that patients in the low-risk groups demonstrated better OS and PFS than patients in the high-risk groups in train and test sets. Univariate and multivariate Cox regression analyses and ROC curves were performed. Remarkably, the predictive capability of these traditional clinical parameters (age, grade, and stage) was significantly lower than that of the LMGRs risk score of train and test sets. ROC curves, assessing the accuracy of this risk model in two sets, indicated that the LMGRs model is reliable and precise. Risk score could be used as an effective prognostic marker in EC.