目的:筛选影响壶腹癌(AC)患者根治术后复发风险的危险因素。然后构建基于Lasso-Cox回归的风险预测模型并进行可视化。
方法:收集2011年1月至2022年1月在河北省肿瘤医院接受胰十二指肠切除术治疗的162例患者的临床资料。训练组采用Lasso回归法筛选复发的危险因素。使用Delong检验比较Lasso-Cox回归和随机生存森林(RSF)模型,以确定基于风险因素的最佳模型。最后,使用来自验证组的临床数据对所选模型进行验证.
结果:患者分为两组,训练和验证的比例为7:3。通过Lasso回归筛选的变量,如CA19-9/GGT,AJCC第8版TNM分期,淋巴结侵入,差异化,肿瘤大小,CA19-9,性别,GPR,PLR,饮酒史,和并发症,在训练组中使用Lasso-Cox回归模型(C指数=0.845)和RSF模型(C指数=0.719)进行建模。根据Delong检验,我们选择了Lasso-Cox回归模型(P=0.019),并用时间相关的接收器工作特性曲线(tdROC)验证了其性能,校正曲线,和决策曲线分析(DCA)。训练组1年、3年和5年的tdROC曲线下面积分别为0.855、0.888和0.924,验证组分别为0.841、0.871和0.901,分别。校准曲线表现良好,以及DCA使用预测模型显示更高的净收益和更广泛的阈值概率。列线图可视化用于显示所选模型的结果。
结论:该研究基于Lasso-Cox回归模型建立了预测AC患者复发的列线图。与通过其他方法构建的列线图相比,这一个更健壮和准确。
OBJECTIVE: To screen the risk factors affecting the recurrence risk of patients with ampullary carcinoma (AC)after radical resection, and then to construct a model for risk prediction based on Lasso-Cox regression and visualize it.
METHODS: Clinical data were collected from 162 patients that received pancreaticoduodenectomy treatment in Hebei Provincial Cancer Hospital from January 2011 to January 2022. Lasso regression was used in the training group to screen the risk factors for recurrence. The Lasso-Cox regression and Random Survival Forest (RSF) models were compared using Delong test to determine the optimum model based on the risk factors. Finally, the selected model was validated using clinical data from the validation group.
RESULTS: The patients were split into two groups, with a 7:3 ratio for training and validation. The variables screened by Lasso regression, such as CA19-9/GGT, AJCC 8th edition TNM staging, Lymph node invasion, Differentiation, Tumor size, CA19-9, Gender, GPR, PLR, Drinking history, and Complications, were used in modeling with the Lasso-Cox regression model (C-index = 0.845) and RSF model (C-index = 0.719) in the training group. According to the Delong test we chose the Lasso-Cox regression model (P = 0.019) and validated its performance with time-dependent receiver operating characteristics curves(tdROC), calibration curves, and decision curve analysis (DCA). The areas under the tdROC curves for 1, 3, and 5 years were 0.855, 0.888, and 0.924 in the training group and 0.841, 0.871, and 0.901 in the validation group, respectively. The calibration curves performed well, as well as the DCA showed higher net returns and a broader range of threshold probabilities using the predictive model. A nomogram visualization is used to display the results of the selected model.
CONCLUSIONS: The study established a nomogram based on the Lasso-Cox regression model for predicting recurrence in AC patients. Compared to a nomogram built via other methods, this one is more robust and accurate.