基于R软件的逐步Cox回归模型拟合及Nomogram预测图绘制

目的建立基于R软件的逐步Cox回归模型拟合及列线图（Nomogram）绘制方法。方法以R软件的Survival包分别对R自带的Lung数据集228例晚期肺癌病人数据和美国监测、流行病、终点结局（surveillance, epidemiology and end results，SEER）数据库获取的6 341例胰腺癌病人数据拟合逐步Cox回归模型，采用rms包绘制列线图，以校准曲线对模型效果进行验证评价。结果Cox回归模型显示性别、ph.ecog评分是晚期肺癌病人生存的独立影响因素，年龄、肿瘤部位、肿瘤分化程度、TNM分期、肿瘤大小与淋巴结阳性率为胰腺癌病人生存的独立影响因素。
基于以上影响因素绘制的2个列线图可较准确地评估肺癌和胰腺癌病人1年、2年生存概率，校准曲线显示1年、2年实际生存概率与预测概率大致接近。结论R绘制列线图方便，可直观地预测个体生存概率。

How to control for unmeasured confounding in an observational time-to-event study with exposure incidence information:the treatment choice Cox model

ObjectiveTo establish the methods of fitting the stepwise Cox regression model and constructing Nomogram based on R software.MethodsDataset on built-in 228 advanced lung cancer patients and dataset including 6 341 pancreatic cancer patients were downloaded from Surveillance, Epidemiology and End Results(SEER)program of USA were adopted to fit the stepwise Cox regression model with survival package of R, respectively. The rms package was used to construct Nomogram. The forecasting effects of Nomogram were validated by calibration curve.ResultsThe multivariate Cox regression demonstrated that gender and score of ph.
ecog were independent prognostic factors for overall survival of the advanced lung cancer patients, and showed that age, location of carcinoma in pancreas, tumor grade, TNM stage, size of carcinoma together with lymph node ratio(LNR)were independent survival predictors for pancreatic cancer patients, respectively. The Nomograms based on above prognostic factors could precisely calculate the 1-year and 2-year survival probability of patients with advanced lung cancer and pancreatic cancer, respectively. The calibration curve demonstrated the actual 1-year and 2-year survival probability was close to the predicting probability.[Conclusions]R software can conveniently construct the Nomogram, visually predicting the survival probability of patients.