摘要
目的试图建立去势抵抗性前列腺癌(CRPC)患者预后预测模型,并评估其准确性。方法回顾分析2007年1月至2012年1月在中山大学附属第三医院就诊的去势抵抗性前列腺癌病例,收集入组时年龄、基线PSA值、PSA倍增时间、血红蛋白、碱性磷酸酶、白蛋白、是否采用多西他赛三周化疗方案治疗等因素,运用Cox单因素分析筛选潜在预后预测变量后,分别以Cox多因素回归分析与部分指数回归人工神经网络建立预后预测模型,以ROC曲线下面积评估模型准确性。结果 Cox多因素回归模型的ROC曲线下面积为0.69,部分指数回归人工神经网络模型ROC曲线下面积为0.84。结论通过部分指数人工神经网络建立的预后预测模型可以纳入不符合比例风险假定的临床数据,改善预测准确度,较好的预测去势抵抗性前列腺癌患者的预后情况。
Objectives To ereat and evaluate the efficacy of prognosis prediction model for castration resistant prostate cancer （CRPC） patients. Methods Records of patients diagnosed with CRPC from January 2007 to January 2012 in the Third Affiliated Hospital of Sun Yat-sen University were reviewed. Data of age, PSA baseline, PSA doubling time, hemoglobin, alkaline phosphatase, albumin, treatment regime, liver metastasis and other candidate prognosis factors were collected. After excluding unrelated factors with Cox univariate regression, prognosis models were created with Cox multivariate regression and partial logistic artificial neural network （PLANN）, respectively. Both models were evaluated with area under ROC curve method. Results Area under ROC curve for Cox multivariate regression model was 0.69, and 0.84 for PLANN model. Conclusions Variables that don＇t satisfy proportional hazard assumption can be included with PLANN method, improving the efficacy of the model, and providing more accurate prediction on prognosis of CRPC patients.
出处
《中华腔镜泌尿外科杂志（电子版）》
2013年第6期7-10,共4页
Chinese Journal of Endourology(Electronic Edition)
基金
国家自然科学基金(81172430)
中山大学临床研究5010计划项目(2007028)
关键词
前列腺癌
预后
人工神经网络
Prostate cancer
Prognosis
Artificial neural networks
