前列腺癌包膜侵犯预测模型的建立

Establishment of regression model predicting the probability of the capsular penetration of prostate cancer

目的 建立基于中国人样本资料的回归模型,根据术前临床指标预测前列腺癌包膜侵犯的发生概率.方法 回顾性分析2006年1月至2010年4月在我院接受前列腺癌根治术患者的资料.根据术后病理是否出现包膜侵犯分为器官局限组与局部晚期组并排除区域淋巴结发生转移的病例.收集两组病例的前列腺特异性抗原检测值、Gleason分级、临床分期等资料.根据术前临床指标及术后病理建立多重Logistic回归模型来预测前列腺癌包膜侵犯的发生率.利用受试者工作特征曲线评估模型的准确性.结果 器官局限组83例,年龄（66.8±5.8）岁,局部晚期组36例,年龄（66.0±6.8）,两组间年龄的差异无统计学意义（t=0.650,P=0.517）.局部晚期组前列腺特异性抗原检测值较高（Wilcoxon W=4562.0,P=0.016）,Gleason分级也较高（Wilcoxon W=4586.5,P=0.016）,临床分期较晚（Wilcoxon W=4444.5,P=0.001）,差异均具有统计学意义.多重Logistic回归模型判别公式为Logit P=0.488×Gleason分级＋0.104×临床分期-6.187.该模型自由度为2,似然比卡方检验χ2=11.263,P=0.004.利用本模型预测本组病例术后包膜侵犯概率的受试者工作特征曲线下面积为0.696,其95%置信区间为0.598～0.793,P=0.001.结论 基于中国前列腺癌人群资料建立的多重Logistic回归模型可以较为准确地预测包膜侵犯的发生概率,具有较高的内部准确性,可供临床决策时参考.

Objective To establish the regression model predicting the probability of the capsular penetration according to the Chinese prostate cancer samples. Methods Men enrolled in the Fudan University Shanghai Cancer Centre and undergoing radical prostatectomy between January 2006 and April 2010 were used to establish the predicting model. According to the pathology after radical prostatectomy, all cases were divided into two groups： organ confined disease group and locally advanced disease group, the difference of which were whether had the capsular penetration. The cases with regional lymph node metastasis were excluded. Serum prostate specific antigen level, Gleason grade, clinical stage were collected. Multiple Logistic regression model was established according to preoperative clinical data and postoperative pathological data to predict the incidence of capsular penetration. Receiver operating characteristic curve was used for the internal validation of the model. Results 83 Chinese men were identified in the organ confined disease group with the age of 66. 8 ± 5.8 years, and 36 in the locally advanced disease group with the age of 66. 0 ± 6. 8 years. The difference of the age between the two groups were of no statistic significance（t = 0. 650, P = 0. 517）. The serum prostate specific antigen level （Wilcoxon W =4562.0, P =0. 016）, Gleason score（Wilcoxon W =4586. 5, P =0. 016）, and clinical stage （Wilcoxon W= 4444. 5, P= 0. 001） of the locally advanced disease group were higher than the other group. The equation of the multiple Logistic regression model was Logit P = 0. 488 × Gleason score ＋ 0. 104 × clinical stage - 6. 187, with the freedom degree of two and the likelihood ratio χ2 test of 11. 263 （P= 0. 001） . The area under the ROC curve （AUC） for capsular penetration was 0.696 （P=0.001）, with the 95% onfidence interval of 0.598-0.793. Conclusion The multiple Logistic regression model based on the Chinese. population can accurately predict the probability of capsular penetration of the prostate cancer and work on well with high internal accuracy when clinical decisions are made.