前列腺特异性抗原水平与前列腺癌患者生存数据的联合建模分析  

Joint-modeling of prostate-specific antigen levels and survival data of prostate cancer patients

作　　者：吴梦娟 高春洁 张涛[1] 赵婷[3] 袁帅[4] 牛越[4] 王蕾[2] WU Mengjuan;GAO Chunjie;ZHANG Tao;ZHAO Ting;YUAN Shuai;NIU Yue;WANG Lei(Xinjiang Medical University,Urumqi,Xinjizng 830017,Chinz;Affiliated Cancer Hospital of Xinjiang Medical University,Urumqi,Xinjizng 830011,Chinz)

机构地区：[1]新疆医科大学公共卫生学院,新疆乌鲁木齐830017 [2]新疆医科大学医学工程技术学院,新疆乌鲁木齐830017 [3]新疆医科大学附属肿瘤医院病案管理科,新疆乌鲁木齐830011 [4]新疆医科大学附属肿瘤医院泌尿科,新疆乌鲁木齐830011

出　　处：《中华肿瘤防治杂志》2024年第15期933-940,共8页Chinese Journal of Cancer Prevention and Treatment

基　　金：国家自然科学基金(12061079);省部共建中亚发病成因与防治国家重点实验室(SKL-HIDCA-2022-JZ1);“天山英才”青年科技创新人才培养(2022TSYCCX0108)。

摘　　要：目的探讨前列腺特异性抗原(PSA)水平的动态变化对晚期前列腺癌患者生存预后的影响,为前列腺癌患者个性化治疗提供一定的理论依据。方法本研究为回顾性队列研究。连续性收集2011-01-01-2017-12-31新疆医科大学附属肿瘤医院经病理学检查确诊为前列腺癌的176例患者作为研究对象,根据治疗方案分为比卡鲁胺联合戈舍瑞林组(n=126)和氟他胺联合戈舍瑞林组(n=50)。采用线性混合效应模型和Cox比例风险模型分别拟合晚期前列腺癌患者血清PSA水平的动态变化及其生存数据,进而根据共享随机效应构建极大似然估计和贝叶斯估计法下的联合模型。通过赤池信息准则(AIC)、贝叶斯信息准则(BIC)以及对数似然函数(LLF)值评估2类联合模型的拟合优度。采用受试者工作特征曲线下面积(AUC)和预测误差(PE)比较2类联合模型的预测性能。通过独立样本t检验、Mann-Whitney U秩和检验、χ^(2)检验或Fisher确切概率法比较不同治疗方案下患者基线数据的组间差异。结果共纳入176例前列腺癌患者,年龄45~90岁,平均年龄(71.76±7.86)岁,随访1.30~36.77个月。极大似然估计下的联合模型结果显示,相比于患者血清PSA水平未增加时,当PSA水平随时间增至10倍后,患者死亡风险增加0.94倍(HR=1.94,95%CI:1.74~2.16,P<0.001);贝叶斯估计下的联合模型结果显示,相比于患者血清PSA水平未增加时,当PSA水平随时间增至10倍后,患者死亡风险增加1.18倍(HR=2.18,95%CI:1.77~2.73,P<0.001)。此外,极大似然估计下的联合模型展现出更好的拟合优度(AIC=3265.01,BIC=3303.06,LLF=-1620.51),而贝叶斯估计下联合模型的AUC值(0.70~0.88)更大、PE值(0.04~0.10)更小,提示其具有更强的预测性能。结论晚期前列腺癌患者血清PSA水平升高是其生存预后的危险因素,临床上应密切监测前列腺癌患者血清PSA水平的动态变化,以便更准确地制定个性化治疗方案。Objective To explore the effect of dynamical changes of prostate-specific antigen(PSA)on the survival prognosis of advanced prostate cancer patients,and provide a theoretical basis for individualized treatment of prostate cancer patients.Methods This study was a retrospective cohort analysis.A total of 176 patients diagnosed with prostate cancer through pathological examination at the Affiliated Cancer Hospital of Xinjiang Medical University from January 01,2011 to December 31,2017 were continuously collected as the study subjects.According to the treatment regimen,the patients were divided into bicalutamide combined with goserelin group(n=126)and flutamide combined with goserelin group(n=50).Linear mixed effects model and Cox proportional risk model were separately used to fit the dynamical changes of serum PSA and survival data of advanced prostate cancer patients.Under the shared random effects,two joint models based on maximum likelihood and Bayesian estimation methods were further constructed,respectively.In addition,the goodness of fit for two joint models were separately assessed by the Akaike information criterion(AIC),Bayesian information criterion(BIC),and Log-Likelihood function(LLF)values.The area under the receiver operating characteristic curve(AUC)and prediction error(PE)were applied to compare the predictive performance of two joint models,respectively.The independent sample t test,Mann-Whitney U rank sum test,χ^(2) test,or Fisher exact probability method were applied in comparing the baseline data differences between groups with different treatment regimens.Results A total of 176 patients with prostate cancer were included,aged 45—90 years,with an average age of(71.76±7.86)years,and the patients were followed-up for 1.30—36.77months.It was shown from the joint model based on maximum likelihood estimation method that when the serum PSA increases tenfold over time,the risk of death would increase by 0.94 times(HR=1.94,95%CI:1.74—2.16,P<0.001)by comparing with the patients whose serum PSA le

关 键 词：前列腺癌 前列腺特异性抗原 纵向数据 生存预后 联合模型 

分 类 号：R737.25[医药卫生—肿瘤]