基于临床指标和病理指标的三种预测模型用于结直肠癌的预后分析研究
发布时间：2018-08-22 14:23
【摘要】：目的通过支持向量机模型(SVM)将临床指标和病理指标进行整合,分析其对结直肠癌(CRC)患者预后的预测价值。方法选取2002—2011年在上海市第十人民医院胃肠外科行结直肠肿瘤切除术的患者2 951例为研究对象。收集患者的临床指标(性别、年龄、肿瘤大小、肿瘤位置、组织病理类型、T分期、N分期、M分期、G分期)和病理指标[癌胚抗原(CEA)、增殖细胞核抗原(PCNA)、P53、CD_(34)、S-100、NM23、CerB-b2、P21、Ki-67]。按照随访信息是否缺失将患者分为两组,第一组为临床数据完整但随访信息缺失,共2 747例;第二组为临床数据及随访信息均完整,共204例。记录患者生存情况。第一组中,若某一患者有超过4个指标缺失,则该患者被剔除;在余下的患者中,若某一指标的缺失率30.0%,则该指标被剔除;进一步将少发病理类型患者剔除;计算临床指标和病理指标的关联性,随后,将所得关联性在第二组患者中进行验证,若该关联性在第二组中存在,则直接将第二组患者纳入第三组中;若该关联性在第二组中不存在,则采用SVM内部算法剔除部分患者,直至该关联性再次成立,将余下的第二组患者纳入第三组。对第二组中患者的病理指标进行统计,若某一指标的缺失率50.0%,则该指标被剔除。基于SVM对第三组数据进行处理,建立3种预测模型:SVM1基于临床指标、SVM2基于病理指标、SVM3基于临床指标和病理指标的汇总。结果第一组中,共834例患者缺失指标4个,其中性别、年龄、肿瘤位置、组织病理类型、P53、CD_(34)、S-100、CerB-b2、Ki-67共9个指标缺失率30.0%而被保留,剔除5例少见肿瘤患者后,共剩下829例患者。第一组患者年龄与CerB-b2表达情况存在关联性(P0.05)。第二组剔除105例患者后,余下99例患者,患者年龄与CerB-b2表达情况存在关联性(P0.05),将这99例患者纳入第三组。第二组中PCNA、P53、CD_(34)、S-100、CerB-b2共5个指标缺失率50.0%而被保留。第三组患者年龄与S-100、CerB-b2表达情况存在关联性(P0.05);M分期与PCNA表达情况存在关联性(P0.05)。第三组不同T分期、N分期患者生存曲线比较,差异有统计学意义(P0.05)。SVM1纳入9个临床指标(性别、年龄、肿瘤大小、肿瘤位置、组织病理类型、T分期、N分期、M分期、G分期),准确率为83.4%;SVM2纳入5个病理指标(PCNA、P53、CD_(34)、S-100、CerB-b2),准确率为78.8%;初始的SVM3纳入以上9个临床指标及5个病理指标,准确率为74.8%,通过最小冗余最大相关性(MRMR)法对指标进行进一步筛选,得到最终的SVM3,其纳入4个临床指标(肿瘤位置、组织病理类型、T分期、N分期)和2个病理指标(CD_(34)、CerB-b2),准确率为81.8%。不同风险SVM1、SVM2、SVM3患者生存曲线比较,差异有统计学意义(P0.05)。结论临床指标如年龄、M分期与病理指标如CerB-b2、S-100和PCNA存在一定的关联性;借助SVM模型将临床指标和病理指标进行整合分析可对CRC患者预后进行有效预测。
[Abstract]:Objective to analyze the prognostic value of support vector machine (SVM) model (SVM) in predicting the prognosis of colorectal cancer patients with (CRC) by integrating the clinical and pathological indexes. Methods from 2002 to 2011, 2 951 patients underwent colorectal tumor resection in Gastrointestinal surgery Department of the Tenth people's Hospital of Shanghai. The clinical parameters (sex, age, tumor size, tumor location, histopathological type, T staging and M stage G staging) and pathological indexes (carcinoembryonic antigen (CEA), proliferating cell nuclear antigen (PCNA) P53, CD34, S-100 NM23CerB-b2P21 Ki-67) were collected. The patients were divided into two groups according to whether the follow-up information was missing or not. The first group was complete clinical data but the follow-up information was missing in 2 747 cases, and the second group was clinical data and follow-up information complete, 204 cases. The patient's survival was recorded. In the first group, if more than 4 indexes were missing in one patient, the patient was eliminated; in the remaining group, if the missing rate of one index was 30.0, the index was removed. The correlation between clinical and pathological indexes was calculated, and then the correlation was verified in the second group. If the correlation existed in the second group, the second group was directly included in the third group. If the correlation does not exist in the second group, the SVM internal algorithm is used to remove some patients until the association is established again, and the remaining group of patients are included in the third group. The pathological indexes of the patients in the second group were counted, and if the missing rate of one index was 50. 0%, the index was eliminated. The third group of data was processed based on SVM, and three prediction models: SVM1 were established based on clinical index, SVM2 based on pathological index and SVM3 based on clinical and pathological indexes. Results in the first group, there were 4 missing indexes in 834 patients, including sex, age, tumor location, histopathologic type and the deletion rate of 9 indexes (30.0%). After 5 rare tumor patients were excluded, 829 patients were left. In the first group, there was a correlation between age and CerB-b2 expression (P0.05). The second group excluded 105 patients, the remaining 99 patients, the patient age and CerB-b2 expression were correlated (P0.05), the 99 patients were included in the third group. In the second group, the deletion rate of five indexes of PCNAn P53, CD34, S-100 and CerB-b2 was 50.0%. The age of patients in the third group was correlated with the expression of CerB-b2 (P0.05) and the expression of PCNA was correlated with the stage of M (P0.05). In the third group, the survival curve of patients with different T stage and N stage was compared, the difference was statistically significant (P0.05) .SVM1 included 9 clinical indexes (sex, age, tumor size, tumor location), The accuracy was 83.4% and SVM2 was included in five pathological indexes (PCNAn P53 CD34 / S-100 CerB-b2), the accuracy was 78.8%, the initial SVM3 included the above 9 clinical indexes and 5 pathological indexes. The accuracy rate was 74.8. The final SVM3 was obtained by (MRMR) with minimal redundancy and maximum correlation. It included four clinical indexes (tumor location, histopathological type, T staging) and two pathological indexes (CD34 / CerB-b2). The accuracy of SVM3 was 81.8%. There was significant difference in survival curve between SVM1 and SVM2 SVM3 patients with different risk (P0.05). Conclusion there is a certain correlation between the clinical parameters such as age M staging and pathological indexes such as CerB-b2P S-100 and PCNA, and the prognosis of CRC patients can be predicted effectively by integrating the clinical and pathological indexes with SVM model.
【作者单位】： 安徽医科大学上海临床学院;上海市第十人民医院胃肠外科;
【基金】：国家自然科学基金资助项目(81230057)
【分类号】：R735.34
本文编号：2197360
[Abstract]:Objective to analyze the prognostic value of support vector machine (SVM) model (SVM) in predicting the prognosis of colorectal cancer patients with (CRC) by integrating the clinical and pathological indexes. Methods from 2002 to 2011, 2 951 patients underwent colorectal tumor resection in Gastrointestinal surgery Department of the Tenth people's Hospital of Shanghai. The clinical parameters (sex, age, tumor size, tumor location, histopathological type, T staging and M stage G staging) and pathological indexes (carcinoembryonic antigen (CEA), proliferating cell nuclear antigen (PCNA) P53, CD34, S-100 NM23CerB-b2P21 Ki-67) were collected. The patients were divided into two groups according to whether the follow-up information was missing or not. The first group was complete clinical data but the follow-up information was missing in 2 747 cases, and the second group was clinical data and follow-up information complete, 204 cases. The patient's survival was recorded. In the first group, if more than 4 indexes were missing in one patient, the patient was eliminated; in the remaining group, if the missing rate of one index was 30.0, the index was removed. The correlation between clinical and pathological indexes was calculated, and then the correlation was verified in the second group. If the correlation existed in the second group, the second group was directly included in the third group. If the correlation does not exist in the second group, the SVM internal algorithm is used to remove some patients until the association is established again, and the remaining group of patients are included in the third group. The pathological indexes of the patients in the second group were counted, and if the missing rate of one index was 50. 0%, the index was eliminated. The third group of data was processed based on SVM, and three prediction models: SVM1 were established based on clinical index, SVM2 based on pathological index and SVM3 based on clinical and pathological indexes. Results in the first group, there were 4 missing indexes in 834 patients, including sex, age, tumor location, histopathologic type and the deletion rate of 9 indexes (30.0%). After 5 rare tumor patients were excluded, 829 patients were left. In the first group, there was a correlation between age and CerB-b2 expression (P0.05). The second group excluded 105 patients, the remaining 99 patients, the patient age and CerB-b2 expression were correlated (P0.05), the 99 patients were included in the third group. In the second group, the deletion rate of five indexes of PCNAn P53, CD34, S-100 and CerB-b2 was 50.0%. The age of patients in the third group was correlated with the expression of CerB-b2 (P0.05) and the expression of PCNA was correlated with the stage of M (P0.05). In the third group, the survival curve of patients with different T stage and N stage was compared, the difference was statistically significant (P0.05) .SVM1 included 9 clinical indexes (sex, age, tumor size, tumor location), The accuracy was 83.4% and SVM2 was included in five pathological indexes (PCNAn P53 CD34 / S-100 CerB-b2), the accuracy was 78.8%, the initial SVM3 included the above 9 clinical indexes and 5 pathological indexes. The accuracy rate was 74.8. The final SVM3 was obtained by (MRMR) with minimal redundancy and maximum correlation. It included four clinical indexes (tumor location, histopathological type, T staging) and two pathological indexes (CD34 / CerB-b2). The accuracy of SVM3 was 81.8%. There was significant difference in survival curve between SVM1 and SVM2 SVM3 patients with different risk (P0.05). Conclusion there is a certain correlation between the clinical parameters such as age M staging and pathological indexes such as CerB-b2P S-100 and PCNA, and the prognosis of CRC patients can be predicted effectively by integrating the clinical and pathological indexes with SVM model.
【作者单位】： 安徽医科大学上海临床学院;上海市第十人民医院胃肠外科;
【基金】：国家自然科学基金资助项目(81230057)
【分类号】：R735.34
【相似文献】
相关期刊论文 前1条
1 王钟秀,李义清,冯逸松;以病理指标的多因素分析对胃癌患者术后的予测[J];肿瘤;1983年03期
，本文编号：2197360
本文链接：https://www.wllwen.com/yixuelunwen/zlx/2197360.html
