A study of optimized model of serum tumor markers of colorectal cancer based on intelligent algorithms

作者：Hou, Panzhang; Luo, Jianchao^*; Zhang, Jianhua

来源：International Journal of Clinical and Experimental Medicine, 2016, 9(9): 17153-17164.

• Colorectal cancer
• serum marker
• optimized model
• logistic regression
• Bhattacharyya-SVM

Objective: To screen out serum tumor markers that has the best effects on the diagnosis, conditions monitoring and therapeutic evaluation in colorectal cancer patients and thus to provide the optimal model of simple, sensitive and noninvasive with serum markers in early auxiliary diagnosis and monitor of colorectal cancer patients. Methods: Literatures used in early supervision of colorectal cancer and published publicly during 1990-2013 were retrieved and Meta analysis was performed to screen serum markers that were of high detection value. Then 100 colorectal cancer patients and 50 patients with benign colorectal disease were recruited to test the level of serum markers which were screened out via Meta analysis with ELISA, and Logistic regression analysis, ROC curve and Bhattacharyya-SVM analysis were used to screen the optimal combination of serum markers of colorectal cancer. Results: The result of Meta analysis showed that 12 serum markers had certain correlation with colorectal cancer. At the same time, analysis of clinical data indicated that the area under the ROC curve (AUC) of CEA, CA19-9 and HSP60 when they were tested individually and jointly were 0.762, 0.752, 0.825 and 0.906. Their accuracy was 82.67%; sensitivity was 96.90%; and specificity was 90.57%. Bhattacharyya-SVM respectively adopted 12 indicators, 4 indicators (whose Bhattacharyya distance was more than 3) and 7 indicators (Bhattacharyya distance more than 2) to establish three SVM models. The classification accuracy rates of them were respectively 76.7%, 83.3% and 90.0%, sensitivity 80.0%, 85.0% and 90.0%, specificity 70.0%, 80.0% and 90.0%. The SVM prediction model established by CEA, CA50, CYFRA21-1, CA199, CA724, CA125 and UGT1A8 had the highest classification accuracy. Conclusion: The 12 serum markers such as CEA, CA242, and HSP60 are of high value for diagnosis of colorectal cancer. And the SVM models established in this study on basis of clinical validation results of these serum markers possess good predictive efficacy, which should be widely applied to clinical practice.

• 出版日期2016
• 单位郑州大学; 河南省人民医院