基于机器学习的肾癌患者术后复发风险预测模型的构建与评价

Recurrence risk prediction models of postoperative patients with renal cell carcinoma based on machine learning

摘要

目的 探索肾癌患者术后复发的影响因素，构建复发风险的机器学习预测模型并评价其效能。方法 回顾性收集山西医科大学第一医院泌尿外科2013-2021年间915例行手术治疗的肾癌患者的临床资料。将纳入研究的患者按7∶3的比例随机划分为训练集（n=510）和验证集（n=218）。在训练集中使用最小绝对收缩和选择算子（LASSO）回归算法筛选重要变量，并构建肾癌患者术后复发风险的机器学习预测模型，在验证集中结合模型的受试者工作特征（ROC）曲线下面积（AUC）和准确率、F1值等指标对模型效能进行比较。结果 LASSO回归筛选出影响肾癌患者术后复发的风险因素有吸烟史、肿瘤最大径、N分期、Fuhrman分级、凝血酶时间和纤维蛋白原水平，据此构建的logistic模型、决策树模型、随机森林模型和贝叶斯模型在验证集中的AUC值分别为0.862、0.792、0.843和0.861，准确率分别为0.917、0.908、0.904和0.927，F1值分别为0.357、0.286、0.323和0.600。综合比较，贝叶斯模型的性能最稳定且具有较好的区分度。结论 在本数据集中，基于贝叶斯算法的机器学习预测模型效能较好，可为临床决策提供参考。

Abstract

Objective To explore the influencing factors of recurrence in postoperative patients with renal cell carcinoma, construct machine learning prediction models and evaluate their performance. Methods Clinical data of 915 patients with renal cell carcinoma treated in our hospital during 2013 and 2021 were retrospectively collected. The data were randomly divided into a training set (n=510) and a validation set (n=218) in a 7∶3 ratio. In the training set, LASSO regression algorithm was used to screen important variables, and machine learning prediction models were constructed to predict the recurrence risk. In the validation set, the effectiveness of the models was compared combined with the area under receiver operating characteristic curve (AUC), accuracy rate, F1 value and other indicators. Results LASSO regression screened out the risk factors, including smoking history, tumor size, N stage, Fuhrman grade, thrombin time and fibrinogen, based on which, the logistic model, decision tree model, random forest model, and Bayes model were constructed. In the validation set, the AUC of the above 4 models was 0.862, 0.792, 0.843 and 0.861, respectively; the accuracy was 0.917, 0.908, 0.904 and 0.927, respectively; F1 value was 0.357, 0.286, 0.323 and 0.600, respectively. The Bayes model had the most stable performance and best differentiation. Conclusion In this dataset, the prediction model based on Bayes algorithm has a good performance and can provide reference for clinical decision making.

关键词

肾细胞癌 / 复发 / 机器学习 / 预测模型 / 逻辑回归 / 决策树 / 随机森林 / 贝叶斯

Key words

renal cell carcinoma / recurrence / machine learning / prediction model / logistic / decision tree / random forest / Bayes

参考文献

[1] YAO L, DU Y C, DONG W, et al. Chinese quality control indices for standardized diagnosis and treatment of renal cancer (2022 edition)[J]. J Natl Cancer Cent, 2024, 4(1): 6-13.

[2] SUNG H, FERLAY J, SIEGEL RL, et al. Global Cancer Statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries[J]. CA Cancer J Clin, 2021, 71(3): 209-249.

[3] HAN B F, ZHENG R S, ZENG H M, et al. Cancer incidence and mortality in China, 2022[J]. J Natl Cancer Cent, 2024, 4(1): 47-53.

[4] 张欣悦， 郑霞， 刘洋. 肾癌术后局部复发不同治疗模式的疗效分析[J]. 中华泌尿外科杂志， 2023， 44（ 1）： 1-6.

[5] USHER-SMITH JA, LI L, ROBERTS L, et al. Risk models for recurrence and survival after kidney cancer: a systematic review[J]. BJU Int, 2022, 130(5): 562-579.

[6] DHIMAN P, MA J, ANDAUR NAVARRO CL, et al. Over-interpretation of findings in machine learning prediction model studies in oncology: a systematic review[J]. J Clin Epidemiol, 2023, 157: 120-133.

[7] 王雷， 李怀康， 彭程， 等. 基于SEER数据库分析老年局限性肾癌术后患者的临床特征及预后影响因素[J]. 现代泌尿外科杂志， 2024， 29（2）： 168-174.

[8] 邓兴宇， 杨楠， 薛宇廷， 等. 基于危险因素和常规实验室指标的子痫前期风险预测模型研究[J]. 临床检验杂志， 2021， 39（8）： 566-572.

[9] CORNELIUS ME, LORETAN CG, JAMAL A, et al. Tobacco product use among adults-United States, 2021[J]. MMWR Morb Mortal Wkly Rep, 2023, 72(18): 475-483.

[10] MOCH H, CUBILLA AL, HUMPHREY PA, et al. The 2016 WHO classification of tumours of the urinary system and male genital organs part a: renal, penile, and testicular tumours[J]. Eur Urol, 2016, 70(1): 93-105.

[11] AMIN MB, GREENE FL, EDGE SB, et al. The eighth edition AJCC cancer staging manual: continuing to build a bridge from a population-based to a more “personalized” approach to cancer staging[J]. CA Cancer J Clin, 2017, 67(2): 93-99.

[12] FUHRMAN SA, LASKY LC, LIMAS C. Prognostic significance of morphologic parameters in renal cell carcinoma[J]. Am J Surg Pathol, 1982, 6(7): 655-663.

[13] ROBERT T. Regression shrinkage and selection via the Lasso[J]. J R Stat Soc Series B Stat Methodol, 1996, 58(1): 267-288.

[14] SCHOBER P, VETTER TR. Logistic regression in medical research[J]. Anesth Analg, 2021, 132(2): 365-366.

[15] QUINLAN JR. Induction of decision trees[J]. Mach Learn, 1986, 1: 81-106.

[16] BREIMAN L. Random Forests[J]. Mach Learn, 2001, 45(1): 5-32.

[17] RISHI. An empirical study of the naive Bayes classifier[J]. J Univers Comput Sci, 2001, 1(2): 127.

[18] 周航宇， 冯国双. 医学论文中logistic回归分析常见逻辑错误与案例分析[J]. 中华全科医师杂志， 2020， 19（12）： 1194-1196.

[19] HE P, MA Y, WU Y, et al. Exploring PANoptosis in breast cancer based on scRNA-seq and bulk-seq[J]. Front Endocrinol, 2023, 14: 1164930.

[20] 陈旭， 尚攀峰. 局限性肾癌术后复发与手术相关因素的研究进展[J]. 中国微创外科杂志， 2022， 28（11）： 903-908.

[21] 韩雪阳， 高瞻， 刘勇. 术前全身炎症反应指数和纤维蛋白原在肾癌切除术后肿瘤预后的影响[J]. 临床泌尿外科杂志， 2021， 36（7）： 567-572.

[22] YUEJUN T, MEI H, SUOSHI J, et al. Clinical and prognostic effect of plasma fibrinogen in renal cell carcinoma: a meta-analysis[J]. Int BR, 2017: 20179591506.

[23] SHI X, WANG X, YAO W, et al. Mechanism insights and therapeutic intervention of tumor metastasis: latest developments and perspectives[J]. Signal Transduct Target Ther, 2024, 9(1): 192.

[24] 谢瑞， 洪含霞. 血栓弹力图评估肾肿瘤患者术前凝血状态的临床价值研究[J]. 国际泌尿系统杂志， 2021， 41（2）： 287-290.

[25] GUYON I, ELISSEEFF A. An introduction to variable and feature selection[J]. J Mach Learn Res, 2003, 3: 1157-1182.

[26] WILLIAMS CKI, BARBER D. Bayesian classification with Gaussian processes[J]. IEEE Trans Pattern Anal Mach Intell, 1998, 20(12): 1342-1351.