基于随机森林算法的胰腺癌术后预测模型构建

doi:10.13481/j.1671-587X.20220220

Abstract

Abstract: Objective

To establish a prediction model of 5-year survival after operation in the pancreatic cancer patients by random forest algorithm， and to provide guidance for prognosis evaluation after operation in the pancreatic cancer patients.

Methods

Surveillance， Epidemzology，and End Results Program（SEER） database was used to collect 42，618 data of the pancreatic cancer patients and their prognosis which met the requirements of this study. After data screening， 4，020 data of the patients were finally included. The subjects were randomly divided into training set and test set， and the characteristic variables were selected by means of χ^２test， multivariate Logistic regression analysis and random forest variable importance ranking； Synthetic Minority Oversampling Technique（SMOTE） was used to adjust the training set to a balanced data set； Based on the balanced data set， the prediction model was constructed by using random forest algorithm； Using the test set， based on the four evaluation indexes of G-mean index， sensitivity， specificity， and area under receiver operating characteristic（ROC） curve（AUC）， the prediction model was evaluated by comparing with logistic regression analysis， support vector machine， decision tree and artificial neural network algorithm.

Results

The original training set contained 2，814 samples， of which 196 patients with a survival time ≥5 years， accounting for about 1/13. It was an unbalanced data set. The balanced data set was obtained after adjustment by SMOTE， and the number of two classification samples was basically balanced. The final variables included in the model were age， race， primary tumor location， tumor differentiation degrees， radiotherapy or not， T stage， N stage， marital status， tumor size and lymph node positive rate. The G-mean index of pancreatic cancer prediction model based on random forest algorithm was 0.830， and the AUC was 0.833， which was better than Logistic regression analysis， support vector machine， decision tree and artificial neural network.

Conclusion

The prediction performance of prediction model of pancreatic cancer based on random forest algorithm is better than other common machine learning methods in predicting the 5-year survival rate of patients with pancreatic cancer. It can provide a basis for clinicians to improve the prognosis and survival of patients with pancreatic cancer.

Key words: Pancreatic neoplasms, Random forest algorithm, Prediction model, Prognosis, Model comparison

CLC Number:

R735.9

Chengsheng LI,Qihan BAO,Xiaoyan HAO,Qingzhong PAN,Suzhen WANG,Fuyan SHI. Establishment of prediction model for postoperative pancreatic cancer based on random forest algorithm[J].Journal of Jilin University(Medicine Edition), 2022, 48(2): 426-435.

Figures/Tables 9

Tab. 1

Tab. 2

Tab. 3

Tab. 4

Tab. 5

Tab. 6

Tab. 7

Tab. 8

Tab. 9

References 29

1	LUCHINI C， CAPELLI P， SCARPA A. Pancreatic ductal adenocarcinoma and its variants［J］. Surg Pathol Clin， 2016， 9（4）： 547-560.
2	SIEGEL R L， MILLER K D， FUCHS H E， et al. Cancer statistics，2021［J］.CA Cancer J Clin，2021，71（1）： 7-33.
3	曹毛毛，陈万青. 中国恶性肿瘤流行情况及防控现状［J］. 中国肿瘤临床， 2019， 46（3）： 145-149.
4	杨永超，李宜雄. 胰腺癌外科治疗的历史和现状［J］. 中国普通外科杂志， 2018， 27（3）： 269-283.
5	肿瘤医学论坛.2020年全球癌症最新数据解读［J］. 中国肿瘤临床与康复， 2021， 28（3）： 301.
6	MAHAJAN U M， LANGHOFF E， GONI E， et al. Immune cell and stromal signature associated with progression-free survival of patients with resected pancreatic ductal adenocarcinoma［J］. Gastroenterology， 2018， 155（5）： 1625-1639.e2.
7	STRIJKER M， SOER E C， DE PASTENA M， et al. Circulating tumor DNA quantity is related to tumor volume and both predict survival in metastatic pancreatic ductal adenocarcinoma［J］. Int J Cancer， 2020， 146（5）： 1445-1456.
8	YU S， LI Y， LIAO Z， et al. Plasma extracellular vesicle long RNA profiling identifies a diagnostic signature for the detection of pancreatic ductal adenocarcinoma［J］. Gut， 2020， 69（3）： 540-550.
9	张峻烽，刘淞淞，王槐志. 基于WGCNA和ss GSEA的胰腺癌预后模型构建［J］.中国临床新医学，2020，13（11）： 1084-1090.
10	周放. 营养相关指标预测胰腺癌预后的模型构建和验证［D］. 苏州：苏州大学， 2020.
11	王月，赵茂先. 基于最大最小爬山算法的肺癌预后模型［J］.山东科技大学学报（自然科学版），2020，39（2）： 105-110.
12	张灵敏，路宁，袁慧，等. 影响胰腺癌特异性生存率的相关因素分析［J］.山西医科大学学报，2019，50（11）： 1531-1537.
13	吴君君. 利用机器学习算法建立胰腺癌远处转移预测模型［D］. 南昌：南昌大学， 2020.
14	衡阳，林海. 胰腺癌预后因素的研究进展［J］. 世界最新医学信息文摘（连续型电子期刊），2020，20（8）：164-165， 169.
15	程斌，江振辉，陈晓鹏，等. 胰腺恶性肿瘤患者术后预后风险模型的构建［J］. 中国普外基础与临床杂志， 2020， 27（12）： 1506-1514.
16	纪宇楠. 基于随机森林构建滤泡型甲状腺癌远处转移预测模型［D］. 沈阳：中国医科大学， 2018.
17	王奕森，夏树涛. 集成学习之随机森林算法综述［J］. 信息通信技术， 2018， 12（1）： 49-55.
18	HIDALGO M. Pancreatic cancer［J］. N Engl J Med， 2010， 362（17）： 1605-1617.
19	QIU W， DUAN N， CHEN X， et al. Pancreatic ductal adenocarcinoma： machine learning-based quantitative computed tomography texture analysis for prediction of histopathological grade［J］. Cancer Manag Res， 2019， 11： 9253-9264.
20	马作红，宫学宇，尚海，等. 基于控制营养状况评分生存预测模型构建及其预测晚期胰腺癌患者预后价值［J］. 临床军医杂志， 2020， 48（7）： 787-790.
21	BRADLEY A， MEER RVAN DER， MCKAY C J. A prognostic Bayesian network that makes personalized predictions of poor prognostic outcome post resection of pancreatic ductal adenocarcinoma［J］. PLoS One， 2019， 14（9）： e0222270.
22	邢晓蕊. 基于机器学习算法的胰腺癌诊断模型研究［D］. 长春：吉林大学， 2018.
23	ILIC M， ILIC I. Epidemiology of pancreatic cancer［J］. World J Gastroenterol， 2016， 22（44）： 9694-9705.
24	YADAV D， LOWENFELS A B. The epidemiology of pancreatitis and pancreatic cancer［J］. Gastroenterology， 2013， 144（6）： 1252-1261.
25	KANG M J， JANG J Y， CHANG Y R， et al. Revisiting the concept of lymph node metastases of pancreatic head cancer： number of metastatic lymph nodes and lymph node ratio according to N stage［J］. Ann Surg Oncol， 2014， 21（5）： 1545-1551.
26	罗国培，倪泉兴，虞先濬. 胰腺癌中淋巴结转移率的临床应用价值［J］. 中华消化外科杂志， 2015， 14（8）： 686-688.
27	方向，麦刚，刘忠亮，等. 胰腺癌淋巴结转移情况对预后的影响［J］. 肝胆胰外科杂志， 2017， 29（1）： 27-31.
28	SALA ELARRE P， OYAGA-IRIARTE E， YU K H， et al. Use of machine-learning algorithms in intensified preoperative therapy of pancreatic cancer to predict individual risk of relapse［J］. Cancers， 2019，11（5）： 606.
29	WARSCHKOW R， WIDMANN B， BEUTNER U， et al. The more the better-lower rate of stage migration and better survival in patients with retrieval of 20 or more regional lymph nodes in pancreatic cancer： a population-based propensity score matched and trend SEER analysis［J］. Pancreas， 2017， 46（5）： 648-657.

Variable	Assignment	Records in SEER database	Number	Percentage(η/%)
Age(year)	1	≤49	253	6.3
	2	50－59	789	19.6
	3	60－69	1 414	35.2
	4	70－79	1 180	29.4
	5	≥80	384	9.6
Gender	1	Female	1 962	48.8
	2	Male	2 058	51.2
Race	1	Black	413	10.3
	2	White	3 263	81.2
	3	Other	344	8.6
PrimarySite	1	Pancreatic head	3 063	76.2
	2	Pancreatic body	274	6.8
	3	Pancreatic tail	355	8.8
	4	Pancreatic overlap	179	4.5
	5	Other	149	3.7
Grade	1	GradeⅠ	344	8.6
	2	GradeⅡ	2 022	50.3
	3	GradeⅡ	1 619	40.3
	4	GradeⅣ	35	0.9
Chemotherapy	1	No	1 157	28.8
	2	Yes	2 863	71.2
Radiotherapy	1	No	2 759	68.6
	2	Yes	1 261	31.4
Number of lymph node dissections	1	0－3	180	4.5
	2	≥4	3 840	95.5
T stage	1	T1	154	3.8
	2	T2	354	8.8
	3	T3	3 296	82.0
	4	T4	216	5.4
N stage	1	N0	1 062	26.4
	2	N1	2 958	73.6
M stage	1	M0	3 785	94.2
	2	M1	235	5.8
Marital status	1	Married	2 604	64.8
	2	Single	514	12.8
	3	Other	902	22.4
Tumor size(l/mm)	1－540	－	－	－
Lymph node positive ratio	0－1	－	－	－
“－”:No data.

Variable	Training set （n=2 814）	Test set （n=1 206）	Statistic	P
Age(year)			Z=－0.553	0.580
≤49	168	77
50－59	561	210
60－69	990	441
70－79	821	375
≥80	274	103
Gender			χ^２=0.088	0.767
Female	1 388	601
Male	1 426	605
Race			χ^２=1.233	0.540
Black	285	135
White	2 278	970
Other	251	101
Primary site			χ^２=4.405	0.354
Pancreatic head	2 145	938
Pancreatic body	192	78
Pancreatic tail	247	110
Pancreatic overlap	124	37
Other	106	43
Grade			Z=－0.234	0.815
GradeⅠ	246	110
GradeⅡ	1 433	612
GradeⅢ	1 111	474
GradeⅣ	24	10
Chemotherapy			χ^２=0.011	0.915
No	812	346
Yes	2 002	860
Radiotherapy			χ^２=0.658	0.417
No	1 922	808
Yes	892	398
Number of lymph node dissections			χ^２=0.262	0.609
0－3	125	58
≥4	2 689	1 148
T stage			Z=－0.991	0.322
T1	105	40
T2	259	97
T3	2 293	1 004
T4	157	65
N stage			χ^２=0.193	0.661
N0	728	320
N1	2 086	886
M stage			χ^２=0.683	0.409
M0	2 646	1 142
M1	168	64
Marital status			χ^２=4.613	0.100
Married	1 826	763
Single	352	181
Other	636	262
Tumor size	2 814	1 206	Z=－0.002	0.998
Lymph node positive ratio	2 814	1 206	Z=－0.257	0.797

Variable	Training set		Variable	Training set
Variable	Statistic	P	Variable	Statistic	P
Age（year）	Z=－2.841	0.004	Chemotherapy	χ^２=11.289	0.001
≤49			No
50－59			Yes
60－69			Radiotherapy	χ^２=11.033	0.001
70－79			No
≥80			Yes
Gender	χ^２=0.118	0.731	Number of lymph node dissections	χ^２=0.064	0.800
Female			0－3
Male			≥4
Race	χ^２=0.066	0.968	T stage	Z=－5.501	<0.001
Black			T1
White			T2
Other			T3
Primary site	χ^２=1.025	0.906	T4
Pancreatic head			N stage	χ^２=110.965	<0.001
Pancreatic body			N0
Pancreatic tail			N1
Pancreatic overlap			M stage	χ^２=4.387	0.340
Other			M0
Grade	Z=－12.723	<0.001	M1
GradeⅠ			Marital status	χ^２=2.160	0.340
GradeⅡ			Married
GradeⅢ			Single
GradeⅣ			Other
			Tumor size	Z=－6.460	<0.001
			Lymph node positive ratio	Z=－10.771	<0.001

Variable	Training set
Variable	B	Std.Error	Wald	P	Exp(B)	95% Exp(B)
Grade	－0.425	0.120	12.599	<0.001	0.654	(0.517,0.822)
Chemotherapy	0.597	0.211	7.982	0.005	1.816	(1.201,2.784)
Radiotherapy	0.335	0.169	3.937	0.047	1.398	(1.004,1.947)
T stage	－0.268	0.119	5.048	0.025	0.765	(0.606,0.966)
N stage	－0.595	0.214	7.720	0.005	0.552	(0.363,0.839)
Tumor size	－0.023	0.006	12.570	<0.001	0.978	(0.965,0.990)
Lymph node positive rate	－3.980	0.927	18.432	<0.001	0.019	(0.003,0.115)

SMOTE dataset	Perc.over	Perc.under	Negative sample size	Positive sample size	Class.error(－)	Class.error(+)	OOB error(%)
Training set	－	－	2 618	196	0.001	0.995	7.00
1	1 300	100	2 548	2 744	0.082	0.122	10.26
2	1 200	100	2 352	2 548	0.085	0.113	9.94
3	1 100	100	2 156	2 352	0.089	0.115	10.27
4	1 000	100	1 960	2 156	0.102	0.122	11.20
5	900	100	1 764	1 960	0.109	0.132	12.11
6	800	100	1 568	1 764	0.116	0.128	12.24
7	700	100	1 372	1 568	0.141	0.140	14.01
8	600	100	1 176	1 372	0.153	0.144	14.80
9	500	100	980	1 176	0.173	0.147	15.91
10	400	100	784	980	0.203	0.158	17.80
11	300	100	588	784	0.248	0.124	17.71
12	200	100	392	588	0.316	0.151	21.73
13	100	100	v196	392	0.398	0.110	20.58
14	600	200	2 352	1 372	0.064	0.219	12.14
15	500	200	1 960	1 176	0.068	0.222	12.60
16	400	200	1 568	980	0.101	0.237	15.31
17	300	200	1 176	784	0.102	0.236	15.56
18	200	200	784	588	0.159	0.260	20.26
19	100	200	392	392	0.306	0.242	27.42
“－”： No data.

Establishment of prediction model for postoperative pancreatic cancer based on random forest algorithm

RichHTML

PDF (PC)

Like

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 9

References 29

Related Articles 15

Metrics

Comments

Recommended 0

Variable	Mean decrease accuracy	Mean decrease Gini
Lymph node positive rate	0.147	828.022
N stage	0.133	224.008
Tumor size	0.100	573.534
T stage	0.057	149.560
Age	0.032	151.944
Grade	0.028	105.018
Primary site	0.026	112.608
Marital status	0.014	75.512
Radiotherapy	0.013	42.982
Race	0.012	62.160
Chemotherapy	0.010	41.365
Gender	0.007	39.164
Number of lymph node dissections	0.005	24.583
M stage	0.002	12.427

Model	Variable set	Sensitivity	Specificity	G-mean	AUC
RF1	Variable set 1：	0.888	0.774	0.829	0.831
	Lymph node positive rate
	N stage
	Tumor size
	T stage
	Age
	Grade
	PrimarySite
	Marital status
	Radiotherapy
	Race
	Chemotherapy
RF2	Variable set 2：	0.891	0.774	0.830	0.833
	Lymph node positive rate
	N stage
	Tumor size
	T stage
	Age
	Grade
	PrimarySite
	Marital status
	Radiotherapy
	Race
RF3	Variable set 3：	0.887	0.774	0.829	0.830
	Lymph node positive rate
	N stage
	Tumor size
	T stage
	Age
	Grade
	PrimarySite
	Marital status
	Radiotherapy
RF4	Variable set 4：	0.889	0.750	0.817	0.820
	Lymph node positive rate
	N stage
	Tumor size
	T stage
	Age
	Grade
	PrimarySite
	Marital status

Model	Variable set	Number	Negative/Positive	Sensitivity	Specificity	G-mean	AUC
RF2	Balanced dataset	4 900	2 352/2 548	0.891	0.774	0.830	0.833
RF	Original training set	2 814	2 618/196	0.999	0.119	0.378	0.559

Model	Sensitivity	Specificity	G-mean	AUC	P	95%CI
Logistic regression	0.740	0.643	0.690	0.738	<0.05	(0.679,0.792)
Support vector machine	0.746	0.583	0.659	0.665	<0.05	(0.610,0.719)
Decision tree	0.791	0.583	0.679	0.687	<0.05	(0.633,0.742)
Artificial neural network	0.625	0.762	0.690	0.720	<0.05	(0.677,0.789)
RF2	0.891	0.774	0.830	0.833	<0.05	(0.784,0.876)

[1]	Ying ZHAO,Danyu ZHAO,Chao LIU. Bioinformatics analysis on prognostic evaluation value of TXNDC11 gene in pan-cancer and its immunity regulation [J]. Journal of Jilin University(Medicine Edition), 2022, 48(1): 142-153.
[2]	Lili QIN,Xiaobo MA,Tianye ZHAO,Xuerong TAO,Min ZHENG,Xueying WANG,Jiaxin YI,Yanhua WU,Jing JIANG. Effects of MMP-9 and TIMP-1 expressions on prognostic evaluation of gastric cancer patients after radical gastrectomy [J]. Journal of Jilin University(Medicine Edition), 2022, 48(1): 163-171.
[3]	Shuzhen LI,Yajie CAO,Haiying GENG,Zengxiaorui CAI,Chunmei DAI,Youfeng WEN,Ning LI. Bioinformatics analysis based on expression level of HMMR in lung adenocarcinoma tissue and its impact on prognosis of LUAD patients [J]. Journal of Jilin University(Medicine Edition), 2021, 47(6): 1502-1509.
[4]	Yang YU,Sainan LIU,Yunkai LIU,Yong LI,Yichun QIAO,Yi CHENG. Bioinformatic analysis on expression characteristics of PRPS2 and its relationship with prognosis of breast cancer [J]. Journal of Jilin University(Medicine Edition), 2021, 47(5): 1229-1236.
[5]	Aiping WEN,Le LUO,Zhiwei MENG,Jingji JIN,Honggui ZHOU,Jihong ZHU. Expression of Tip60 protein in endometrial adenocarcinoma tissue and its clinical significance [J]. Journal of Jilin University(Medicine Edition), 2021, 47(5): 1244-1249.
[6]	Lu LU,Dongming LI,Xueguo WANG,Dan SONG,Taicheng WANG,Hongyan ZHAO,Xiaoyong WU. Effect of chloroquine on gemcitabine-resistant cells by affecting autophagy and mitochondrial function of pancreatic cancer cells and its mechanism [J]. Journal of Jilin University(Medicine Edition), 2021, 47(4): 926-933.
[7]	Dan SU, Yi LIU, Manman CUI, Nian YANG, Yu HUANG, Wenjing HE. Evaluation of clinical effect of screening chemotherapy regimens in treatment of ovarian cancer based on miniPDX animal models [J]. Journal of Jilin University(Medicine Edition), 2021, 47(3): 731-739.
[8]	Dongkui ZHOU, Mingqian LU, Yaxin LIN, Xuesong FENG, Yan GAO, Hao SONG. Small cell carcinoma of gallbladder with liver and retroperitoneal lymph node metastasis: A case report and literature review [J]. Journal of Jilin University(Medicine Edition), 2021, 47(3): 747-752.
[9]	GUO Fei, ZHU Lin, XU Hong, XIE Zongyu, ZHANG Li, DENG Xuefei. Analysis on correlation between image features of chest MSCT and prognosis in patients with novel coronavirus pneumonia [J]. Journal of Jilin University(Medicine Edition), 2020, 46(04): 867-874.
[10]	SONG Hongyun, PAN Ying, AN Furun, ZHANG Jiakui, YANG Dongdong, ZHAI Zhimin. Clinical characteristics of acute leukemia patients with complex karyotypes and their effects on prognosis [J]. Journal of Jilin University(Medicine Edition), 2020, 46(02): 377-382.
[11]	LIU Xiuhua, WANG Man, DONG Xinjie, LIU Jingnan, HAN Wei, GUAN Yinghui. Expression of filamin A in cancer tissue of cervical cancer patients and its clinical significance [J]. Journal of Jilin University(Medicine Edition), 2019, 45(06): 1238-1242.
[12]	LYU Xiuyun, YANG Ting, XU Lei, WANG Lihong. Dynamic changes of MIP-1α and IL-13 levels of patients with severe asthma and their valuesin prognosis evaluation [J]. Journal of Jilin University(Medicine Edition), 2019, 45(06): 1401-1407.
[13]	FENG Xiunan, YUAN Yi, SU Kaisheng, JIANG Zhenyu. Distribution of peripheral blood T lymphocyte subsets of patients with systemic lupus erythematosus and its relationships with disease activity and prognosis [J]. Journal of Jilin University(Medicine Edition), 2019, 45(06): 1415-1421.
[14]	YU Xiuyan, WAN Guangcai, SUN Hongshuai, ZHU Hua, GAO Haiyan, WU Xuefeng. Relationship between fibrinogen and prognosis of breast cancer patients:A Meta-analysis [J]. Journal of Jilin University(Medicine Edition), 2019, 45(05): 1092-1097.
[15]	HUO Xiaolei, PEI Zhen, YANG Hao, ZHANG Yiqiang, JIA Jiantao, HAN Lingna. Effect of LncRNA-BLACAT1 on cell proliferation of non-small cell lung cancer through regulation of CyclinD1/CDKN2B axis and its mechanism [J]. Journal of Jilin University(Medicine Edition), 2019, 45(04): 759-765.