Construction and validation of a machine learning-based prediction model for very early recurrence after curative-intent resection for gallbladder cancer

Abstract

Abstract:

ObjectiveTo explore the risk factors for very early recurrence (VER) after curative-intent resection for gallbladder cancer (GBC) patients and construct prediction models for VER based on various machine learning (ML) algorithms.MethodsA retrospective study was conducted on 329 GBC patients who underwent curative-intent surgery at our hospital between January 2016 and December 2020. Risk factors for VER were identified, and prediction models were constructed, validated and compared with multiple ML algorithms[logistic regression (LR), support vector machine (SVM), naive Bayes (NB), random forest (RF), light gradient boosting machine (LGB), and extreme gradient boosting (XGB)]based on independent associated factors for VER.ResultsAmong the 329 patients who underwent curative-intent resection in patients with GBC, 162 (49.2%) patients experienced recurrence, including 69 (42.6%) with VER（<6 months） and 93 (57.4%) with non-VER（≥6 months）. Survival analysis showed that patients with VER had significantly worse median overall survival compared to those with non-VER (6 monthsvs. not arrived，χ²=398.2,P<0.001). Univariate analysis showed that carcinoembryonic antigen (CEA), carbohydrate antigen (CA)19-9, CA-125, tumor differentiation, pathological type, liver involvement, vascular invasion, perineural invasion, TNM stage, T stage and N stage were risk factors of VER (P<0.05), whereas adjuvant chemotherapy was protective factor (P<0.05). Multivariate analysis confirmed CA-125, tumor differentiation, pathological type, vascular invasion and N stage as independent risk factors (P<0.05), whereas adjuvant chemotherapy was independent protective factor (P<0.05). XGB model achieved the best performance with an area under curve (AUC) of 0.841 and an accuracy (ACC) of 83.0% in the validation set. Shapley additive explanations (SHAP) bar plots highlighted tumor differentiation, N stage, pathological type of tumor, and CA-125 the top four features contributing to the model, each positively influencing the predicted probability of VER.ConclusionsCA-125, tumor differentiation, pathological type, vascular invasion, N stage and adjuvant chemotherapy are independent factors associated with VER of GBC following curative-intent resection. ML-based prediction models incorporating these factors have the potential to some extent to effectively identify high-risk patients, providing a valuable reference for VER surveillance in GBC.

Key words:Gallbladder cancer(GBC),Very early recurrence(VER),Prognosis,Machine learning, Prediction model

CLC Number:

R735.8

TANG Zhenqi, LI Qi, LIU Hengchao, ZHANG Dong, GENG Zhimin. Construction and validation of a machine learning-based prediction model for very early recurrence after curative-intent resection for gallbladder cancer[J]. Journal of Surgery Concepts & Practice, 2025, 30(04): 316-324.

Figures/Tables6

Tab 1

Baseline characteristics of GBC patients in the training and validation sets[n(%)]

Item	Training sets (n=229)	Validation sets (n=100)	χ²value	Pvalue
Gender
Male	71 （31.0）	30 （30.0）	0.033	0.856
Female	158 （69.0）	70 （70.0）	0.033	0.856
Age (years)
≤65	99 （43.2）	43 （43.0）	0.002	0.969
>65	130 （56.8）	57 （57.0）	0.002	0.969
Gallstones
No	85 （37.1）	43 （43.0）	1.013	0.314
Yes	144 （62.9）	57 （57.0）	1.013	0.314
TBIL (mmol/L)
≤34.2	203 （88.6）	93 （93.0）	1.462	0.227
>34.2	26 （11.4）	7 （7.0）	1.462	0.227
CEA (μg/L)
≤5.0	185 （80.8）	77 （77.0）	0.615	0.433
>5.0	44 （19.2）	23 （23.0）	0.615	0.433
CA19-9(U/mL)
≤39.0	162 （70.7）	72 （72.0）	0.054	0.817
>39.0	67 （29.3）	28 （28.0）	0.054	0.817
CA-125(U/mL)
≤35.0	160 （69.9）	74 （74.0）	0.065	0.798
>35.0	69 （30.1）	26 （26.0）	0.065	0.798
Unexpected gallbladder carcinoma
No	85 （37.1）	43 （43.0）	0.578	0.447
Yes	144 （62.9）	57 （57.0）	0.578	0.447
Tumor differentiation
Well	31 （13.5）	15 （15.0）	0.128	0.938
Moderate	125 （54.6）	54 （54.0）	0.128	0.938
Poor	73 （31.9）	31 （31.0）
Pathological type
Adenocarcinoma	203 （88.6）	87 （87.0）	0.181	0.671
Non-adenocarcinoma	26 （11.4）	13 （13.0）	0.181	0.671
Liver involvement
No	164 （71.6）	74 （74.0）	0.198	0.657
Yes	65 （28.4）	26 （26.0）	0.198	0.657
Vascular invasion
No	217 （94.8）	93 （93.0）	0.396	0.529
Yes	12 （5.2）	7 （7.0）	0.396	0.529
Perineural invasion
No	206 （90.0）	86 （86.0）	1.092	0.296
Yes	23 （10.0）	14 （14.0）	1.092	0.296
AJCC 8th edition TNM stage
Ⅰ-Ⅱ	47 （20.5）	16 （16.0）
Ⅲ	148 （64.6）	64 （64.0）	1.874	0.392
Ⅳ	34 （14.8）	20 （20.0）
AJCC 8th edition T stage
T1-2	51 （22.3）	19 （19.0）
T3	161 （70.3）	71 （71.0）	0.918	0.632
T4	17 （7.4）	10 （10.0）
AJCC 8th edition N stage
N0	163 （71.2）	67 （67.0）
N1	45 （19.7）	23 （23.0）	0.603	0.740
N2	21 （9.2）	10 （10.0）
Surgical margin
R0	219 （95.6）	93 （93.0）	0.985	0.321
R1	10 （4.4）	7 （7.0）	0.985	0.321
Adjuvant chemotherapy
No	178 （77.7）	83 （83.0）	1.179	0.277
Yes	51 （22.3）	17 （17.0）	1.179	0.277
VER
No	182 （79.5）	78 （78.0）	0.091	0.762
Yes	47 （20.5）	22 （22.0）	0.091	0.762

Tab 1

Fig 1

Tab 2

Univariate and multivariate analysis of postoperative VER in patients with GBC after curative-intent resection

Item	Univariate analysis		Multivariate analysis
Item	OR (95%CI)	Pvalue	OR (95%CI)	Pvalue
Gender
Femalevs.Male	0.391(0.379-1.462)	0.391
Age (years)
>65vs.≤65	1.829(0.927-3.606)	0.081
Gallstones
Yesvs. No	1.508(0.754-3.013)	0.245
TBIL (mmol/L)
>34.2vs.≤34.2	2.299(0.952-5.551)	0.064
CEA(μg/L)
>5.0vs.≤5.0	3.724(1.813-7.650)	<0.001	1.459(0.503-4.232)	0.487
CA19-9(U/mL)
>39.0vs.≤39.0	3.006(1.546-5.845)	0.001	0.791(0.290-2.154)	0.646
CA-125(U/mL)
>35.0vs.≤35.0	3.227(1.645-6.328)	0.001	3.283(1.208-8.918)	0.020
Unexpected gallbladder carcinoma
Yesvs. No	0.563(0.262-1.210)	0.141
Tumor differentiation
Poorvs.Well	9.022(1.996-40.782)	0.004	3.154(1.435-6.931)	0.004
Moderatevs.Well	2.282(0.498-10.451)	0.288	3.154(1.435-6.931)	0.004
Pathological type
Adenocarcinomavs. Non-adenocarcinoma	4.114(1.754-9.652)	0.001	9.806(2.758-34.865)	<0.001
Liver involvement
Yesvs. No	5.116(2.592-10.096)	<0.001	1.197(0.453-3.161)	0.717
Vascular invasion
Yesvs. No	9.128(2.617-31.837)	0.001	20.895(2.925-149.260)	0.002
Perineural invasion
Yesvs. No	4.329(1.771-10.580)	0.001	3.187(0.872-11.651)	0.080
AJCC 8th edition TNM stage
Ⅳvs.Ⅰ-Ⅱ	16.392(4.292-62.610)	<0.001	0.661(0.178-2.453)	0.536
Ⅲvs.Ⅰ-Ⅱ	3.030(0.871-15.537)	0.081	0.661(0.178-2.453)	0.536
AJCC 8th edition T stage
T4vs.T1-2	22.000 (5.293-91.448)	<0.001	1.227(0.384-3.914)	0.730
T3vs.T1-2	3.000(1.008-8.932)	0.048	1.227(0.384-3.914)	0.730
AJCC 8th edition N stage
N2vs.N0	7.273(1.380-6.569)	<0.001	3.701(1.459-9.383)	0.006
N1vs.N0	3.011(1.380-18.580)	0.006	3.701(1.459-9.383)	0.006
Surgical margin
R1vs.R0	2.729(0.737-10.096)	0.133
Adjuvant chemotherapy
Yesvs. No	0.267(0.091-0.784)	0.016	0.162(0.037-0.709)	0.016

Tab 2

Tab 3

Prediction performance of different algorithms in the training and validation sets

	Algorithm	AUC（95% CI）	ACC	F1-score	Sensitivity	Specificity
Training sets	LGB	0.844（0.780-0.903）	0.728	0.557	0.830	0.702
	LR	0.866（0.801-0.920）	0.855	0.602	0.532	0.939
	NB	0.856（0.787-0.915）	0.833	0.604	0.617	0.890
	RF	0.898（0.847-0.943）	0.825	0.649	0.787	0.834
	SVM	0.877（0.821-0.926）	0.873	0.688	0.681	0.923
	XGB	0.890（0.832-0.941）	0.873	0.695	0.702	0.917
Validation sets	LGB	0.828（0.742-0.909）	0.750	0.603	0.864	0.718
	LR	0.833（0.733-0.918）	0.850	0.634	0.591	0.923
	NB	0.814（0.694-0.912）	0.820	0.591	0.591	0.885
	RF	0.831（0.736-0.913）	0.800	0.615	0.727	0.821
	SVM	0.826（0.699-0.919）	0.830	0.622	0.636	0.885
	XGB	0.841（0.746-0.922）	0.830	0.622	0.636	0.885

Tab 3

Fig 2

Fig 3

References32

[1]	VALLE J W, KELLEY R K, NERVI B, et al. Biliary tract cancer[J].Lancet,2021,397(10272):428-444. doi:10.1016/S0140-6736(21)00153-7pmid:33516341
[2]	中华医学会外科学分会胆道外科学组, 中国医师协会外科医师分会胆道外科专业委员会. 胆囊癌诊断和治疗指南(2019版)[J].中华外科杂志,2020,58(4):243-251.
	Branch of Biliary Surgery, Chinese Surgical Society, Chinese Committee of Biliary Surgeons. Guideline for the diagnosis and treatment of gallbladder carcinoma (2019 edition)[J].Chin J Surg,2020,58(4):243-251.
[3]	NISHIO H, EBATA T, YOKOYAMA Y, et al. Gallbladder cancer involving the extrahepatic bile duct is worthy of resection[J].Ann Surg,2011,253(5):953-960. doi:10.1097/SLA.0b013e318216f5f3pmid:21490453
[4]	KISHI Y, SHIMADA K, HATA S, et al. Definition of T3/4 and regional lymph nodes in gallbladder cancer: which is more valid, the UICC or the Japanese staging system?[J].Ann Surg Oncol,2012,19(11):3567-3573. doi:10.1245/s10434-012-2599-5pmid:22890597
[5]	SAHARA K, TSILIMIGRAS D I, KIKUCHI Y, et al. Defining and predicting early recurrence after resection for gallbladder cancer[J].Ann Surg Oncol,2021,28(1):417-425.
[6]	SHIMIZU Y, ASHIDA R, SUGIURA T, et al. Early recurrence in resected gallbladder carcinoma: clinical impact and its preoperative predictive score[J].Ann Surg Oncol,2022,29(9):5447-5457. doi:10.1245/s10434-022-11937-ypmid:35666409
[7]	YANG S Q, FENG H, TIAN Y, et al. Unraveling early recurrence of risk factors in gallbladder cancer: a syste-matic review and meta-analysis[J].Eur J Surg Oncol,2024,50(7):108372.
[8]	SHIMIZU Y, OHTSUKA M, ITO H, et al. Should the extrahepatic bile duct be resected for locally advanced gallbladder cancer?[J].Surgery,2004,136(5):1012-1017.
[9]	TSILIMIGRAS D I, HYER J M, PAREDES A Z, et al. The optimal number of lymph nodes to evaluate among patients undergoing surgery for gallbladder cancer: correlating the number of nodes removed with survival in 6531 patients[J].J Surg Oncol,2019,119(8):1099-1107. doi:10.1002/jso.25450pmid:30864246
[10]	LI Q, LI N, GAO Q, et al. The clinical impact of early recurrence and its recurrence patterns in patients with gallbladder carcinoma after radical resection[J].Eur J Surg Oncol,2023,49(10):106959.
[11]	张东, 李起, 郭伟, 等. 胆囊癌意向性根治术后肿瘤早期复发影响因素及辅助化疗效果分析的全国多中心临床研究[J].中华消化外科杂志,2024,23(1):125-133.
	ZHANG D, LI Q, GUO W, et al. Analysis of influencing factors for early tumor recurrence and efficacy of adjuvant chemotherapy in gallbladder carcinoma patients after curative-intent resection: a nationwide, multicenter clinical study[J].Chin J Dig Surg,2024,23(1):125-133.
[12]	LIM H, SEO D W, PARK D H, et al. Prognostic factors in patients with gallbladder cancer after surgical resection: analysis of 279 operated patients[J].J Clin Gastroenterol,2013,47(5):443-448. doi:10.1097/MCG.0b013e3182703409pmid:23188077
[13]	KIM Y, AMINI N, WILSON A, et al. Impact of chemotherapy and external-beam radiation therapy on outcomes among patients with resected gallbladder cancer: a multi-institutional analysis[J].Ann Surg Oncol,2016,23(9):2998-3008. doi:10.1245/s10434-016-5262-8pmid:27169772
[14]	雷建军, 张健, 陈晨, 等. 胆囊癌临床和病理学因素与神经浸润的相关性研究[J].中华外科杂志,2022,60(7):695-702.
	LEI J J, ZHANG J, CHEN C, et al. Analysis of perineural invasion with clinicopathological factors and prognosis for curatively resected gallbladder carcinoma[J].Chin J Surg,2022,60(7):695-702.
[15]	SHEN F Z, ZHANG B Y, FENG Y J, et al. Current research in perineural invasion of cholangiocarcinoma[J].J Exp Clin Cancer Res,2010,29(1):24.
[16]	LI J, KANG R, TANG D. Cellular and molecular mechanisms of perineural invasion of pancreatic ductal adenocarcinoma[J].Cancer Commun (Lond),2021,41(8):642-660.
[17]	GOEL S, AGGARWAL A, IQBAL A, et al. Multimodality management of gallbladder cancer can lead to a better outcome: experience from a tertiary care oncology centre in North India[J].World J Gastroenterol,2021,27(45):7813-7830.
[18]	WANG Z, LIU H, HUANG Y, et al. Comparative analysis of postoperative curative effect of liver wedge resection and liver Ⅳb + Ⅴ segment resection in patients with T2b gallbladder cancer[J].Front Surg,2023,10:1139947.
[19]	PARK J S, YOON D S, KIM K S, et al. Actual recurrence patterns and risk factors influencing recurrence after curative resection with stage Ⅱ gallbladder carcinoma[J].J Gastrointest Surg,2007,11(5):631-637.
[20]	LV T R, WANG J K, LI F Y, et al. Prognostic factors for resected cases with gallbladder carcinoma: a systematic review and meta-analysis[J].Int J Surg,2024,110(7):4342-4355.
[21]	胡成宇, 杨建宇, 徐燕楠, 等. 可切除胰腺癌根治术后早期复发的危险因素分析及列线图模型构建[J].中华胰腺病杂志,2025,25(2):104-111.
	HU C Y, YANG J Y, XU Y N, et al. Analysis of risk factors and development of a nomogram model for early recurrence following curative resection of resectable pancreatic cancer[J].Chin J Pancreat Dis,2025,25(2):104-111.
[22]	沈雷斌, 杨沔, 金梁斌, 等. 结直肠癌肝转移患者术后早期复发的高危因素分析[J].中华普通外科杂志,2025,40(3):207-212.
	SHEN L B, YANG M, JIN L B, et al. High-risk factors for early postoperative recurrence in patients with colorectal cancer liver metastasis[J].Chin J Gen Surg,2025,40(3):207-212.
[23]	秦克旺, 揭志刚, 刘寄尘, 等. 检测血清 CA125 水平对胃癌术后腹膜复发的诊断价值[J].中华胃肠外科杂志,2004,7(3):201.
	QIN K W, JIE Z G, LIU J C, et al. Diagnostic value of serum CA125 level for peritoneal recurrence after gastric cancer surgery[J].Chin J Gastrointest Surg,2004,7(3):201.
[24]	AYABE R I, WACH M M, RUFF S M, et al. Gallbladder squamous cell carcinoma: an analysis of 1084 cases from the National Cancer Database[J].J Surg Oncol,2020,122(4):716-722.
[25]	LEIGH N, SOLOMON D, PLETCHER E, et al. Adeno-squamous and squamous cell carcinoma of the gallbladder: the importance of histology in surgical management[J].Am J Surg,2020,220(5):1242-1248. doi:10.1016/j.amjsurg.2020.06.050pmid:32646581
[26]	SAMUEL S, MUKHERJEE S, AMMANNAGARI N, et al. Clinicopathological characteristics and outcomes of rare histologic subtypes of gallbladder cancer over two decades: a population-based study[J].PLoS One,2018,13(6):e0198809.
[27]	SONG H W, CHEN C, SHEN H X, et al. Squamous/adenosquamous carcinoma of the gallbladder: analysis of 34 cases and comparison of clinicopathologic features and surgical outcomes with adenocarcinoma[J].J Surg Oncol,2015,112(6):677-680.
[28]	ZOU R Q, HU H J, LV T R, et al. Clinicopathological characteristics and outcome of primary sarcomatoid carcinoma of the gallbladder[J].Front Oncol,2022,12:1009673.
[29]	ZOU R Q, HU H J, LIU F, et al. Comparison of clinicopathological characteristics of mucinous adenocarcinoma and conventional adenocarcinoma of gallbladder[J].Asian J Surg,2023,46(1):283-290.
[30]	LI X F, MA T T, LI T. Risk factors and survival prediction model establishment for prognosis in patients with radical resection of gallbladder cancer[J].World J Gastrointest Surg,2024,16(10):3239-3252.
[31]	WU D, JIN W, ZHANG Y, et al. Insights from the analysis of clinicopathological and prognostic factors in patients with gallbladder cancer[J].Front Oncol,2022,12:889334.
[32]	THENG D, BHOYAR K K. Feature selection techniques for machine learning: a survey of more than two decades of research[J].Knowl Inf Syst,2024,66:1575-1637.