儿童白血病脐带血移植原发性植入失败预测模型的构建

doi:10.12372/jcp.2025.24e0304

摘要/Abstract

摘要：

目的基于过采样算法构建儿童白血病脐带血移植后发生原发性植入失败（PGF）的预测模型，为优化脐带血移植方案提供依据。方法回顾性分析2017年1月至2022年12月接受脐带血移植白血病患儿的病历资料，根据是否发生PGF分为植入失败组与植入成功组。基于随机过采样少数类（ROSE）与合成少数类过采样（SOMTE）算法扩充阳性组数据。使用随机分层抽样法将纳入病例以7∶3的比例分为训练集及测试集。采用随机森林、神经网络、logistic回归3种方法进行模型的构建，并使用5折交叉验证法评估算法的稳定性。计算ROC曲线下面积精确率、召回率、F1分数评价模型的性能。选择多因素logistic回归进行危险因素分析。结果共纳入92例白血病患儿，10例患儿发生PGF（10.9%）。5折交叉验证法显示ROSE与SOMTE算法在训练集与测试集中都有较高的准确度。在使用ROSE平衡化处理后的数据集中，各个模型均具有良好的预测效果，表现最佳的是神经网络模型，幼年粒单核细胞白血病、HLA匹配度<9/10、减低强度预处理方案、未发生围植入综合征以及42天内感染EBV是PGF发生的危险因素。结论多种因素可引起白血病患儿脐带血移植后PGF的发生。基于ROSE-神经网络模型具有良好的预测能力，有助于医师早期识别PGF发生的高风险患者，提供个性化治疗，改善患儿预后。

关键词: 白血病, 脐带血移植, 原发性植入失败, 预测模型, 儿童

Abstract:

Objective To build a risk prediction model for primary graft failure (PGF) after umbilical cord blood transplantation (UCBT) in pediatric leukemia based on over-sampling. Methods Patients with leukemia who received umbilical cord blood transplantation from January 2017 to December 2022 were retrospectively analyzed. According to the presence or absence of PGF, the patients were divided into graft failure group and graft success group. Based on the over-sampling algorithm to expand the positive group data, the random forest, neural network and logistic regression were used to construct the mode. The stability of the algorithm was evaluated by using the 5-fold cross-validation method. The model was evaluated by using AUC, precision, recall and F1-score. Results A total of 92 leukemia patients were enrolled, PGF occurred in 10 patients (10.9%). ROSE and SMOTE algorithm demonstrate good stability in 5-fold cross-validation method. In the data set processed by ROSE algorithm, all models have good prediction effect, and the best performance is the neural network model. Juvenile myelomonocytic leukemia, HLA matching<9/10, RIC, no Periengraftment syndrome and EBV infection within 42 days were independent risk factors for PGF. Conclusion Multiple factors may cause PGF after umbilical cord blood transplantation in pediatric leukemia. ROSE-Neural Network model has good predictive ability, which can help doctors to identify patients at high risk of PGF early, provide personalized treatment, and improve the prognosis of children.

Key words: leukemia, umbilical cord blood transplantation, primary graft failure, prediction model, child

中图分类号:

张志奇, 熊若兰, 李泊涵, 吉奇, 王庆伟, 卢俊, 李捷, 肖佩芳, 胡绍燕. 儿童白血病脐带血移植原发性植入失败预测模型的构建[J]. 临床儿科杂志, 2025, 43(7): 511-518.

ZHANG Zhiqi, XIONG Ruolan, LI Bohan, JI Qi, WANG Qingwei, LU Jun, LI Jie, XIAO Peifang, HU Shaoyan. Construction of risk prediction model for primary graft failure after umbilical cord blood transplantation in pediatric leukemia[J]. Journal of Clinical Pediatrics, 2025, 43(7): 511-518.

图/表 6

表1

表2

表3

表4

表5

图1

参考文献 23

[1]	Yabe H. Allogeneic hematopoietic stem cell transplantation for inherited metabolic disorders[J]. Int J Hematol, 2022, 116(1): 28-40. doi: 10.1007/s12185-022-03383-z pmid: 35594014
[2]	Marcais A, Mahloui N, Neven B, et al. Curative allogeneic hematopoietic stem cell transplantation following reduced toxicity conditioning in adults with primary immunodeficiency[J]. Bone Marrow Transplant, 2022, 57(10): 1520-1530.
[3]	刘虎, 肖佩芳, 卢俊, 等. 脐带血移植治疗儿童血液病疗效的回顾性分析[J]. 临床儿科杂志, 2017, 35(10): 769-774.
	Liu H, Xiao PF, Lu J, et al. Retrospective analysis of curative effect of umbilical cord blood transplantation on children with hematologic diseases[J]. Linchuang Erke Zazhi, 2017, 35(10): 769-774.
[4]	Sanchez-Petitto G, Rezvani K, Daher M, et al. Umbilical cord blood transplantation: connecting its origin to its future[J]. Stem Cells Transl Med, 2023, 12(2): 55-71.
[5]	Sun Z, Yao B, Xie H, et al. Clinical progress and preclinical insights into umbilical cord blood transplantation improvement[J]. Stem Cells Transl Med, 2022, 11(9): 912-926.
[6]	Chen J, Pang A, Zhao Y, et al. Primary graft failure following allogeneic hematopoietic stem cell transplantation: risk factors, treatment and outcomes[J]. Hematology, 2022, 27(1): 293-299. doi: 10.1080/16078454.2022.2042064 pmid: 35192779
[7]	尹光丽, 李建勇, 缪扣荣. 造血干细胞移植植入失败的研究进展[J]. 南京医科大学学报(自然科学版), 2020, 40(1): 141-146.
	Yin GL, Li JY, Miao KR. Progress on graft failure of allogeneic hematopoietic stem cell transplantation[J]. Nanjing Yike Daxue Xuebao (Ziran Kexue Ban), 2020, 40(1): 141-146.
[8]	Park JH, Lee JH, Lee JH, et al. Incidence, management, and prognosis of graft failure and autologous reconstitution after allogeneic hematopoietic stem cell transplantation[J]. J Korean Med Sci, 2021, 36(23): e151. doi: 10.3346/jkms.2021.36.e151 pmid: 34128593
[9]	Balduzzi A, Dalle JH, Wachowiak J, et al. Transplantation in children and adolescents with acute lymphoblastic leukemia from a matched donor versus an HLA-identical sibling: Is the outcome comparable? Results from the international BFM ALL SCT 2007 study[J]. Biol Blood Marrow Transplant, 2019, 25(11): 2197-2210.
[10]	Olsson RF, Logan BR, Chaudhury S, et al. Primary graft failure after myeloablative allogeneic hematopoietic cell transplantation for hematologic malignancies[J]. Leukemia, 2015, 29(8): 1754-1762. doi: 10.1038/leu.2015.75 pmid: 25772027
[11]	顾永春, 顾兴全, 武娇, 等. 面向不平衡网络评论数据挖掘的服务质量评价[J]. 小型微型计算机系统, 2021, 42(2): 354-361.
	Gu YC, Gu XQ, Wu J, et al. Service quality evaluation via mining unbalanced network reviews[J]. Xiaoxing Weixing Jisuanji Xitong, 2021, 42(2): 354-361.
[12]	Ozdemir ZN, Civriz Bozdag S. Graft failure after allogeneic hematopoietic stem cell transplantation[J]. Transfus Apher Sci, 2018, 57(2): 163-167. doi: S1473-0502(18)30132-0 pmid: 29724627
[13]	McCann S. Graft failure[J]. Bone Marrow Transplant, 2020, 55(10): 1888-1889.
[14]	费洋, 胡晓霞, 陈琪, 等. 异基因造血干细胞移植植入失败的危险因素分析[J]. 中华血液学杂志, 2020, 41(1): 64-68.
	Fei Y, Hu XX, Chen Q, et al. Risk-factors analysis of graft failure after allogeneic hematopoietic stem cell transplantation[J]. Zhonghua Xueyexue Zazhi, 2020, 41(1): 64-68.
[15]	Arber DA, Orazi A, Hasserjian R, et al. The 2016 revision to the World Health Organization classification of myeloid neoplasms and acute leukemia[J]. Blood, 2016, 127(20): 2391-2405. doi: 10.1182/blood-2016-03-643544 pmid: 27069254
[16]	Mata JR, Zahurak M, Rosen N, et al. Graft Failure Incidence, risk Factors, and outcomes in patients undergoing non-myeloablative allogeneic hematopoietic cell transplantation using post-transplant cyclophosphamide[J]. Transplant Cell Ther, 2024, 30(6): 588-596. doi: 10.1016/j.jtct.2024.03.018 pmid: 38521411
[17]	Hong KT, Kang HJ, Kim NH, et al. Peri-engraftment syndrome in allogeneic hematopoietic SCT[J]. Bone Marrow Transplant, 2013, 48(4): 523-528.
[18]	Park M, Lee SH, Lee YH, et al. Pre-engraftment syndrome after unrelated cord blood transplantation: a predictor of engraftment and acute graft-versus-host disease[J]. Biol Blood Marrow Transplant, 2013, 19(4): 640-646.
[19]	郑昌成. 非血缘脐血移植植入前综合征的临床与基础研究[D]. 山东大学, 2014.
	Zheng CC. Pre-engraftment syndrome after unrelated cord blood transplantation: clinical and basic research[D]. Shandong University, 2014.
[20]	Sawada A, Inoue M, Koyama-Sato M, et al. Umbilical cord blood as an alternative source of reduced-intensity hematopoietic stem cell transplantation for chronic Epstein-Barr virus-associated T or natural killer cell lymphoproliferative diseases[J]. Biol Blood Marrow Transplant, 2014, 20(2): 214-221.
[21]	熊思伟, 刘玉琳. 基于Borderline-SMOTE算法与Stacking集成学习的前列腺肿瘤风险预测研究[J]. 现代肿瘤医学, 2023, 31(16): 3075-3081.
	Xiong SW, Liu YL. Risk prediction study of prostate tumors based on Borderline-SMOTE algorithm and Stacking ensemble learning[J]. Xiandai Zhongliu Yixue, 2023, 31(16): 3075-3081.
[22]	郑志方, 敬小青, 刘利蕊, 等. 儿童原发性肾病综合征复发相关因素分析及风险预测模型构建[J]. 临床儿科杂志, 2023, 41(12): 914-918.
	Zheng ZF, Jing XQ, Liu LR, et al. Related factors analysis of recurrence and risk prediction model construction for primary nephrotic syndrome in children[J]. Linchuang Erke Zazhi, 2023, 41(12): 914-918.
[23]	中华医学会儿科学分会血液学组. 儿童恶性血液病脐带血移植专家共识[J]. 中华儿科杂志, 2016, 54(11): 804-807.
	Chinese Medical Association. Expert consensus on umbilical cord blood transplantation for pediatric malignant hematological diseases[J]. Zhonghua Erke Zazhi, 2016, 54(11): 804-807.

项目	植入失败组（n=10）	植入成功组（n=82）	统计量	P
性别[n(%)]			χ²=2.34	0.126
男	8（80）	40（48.8）
女	2（20）	42（51.2）
年龄[M(P₂₅~P₇₅)/岁]	2.47（1.98~5.52）	3.46（1.73~5.69）	Z=0.95	0.34
疾病类型[n(%)]			χ²=9.10	0.011
JMML	5（50）	8（9.8）
ALL	3（30）	33（40.2）
AML	2（20）	41（50）
HLA[n(%)]			χ²=4.04	0.044
≥9/10	2（20）	44（53.7）
<9/10	8（80）	38（46.3）
供受者血型[n(%)]			χ²=0.35	0.556
相合	4（40）	21（25.6）
不相合	6（60）	61（74.4）
预处理方案强度[n(%)]			χ²=5.83	0.016
MAC	3（30）	60（73.2）
RIC	7（70）	22（26.8）
白消安剂量[M(P₂₅~P₇₅)]/mg·kg^-1	9.74（9.6~12.86）	12.8（9.89~12.8）	Z=1.55	0.121
环磷酰胺剂量[M(P₂₅~P₇₅)]/mg·kg^-1	120（103.94~120.51）	120（115.08~120）	Z=0.42	0.674
MNC输注量[M(P₂₅~P₇₅)]×10⁵·kg^-1	9.31（6.42~11.09）	7.04（5.62~9.23）	Z=1.56	0.120
CD34⁺细胞输注量[M(P₂₅~P₇₅)]×10⁷·kg^-1	3.48（1.74~4.64）	2.54（1.88~3.37）	Z=1.20	0.231
Peri-ES[n(%)]			χ²=9.18	0.002
有	2（20）	60（73.2）
无	8（80）	22（26.8）
42天内EBV感染[n(%)]			χ²=14.93	<0.001
有	4（40）	2（2.4）
无	6（60）	80（97.6）

5折交叉验证	随机森林		神经网络		logistic回归
5折交叉验证	ROSE	SMOTE	ROSE	SMOTE	ROSE	SMOTE
Fold1	1.00	0.91	0.92	0.91	0.87	0.74
Fold2	0.96	0.95	0.91	0.87	0.92	0.96
Fold3	0.96	0.87	1.00	0.95	0.72	0.83
Fold4	0.96	1.00	0.96	0.95	0.57	0.91
Fold5	0.96	0.91	1.00	0.96	0.87	0.91

模型	集合	方法	项	AUC	precision	recall	F1-score
随机森林	训练集	ROSE	值	0.97	1.00	0.97	0.99
			95%CI	0.94~1	0.93~1	0.92~0.99	0.95~0.99
			P	-	<0.001	<0.001	<0.001
	测试集		值	0.98	1.00	0.98	0.99
			95%CI	0.94~1	0.84~1	0.88~1	0.96~0.99
			P	-	<0.001	<0.001	<0.001
	训练集	SMOTE	值	0.96	0.93	0.96	0.94
			95%CI	0.92~0.99	0.82~0.98	0.89~0.98	0.89~0.97
			P	-	<0.001	<0.001	<0.001
	测试集		值	0.94	0.88	0.94	0.91
			95%CI	0.88~1	0.68~0.97	0.82~0.98	0.85~0.94
			P	-	<0.001	<0.001	<0.001
神经网络	训练集	ROSE	值	0.99	1.00	0.97	0.99
			95%CI	0.99~1	0.93~1	0.92~0.99	0.96~0.99
			P	-	<0.001	<0.001	<0.001
	测试集		值	1.00	1.00	0.98	0.99
			95%CI	0.99~1	0.84~1	0.88~1	0.96~1
			P	-	<0.001	<0.001	<0.001
	训练集	SMOTE	值	0.96	0.93	0.94	0.93
			95%CI	0.93~0.99	0.82~0.97	0.87~0.97	0.88~0.96
			P	-	<0.001	<0.001	<0.001
	测试集		值	0.98	0.88	0.94	0.91
			95%CI	0.95~1	0.68~0.97	0.82~0.98	0.85~0.94
			P	-	<0.001	<0.001	<0.001
logistic回归	训练集	ROSE	值	0.90	0.85	0.81	0.83
			95%CI	0.84-0.95	0.73-0.92	0.73-0.87	0.76-0.88
			P	-	<0.001	<0.001	<0.001
	测试集		值	0.92	0.89	0.83	0.86
			95%CI	0.85~0.99	0.70~0.97	0.69~0.91	0.80~0.90
			P	-	<0.001	<0.001	<0.001
	训练集	SMOTE	值	0.96	0.91	0.91	0.91
			95%CI	0.94~0.99	0.79~0.97	0.84~0.95	0.86~0.95
			P	-	<0.001	<0.001	<0.001
	测试集		值	0.92	0.68	0.82	0.74
			95%CI	0.84~1	0.46~0.84	0.67~0.91	0.66~0.81
			P	-	0.11	<0.001	<0.001

方法	集合	模型对比	统计量	P
ROSE	训练集	随机森林与神经网络	-1.81	0.07
		神经网络与logistic	3.34	0.00
		随机森林与logistic	2.45	0.01
	测试集	随机森林与神经网络	-0.97	0.33
		神经网络与logistic	2.09	0.04
		随机森林与logistic	1.51	0.13
SMOTE	训练集	随机森林与神经网络	-0.77	0.44
		神经网络与logistic	0.42	0.67
		随机森林与logistic	-0.66	0.51
	测试集	随机森林与神经网络	0.79	0.43
		神经网络与logistic	-1.00	0.32
		随机森林与logistic	0.63	0.53

项目	回归系数	标准误	χ²值	P	OR	95%CI
HLA<9/10	-1.61	0.73	4.90	0.027	0.20	0.05~0.83
预处理方案强度	-4.62	1.13	16.66	<0.001	0.01	0.001~0.09
围植入综合征	-3.15	0.75	17.55	<0.001	0.04	0.01~0.19
42天内感染EBV	3.87	1.38	7.84	0.005	47.73	3.19~714.97
JMML	4.62	1.17	15.54	<0.001	101.64	10.21~1 011.37