临床儿科杂志 ›› 2025, Vol. 43 ›› Issue (7): 511-518.doi: 10.12372/jcp.2025.24e0304

• 论著 • 上一篇    下一篇

儿童白血病脐带血移植原发性植入失败预测模型的构建

张志奇1, 熊若兰1, 李泊涵1, 吉奇1, 王庆伟1, 卢俊1, 李捷1, 肖佩芳1(), 胡绍燕1,2()   

  1. 1.苏州大学附属儿童医院血液科(江苏苏州 215000)
    2.江苏省儿童血液肿瘤中心(江苏苏州 215000)
  • 收稿日期:2024-04-09 录用日期:2025-01-26 出版日期:2025-07-15 发布日期:2025-06-27
  • 通讯作者: 肖佩芳,胡绍燕 E-mail:xpfdr@163.com;hushaoyan@suda.edu.cn
  • 基金资助:
    科技部重大课题(2022YFC2502700);国家自然基金课题(82170218);国家自然基金课题(82300244);江苏省重大课题(BE2021654);苏州市重点实验室(SZS201615);苏州市重点实验室(SKY2022012);苏州市重点实验室(SZS2023014);苏州大学四方共建(ML13101223)

Construction of risk prediction model for primary graft failure after umbilical cord blood transplantation in pediatric leukemia

ZHANG Zhiqi1, XIONG Ruolan1, LI Bohan1, JI Qi1, WANG Qingwei1, LU Jun1, LI Jie1, XIAO Peifang1(), HU Shaoyan1,2()   

  1. 1. Department of Hematology and Oncology, Children's Hospital of Soochow University, Suzhou 215000, Jiangsu, China
    2. Jiangsu Pediatric Hematology & Oncology, Suzhou 215000, Jiangsu, China
  • Received:2024-04-09 Accepted:2025-01-26 Published:2025-07-15 Online:2025-06-27
  • Contact: XIAO Peifang, HU Shaoyan E-mail:xpfdr@163.com;hushaoyan@suda.edu.cn

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: 

  • R72