基于机器学习的预测模型的开发,用于新生儿细菌脑膜炎的长期不良结果
Ying Chen1, Shengpei Wang2, Jing Wu3
1Capital Institute of Pediatrics, Department of Neonatology, Beijing, China; Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China.
Jornal de pediatria
|November 10, 2025
概括
机器学习有效地预测新生儿细菌脑膜炎 (NBM) 的不良结果. 随机森林模型显示出强大的临床实用性,识别高风险婴儿及时干预.
科学领域:
- 新生儿医学 新生儿医学
- 计算生物学是一种计算生物学.
- 临床信息学是一种临床信息学.
背景情况:
- 新生儿细菌性脑膜炎 (NBM) 对长期不良预后构成重大风险.
- 早期识别高风险婴儿对于有效管理和改善结果至关重要.
- 预测模型可以帮助临床医生根据预后对NBM患者进行分层.
研究的目的:
- 应用机器学习来选NBM中长期不良预后的风险因素.
- 开发和评估一个强大的NBM结果预测模型.
- 确定新生儿细菌脑膜炎不良预后的关键预测因素.
主要方法:
- 包括139名患有NBM的新生儿;分为预后良好 (n=94) 和不良 (n=45) 的组.
- 使用最小绝对收缩和选择运算符,Boruta和递归特征消除来进行特征选择.
- 构建了七个机器学习模型,以AUC,精度,灵敏度和特异性评估性能;使用Shapley添加式解释解释.
主要成果:
- 随机森林模型表现出卓越的临床适用性,具有高精度 (0.881),良好的校准 (布赖尔得分:0.123),以及平衡的灵敏度 (0.887) 和特异性 (0.878).
- 后勤回归显示出高的区分能力 (AUC:0.903).
- 发现的关键预测因素包括脑脊液白细胞计数,脑脊液蛋白质水平和发作.
结论:
- 机器学习模型,特别是随机森林,可靠地预测NBM患者的长期不良结果.
- 这些模型有助于识别患不良预后高风险的婴儿.
- 建议在不同的队列中进行进一步验证,以提高概括性和临床使用.
相关概念视频
01:22Bacterial Meningitis I: Introduction
Bacterial meningitis is a severe, life-threatening inflammation of the meninges, particularly the pia mater and arachnoid mater, affecting the subarachnoid space, ventricles, and cerebrospinal fluid (CSF). If untreated, it can lead to significant neurological complications or death.Causative AgentsCommon pathogens vary with age and immune status. In adults, major organisms include Streptococcus pneumoniae, Neisseria meningitidis, and Haemophilus influenzae. Streptococcus agalactiae (group B...
01:26Bacterial Meningitis II: Pathophysiology
Bacterial meningitis typically begins when pathogens such as Neisseria meningitidis and Streptococcus pneumoniae colonize the nasopharynx and invade the bloodstream. This process is facilitated by bacterial virulence factors, such as polysaccharide capsules, which resist phagocytosis and complement-mediated killing. Less commonly, bacteria reach the central nervous system via contiguous spread from infections like otitis media or sinusitis, through congenital or acquired dural defects, or...