结合机器学习模型的早产儿ICU死亡风险评估与可解释性分析
2025-08-03苏燕凤洪素茹陈钰霜吴夏阳







[摘要] 目的 利用机器学习算法预测早产儿重症监护病房(intensive care unit,ICU)死亡风险,为临床医生提供早期诊断和风险评估的辅助决策工具。方法 回顾性收集儿科重症监护数据库中早产儿病例的临床数据。基于最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)回归分析和多因素Logistic回归分析,筛选影响早产儿预后的关键临床特征。研究通过人工少数类过采样技术算法平衡数据,结合7种机器学习模型构建预测模型并评估其性能。使用沙普利加性解释法(Shapley additive explanations,SHAP)进行模型解释。结果 最终纳入患儿923例,生存组886例,死亡组37例,共收集38个临床特征。LASSO筛选出8个与早产儿ICU死亡密切相关的变量(乳酸、呼吸频率、氯离子浓度、中性粒细胞、红细胞分布宽度等)。多因素Logistic回归分析显示乳酸、呼吸频率是早产儿ICU预后的独立影响因素。内部测试及外部验证显示轻量梯度提升机模型在准确性、精确性等指标上均优于其他模型。SHAP分析显示,呼吸频率和乳酸对早产儿死亡风险的预测贡献最大。结论 本研究为早产儿预后的早期识别和干预提供可靠工具,强调关键指标的重要性。
[关键词] 早产儿;ICU死亡风险;机器学习;轻量梯度提升机模型;风险预测
[中图分类号] R722.6" """"[文献标识码] A """[DOI] 10.3969/j.issn.1673-9701.2025.18.007
Mortality risk assessment and interpretability analysis of preterm infants in the ICU by using machine learning models
SU Yanfeng1, HONG Suru2, CHEN Yushuang3, WU Xiayang3
1.Department of Emergency Medicine, the Second Affiliated Hospital of Xiamen Medical College, Xiamen 361021, Fujian, China; 2.Department of Radiology, Xiamen Children’s Hospital, Xiamen 361006, Fujian, China; 3.Department of Pharmacy, Xiamen Children’s Hospital, Xiamen 361006, Fujian, China
[Abstract] Objective To aim at using machine learning algorithms to predict the risk of neonatal intensive care unit (ICU) mortality, providing clinicians with an early diagnosis and risk assessment tool to assist in decision-making. Methods Clinical data of preterm infants from the paediatric intensive care database retrospectively were collected. By using least absolute shrinkage and selection operator (LASSO) regression analysis and multivariate Logistic regression analysis, key clinical characteristics affecting preterm infant prognosis were screened. The study was balanced the data by using the synthetic minority oversampling technique, combined seven machine learning models to build a predictive model and evaluate its performance. The Shapley additive explanations (SHAP) was used for model interpretation. Results A total of 923 preterm infants were finally included, survival group comprised 886 infants, and death group comprised 37 infants. A total of 38 clinical characteristics were collected. LASSO screening identified 8 variables significantly associated with neonatal ICU mortality, including lactate, respiratory rate, chloride concentration, neutrophils, and red blood cell distribution width etc. Multivariate Logistic regression analysis revealed that lactate and respiratory rate were independent predictors of neonatal ICU outcomes. Internal testing and external validation showed that light gradient boosting machine model outperformed other models in terms of accuracy and precision etc. indicators. SHAP analysis indicated that respiratory rate and lactate levels had the largest predictive contribution to the risk of preterm infants mortality. Conclusion This study provides reliable tools for early identification and intervention in the prognosis of preterm infants, emphasizing the importance of key indicators.
[Key words] Preterm infants; ICU mortality risk; Machine learning; Light gradient boosting machine model; Risk prediction
早产儿因器官发育不成熟,出生后面临严重的健康挑战,尤以重症监护病房(intensive care unit,ICU)死亡率居高不下为甚。研究表明早产儿在新生儿重症监护病房(neonatal intensive care units,NICU)中的呼吸窘迫、中枢神经感染及坏死性小肠结肠炎等发生率明显高于足月儿,提示需加强监护和干预以降低死亡风险[1-2]。……
