Construction of Early Warning Model for Infectious Diseases Based on Multidimensional Parameters Fusion of Blood Routine and Optimization of Decision Threshold
Abstract：Objective To construct an early warning machine learning model for infectious diseases based on multidimensional parameters of complete blood count(CBC),optimize decision thresholds,and improve the accuracy and clinical utility of infection diagnosis.Methods A multicenter retrospective cohort study was conducted,collecting data from 5,317 suspected infectious disease patients in the emergency departments of three research centers from January 2023 to December 2024,including 1,683 cases in the infection group and 3,634 cases in the non-infection control group.CBC results within 24 hours of hospital admission were collected,and routine,derived inflammatory,and cell population data parameters were extracted to construct three hierarchical feature sets.LASSO regression combined with random forest was used to screen features,and the predictive performance of six machine learning algorithms was compared.Six threshold determination methods were systematically evaluated,with the introduction of clinical utility weighting functions and adaptive threshold strategies.Model performance was assessed through internal and external validation.Results The extended fusion hierarchical XGBoost model achieved an AUC of 0.874 on the internal validation set,superior to the basic fusion hierarchical model;the dynamic fusion hierarchical AUC increased to 0.891.The XGBoost model outperformed other algorithms.The model demonstrated stable performance in external validation.Subgroup analysis of bacteremia showed better predictive performance for Gram-negative bacteremia.SHAP analysis indicated that the top five most contributing features were procalcitonin,etc.The adaptive threshold strategy significantly improved the net recclassification index,with a low threshold in the high-risk group achieving a sensitivity of 91.23%.Conclusion The XGBoost model based on multidimensional parameter fusion of CBC can effectively predict infectious diseases,and the introduction of cell population data enhances model performance.The adaptive threshold strategy enables individualized early warning,demonstrating clinical translational potential to provide decision support for early infection identification in emergency departments and primary healthcare institutions.
张瑜. 基于血常规多维参数融合的感染性疾病早期预警模型构建及决策阈值优化研究[J]. 哈尔滨医药, 2026, 46(3): 5-13.
Zhang Yu. Construction of Early Warning Model for Infectious Diseases Based on Multidimensional Parameters Fusion of Blood Routine and Optimization of Decision Threshold. journal1, 2026, 46(3): 5-13.
[1] Tang F,Yuan H,Li X,et al.Effect of delayed antibiotic use on mortality outcomes in patients with sepsis or septic shock:A systematic review and meta-analysis[J].Int Immunopharmacol,2024,129:111616. [2] 刘丹玉,姚佳丽,童红杰,等.中性粒细胞/淋巴细胞比值、血糖变异度对脓毒症患者预后的预测价值[J].中国现代医学杂志,2025,35(5):67-72. [3] Chiu YW,Chang YH,Hsu TY,et al.Monitoring for early prediction of gram-negative bacteremia using machine learning and hematological data in the emergency department[J].Commun Med(Lond),2025,5(1):483. [4] Gao Z,Wang X,Wang W,et al.Association between neutrophil to lymphocyte ratio and the mortality of patients with sepsis:an update systematic review and meta-analysis.Front.Med.12:1637365. [5] 冯晓丹,刘菲菲,杜瑶,等.基于超声影像组学及临床特征的可解释性机器学习模型预测乳腺癌新辅助化疗疗效[J].中国超声医学杂志,2024,40(3):278-282. [6] 李天时,朱轶,张忠满,等.革兰阴性杆菌血流感染的脓毒症相关性脑病患者临床特征分析[J].南京医科大学学报(自然科学版),2025,45(7):1001-1006. [7] Gao Z,Wang X,Wang W,et al.Association between neutrophil to lymphocyte ratio and the mortality of patients with sepsis:an update systematic review and meta-analysis[J].Front Med(Lausanne),2025,12:1637365. [8] 孙丹,吕国忠,曹玲,等.成年严重烧伤患者发生肠内营养不耐受风险预测模型的构建及验证[J].中华烧伤与创面修复杂志,2025,41(12):1-10. [9] 刘建模,罗颢文,俞鹏飞,等.基于机器学习的急性缺血性脑卒中医院感染预测模型建立与评价[J].中国感染控制杂志,2023,22(2):129-135. [10] 丁新焕,彭耀军,黄晶晶,等.急诊科脓毒症患者心脏骤停风险预测模型构建[J].中华急诊医学杂志,2023,32(12):1693-1698.
2026, Vol. 46 
