Zhang Ye, Li Chen, Ji Yilin, Wei Bing, Guo Shubin, Mei Xue, Wang Junyu
Emergency Medicine Clinical Research Center, Beijing Chao-Yang Hospital &Capital Medical University, Beijing, 100000, People's Republic of China.
Shandong University of Traditional Chinese Medicine College of Optometry and Ophthalmology, Jinan, Shandong Province, 250355, People's Republic of China.
J Inflamm Res. 2025 May 19;18:6427-6437. doi: 10.2147/JIR.S502837. eCollection 2025.
The aim of our study was to establish and validate a machine learning-based predictive model for mortality risk in elderly patients with sepsis. By integrating traditional biomarkers, novel biomarkers, clinical data, and established scoring systems, the model seeks to enhance predictive accuracy and thereby improve clinical outcomes in high-risk patient population.
Conducted at Beijing Chao-Yang Hospital from August 2021 to August 2023, our study included 180 emergency department patients meeting Sepsis 3.0 diagnostic criteria. Data collected included patient demographics, vital signs, laboratory parameters, disease-related scores, major comorbidities, and the 28-day mortality. Variables were analyzed using univariate analysis and LASSO regression, and the machine learning model was constructed using R statistical software and validated internally via bootstrap resampling and calibration curves.
The model identified seven significant variables: SOFA, APACHE II, MAP, ALB, PCT, LTB, and VEGF. These variables constituted our final prediction model, which achieved an AUC of 0.845 (95% CI: 0.786, 0.905), with a sensitivity of 75.9% and a specificity of 85.0%. Internal validation yielded a bootstrap-corrected AUC of 0.857 (95% CI: 0.799, 0.912), confirming the model's statistical robustness. The nomogram provided a visual tool for predicting 28-day mortality risk, and decision curve analysis demonstrated strong potential for clinical utility.
The predictive model, which incorporates SOFA, APACHE II, MAP, ALB, PCT, LTB, and VEGF, shows significant potential in predicting the 28-day mortality risk for elderly sepsis patients. It provides a convenient and rapid tool for clinical use. Further research with larger sample sizes and external validation is warranted to confirm these findings and enhance the model's applicability.
本研究旨在建立并验证一种基于机器学习的老年脓毒症患者死亡风险预测模型。该模型通过整合传统生物标志物、新型生物标志物、临床数据和既定评分系统，力求提高预测准确性，从而改善高危患者群体的临床结局。
本研究于2021年8月至2023年8月在北京朝阳医院进行，纳入了180例符合脓毒症3.0诊断标准的急诊科患者。收集的数据包括患者人口统计学信息、生命体征、实验室参数、疾病相关评分、主要合并症以及28天死亡率。采用单因素分析和LASSO回归对变量进行分析，并使用R统计软件构建机器学习模型，通过自助重采样和校准曲线进行内部验证。
该模型识别出七个显著变量：序贯器官衰竭评估（SOFA）、急性生理与慢性健康状况评分系统II（APACHE II）、平均动脉压（MAP）、白蛋白（ALB）、降钙素原（PCT）、淋巴细胞亚群（LTB）和血管内皮生长因子（VEGF）。这些变量构成了我们最终的预测模型，其曲线下面积（AUC）为0.845（95%置信区间：0.786，0.905），灵敏度为75.9%，特异度为85.0%。内部验证得出自助校正后的AUC为0.857（95%置信区间：0.799，0.912），证实了该模型的统计学稳健性。列线图为预测28天死亡风险提供了直观工具，决策曲线分析表明其具有很强的临床应用潜力。
该预测模型纳入了SOFA、APACHE II、MAP、ALB、PCT、LTB和VEGF，在预测老年脓毒症患者28天死亡风险方面显示出显著潜力。它为临床应用提供了便捷快速的工具。有必要进行更大样本量的进一步研究和外部验证，以证实这些发现并提高模型的适用性。
