基于 stacking 模型融合与优化算法的 ICU 脓毒血症患者死亡风险预测研究*

娄一明 ,  赵瑞青 ,  崔宇 ,  仇丽霞

中国卫生统计 ›› 2026, Vol. 43 ›› Issue (3) : 333 -339.

PDF (2942KB)
中国卫生统计 ›› 2026, Vol. 43 ›› Issue (3) : 333 -339. DOI: 10.11783/j.issn.1002-3674.2026.03.003
论著

基于 stacking 模型融合与优化算法的 ICU 脓毒血症患者死亡风险预测研究*

作者信息 +

Prediction of Death Risk of Patients with Sepsis in ICU based on the Fusion of Stacking Model and Optimization Algorithm

Author information +
文章历史 +
PDF (3011K)

摘要

目的 基于MIMIC-Ⅲ(medical information mart for intensive care Ⅲ)数据库,采用stacking 异质集成融合 算法构建脓毒血症患者死亡风险预测模型,有助于早期识别和准确预测脓毒血症患者的死亡风险。方法 利用MIMIC- Ⅲ数据库中5163例脓毒血症患者数据,采用stacking模型预测死亡风险。研究分为五步:通过miceforest处理缺失值,随机 森林提取特征,使用Tomek links 欠采样、合成少数类过采样技术(synthetic minority over-sampling technique, SMOTE)以 及SMOTETomek 组合算法3 种方法处理数据不平衡,比较极端梯度提升、分类梯度提升和轻量梯度提升机模型,优化参 数。最后,以优化后的3 个模型为基础,logistic 回归为元分类器构建stacking 模型,预测脓毒血症患者死亡风险。结果 通过随机森林特征选择确定简化急性生理评分Ⅱ、年龄等14个关键变量,使用网格搜索、灰狼优化和哈里斯鹰优化(Harris hawks optimization,HHO)3 种优化算法提升模型性能,其中 SMOTETomek-lightGBM+HHO 组合最优。最终,使用 SMOTETomek 处理的初级学习器和 logistic 回归元分类器的 stacking 模型准确率为 0.9516、精确度为 0.9382、召回率为 0.9025、AUC 为0.9380、F1分数为0.9516,显示stacking 融合策略有效。结论 本研究构建的stacking 模型在预测准确性和 训练效率上均有优势,有助于重症监护室医师早期识别高风险患者,为临床诊治提供科学依据。

Abstract

Objective Based on the medical information mart for intensive care Ⅲ(MIMIC-Ⅲ) database, constructing a sepsis patient mortality risk prediction model using the stacking heterogeneous ensemble fusion algorithm can help in the early identification and accurate prediction of mortality risks in sepsis patients.Methods Utilizing data from 5163 sepsis cases in the MIMIC-Ⅲ database, the stacking model was employed to predict the risk of mortality. The study was divided into five steps: handling missing values with miceforest, feature extraction with random forest, addressing data imbalance with Tomek links, synthetic minority over-sampling technique (SMOTE) and SMOTETomek, comparing extreme gradient boosting, categorical boosting, light gradient boosting machine models, and optimizing parameters. Finally, based on the optimized three models, logistic regression was used as the meta-classifier to construct the stacking model for predicting the mortality risk of sepsis patients.Results Through random forest feature selection, 14 key variables including simplified acute physiology score Ⅱ and age were identified. The model performance was enhanced using three optimization algorithms: grid search, grey wolf optimizer, and Harris hawks optimization (HHO) algorithm, with the SMOTETomek-lightGBM+HHO combination proving to be the most optimal. Ultimately, the stacking model with a primary learner processed by SMOTETomek and a logistic regression meta-classifier achieved an accuracy of 0. 9516, precision of 0. 9382, recall of 0. 9025, AUC of 0. 9380, and anF1 score of 0. 9516, demonstrating the effectiveness of the stacking ensemble strategy.Conclusion The stacking model constructed in this study has advantages in both predictive accuracy and training efficiency, which helps intensive care unit physicians to early identify high-risk patients and provides a scientific basis for clinical diagnosis and treatment.

关键词

MIMIC-Ⅲ / 脓毒血症 / 机器学习 / 优化算法 / stacking集成

Key words

MIMIC-Ⅲ / Sepsis / Machine learning / Optimization algorithm / Stacking integration

引用本文

引用格式 ▾
娄一明,赵瑞青,崔宇,仇丽霞. 基于 stacking 模型融合与优化算法的 ICU 脓毒血症患者死亡风险预测研究*[J]. 中国卫生统计, 2026, 43(3): 333-339 DOI:10.11783/j.issn.1002-3674.2026.03.003

登录浏览全文

4963

注册一个新账户 忘记密码

参考文献

[1]

Rhodes A, Evans L, Alhazzani W, et al. Surviving Sepsis Campaign: International Guidelines for Management of Sepsis and Septic Shock: 2016 [J].Intensive Care Medicine, 2017, 43(3): 304-377.

[2]

Seymour CW, Liu VX, Iwashyna TJ, et al. Assessment of Clinical Criteria for Sepsis: For the Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis—3) [J].JAMA, 2016, 315(8): 762-774.

[3]

Kooistra EJ, van der Voort PHJ, van der Heijden GJMG, et al. Critical Care Predictive Models: A Systematic Review and Analysis [J].Critical Care Medicine, 2018, 46(11): 1768-1776.

[4]

Silver D, Huang A, Maddison CJ, et al. Mastering the game of Go without human knowledge [J].Nature, 2017, 550(7676): 354-359.

[5]

张平,吴念悦,张浩天,等. 心外科手术患者重症监护室住院时间预测模型研究[J].电子科技大学学报,2022,51(4): 500-505.

[6]

王羽,刘志玄,邱洪斌,等. 心力衰竭重症患者院内死亡率的预测模型:基于MIMIC—Ⅲ数据库的回顾性研究[J].中国动脉硬化杂志,2023,31(3):245-252.

[7]

Seymour CW, Rosengard AM, Angoff GH, et al. The Role of the Microbiome in Sepsis [J].Critical Care Medicine, 2018, 46(12): 2023-2030.

[8]

Johnson AEW, Ghassemi M, Niehaus K, et al. MIMIC—Ⅳ: A Large—Scale Database for Critical Care Research [J].Scientific Data, 2020, 7: 317.

[9]

Rios—Avila L, Zhu J, Paul J, et al. Multiple Imputation by Chained Equations in R: The mice Package [J].Journal of Statistical Software, 2019, 91(5): 1-16.

[10]

Geurts P, Ernst D, Wehenkel L. Extremely Randomized Trees [J].Machine Learning, 2006, 63(1): 3-42.

[11]

Hansen LK, Salamon P. Neural network ensembles [J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 1990, 12(10): 993-1001.

[12]

He H, Bai Y, Garcia EA, et al. ADASYN: Adaptive Synthetic Sampling Approach for Imbalanced Learning [J].IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), 2008, 39(2): 539-549.

[13]

Kohavi R. A Study of Cross—Validation and Bootstrap for Accuracy Estimation and Model Selection [C].Reprinted from the Proceedings of the 14th International Joint Conference on Artificial Intelligence (IJCAI), 1995.

[14]

Chen T, He T, Guestrin C. XGBoost: A Scalable Tree Boosting System [C].In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016.

[15]

Ke G, Meng Q, Finley T, et al. LightGBM: A Highly Efficient Gradient Boosting Decision Tree [C]. InAdvances in Neural Information Processing Systems, 2017.

[16]

Heidari AA, Mirjalili S, Faris H, et al. Harris hawks optimization: Algorithm and applications [J].Future Generation Computer Systems, 2019, 97: 849-872.

[17]

Batista GE, Prati RC, Monard MC, et al. A study of the behavior of several methods for balancing machine learning training data [J].ACM Sigkdd Explorations Newsletter, 2004, 6(1): 20-29.

[18]

Kuncheva LI, Whitaker CJ. Measures of diversity in classifier ensembles and their relationship with the ensemble accuracy [J].Machine Learning, 2003, 51(2): 181-207.

[19]

张海红,王斌,吴晓雅. 基于多因素logistic 回归及列线图模型预测脓毒症休克患者发生脑卒中风险[J].中国卫生统计,2022,39(4):604-606.

[20]

Singer M, Deutschman CS, Seymour CW, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis—3) [J].JAMA, 2016, 315(8): 801-810.

基金资助

*国家自然科学基金(81973155)

AI Summary AI Mindmap
PDF (2942KB)

0

访问

0

被引

详细

导航
相关文章

AI思维导图

/

〈 〉