|
| 使用生物标志物建立机器学习模型用于菌血症的早期预测 |
| Biomarker based machine learning model for early prediction of bacteremia |
| 投稿时间：2024-01-30 修订日期：2024-02-18 |
| DOI： |
| 中文关键词: 机器学习 降钙素原 C-反应蛋白 中性粒细胞明胶酶相关脂质运载蛋白 菌血症 |
| 英文关键词: machine learning procalcitonin C-reactive protein neutrophil gelatinase-associated lipocalin bacteremia |
| 基金项目:福建省自然科学基金(2021J011447) |
|
| 摘要点击次数: 106 |
| 全文下载次数: 0 |
| 中文摘要: |
| 目的 菌血症的早期识别和及时干预是降低发病率和死亡率的关键。血培养是诊断菌血症的金标准,但其敏感性低且周转时间长,难以满足临床诊断的需求。本研究的目的是通过回顾性分析,从若干与感染相关的生物标志物中筛选出最佳特征子集用于构建机器学习模型,用于菌血症的早期预测。方法 采用回顾性分析筛选出符合研究标准的菌血症和局部感染患者,并收集参与者的生物标志物数据。通过特征选择算法筛选最佳特征子集,并以此构建机器学习模型。使用精确率,召回率和准确率综合评估机器学习模型的性能。结果 本研究收集到247例菌血症患者为病例组,262例局部感染患者为对照组,并纳入12项生物标志物。通过特征选择算法,我们发现支持向量机(support vector machine, SVM)模型只须纳入5项生物标志物,其模型准确率为90.2%,AUC高达0.967。结论 通过特征选择算法与机器学习模型相结合的策略,本研究开发了3种菌血症的预测模型,其中SVM的性能最佳,为菌血症的早期诊断提供了数据支持。 |
| 英文摘要: |
| Objective Early detection and timely intervention of bacteremia are critical to reducing morbidity and mortality. Blood culture is the gold standard for the diagnosis of bacteremia, but its long turnaround time and low sensitivity make it difficult to meet the diagnostic needs of clinical practice. The aim of this study was to screen the best combination of several infection-related variables for constructing a machine learning model for early prediction of bacteremia by retrospective analysis. Methods A retrospective analysis was used to screen patients with bacteremia and localized infections who met the study criteria, and biomarker data were collected from participants. The best subset of features was screened by a feature selection algorithm and used to construct a machine learning model. The performance of the machine learning model was evaluated using a combination of precision, recall, and accuracy. Results In this study, 247 patients with bacteremia were collected as a case group and 262 patients with localized infections as a control group, and 12 biomarkers were included. Using a feature selection algorithm, we found that the support vector machine (SVM) model, which had to incorporate only 5 biomarkers, had a model accuracy of 90.2% with an AUC of up to 0.967. Conclusion Through the strategy of combining feature selection algorithms and machine learning models, three prediction models for bacteremia were developed in this study, among which SVM had the best performance and provided data support for the early diagnosis of bacteremia. |
|
View Fulltext
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
