VOCs可能作为慢阻肺、哮喘和PRISm早期评估指标
2025/10/27
背景：慢性呼吸系统疾病（如COPD和哮喘）已成为全球第三大死亡原因，长期威胁人类健康。为减轻疾病负担，亟需开发快速、便捷的COPD、保留比率肺功能受损（PRISm）及哮喘的筛查方法。呼出气挥发性有机化合物（VOCs）可反映疾病的病理生理过程，对呼吸系统疾病具有一定的诊断价值。作者旨在基于机器学习通过具有分类潜力的呼出气VOC标志物构建COPD、哮喘和PRISm的早期诊断模型。
方法：本研究为一项横断面研究，纳入184例COPD患者、66例哮喘患者、72例PRISm患者及45名健康人（2023年8月至2024年6月）。使用便携式微型气相色谱仪（CXBA-Alpha, ChromX Health Co., Ltd.）分析呼出气样本，并通过单变量和多变量分析筛选潜在VOC标志物。结合基线特征（性别、年龄、BMI、吸烟状况及职业暴露），采用五种机器学习算法建立分类模型，并计算其敏感性、特异性和准确性评估鉴别能力。
结果：共纳入367例受试者。呼吸组学特征筛选结果显示：可鉴别COPD与健康人、PRISm与健康人、哮喘与健康人、COPD与哮喘和PRISm与哮喘的VOC分别有9种、9种、5种、5种和7种。五种算法中，随机森林模型在区分COPD与健康人时表现最佳（ROC曲线下面积[AUC] = 0.92 ± 0.01）；支持向量机（SVC）模型在区分PRISm与健康人时AUC为0.78 ± 0.01；逻辑回归在区分哮喘与PRISm（AUC = 0.74 ± 0.02）以及哮喘与COPD（AUC = 0.92 ± 0.01）时表现最佳；随机森林模型在区分哮喘与健康对照时AUC为0.81 ± 0.02。
结论：基于VOC的分类模型可作为区分慢性呼吸系统疾病的新策略。使用便携式微型气相色谱可实现慢性呼吸系统疾病的快速检测，尤其是在群体中可快速识别PRISm个体。
关键词：呼出气；挥发性有机化合物；慢性阻塞性肺疾病；哮喘；保留比率肺功能受损；机器学习；早期检测；生物标志物
方法：本研究为一项横断面研究，纳入184例COPD患者、66例哮喘患者、72例PRISm患者及45名健康人（2023年8月至2024年6月）。使用便携式微型气相色谱仪（CXBA-Alpha, ChromX Health Co., Ltd.）分析呼出气样本，并通过单变量和多变量分析筛选潜在VOC标志物。结合基线特征（性别、年龄、BMI、吸烟状况及职业暴露），采用五种机器学习算法建立分类模型，并计算其敏感性、特异性和准确性评估鉴别能力。
结果：共纳入367例受试者。呼吸组学特征筛选结果显示：可鉴别COPD与健康人、PRISm与健康人、哮喘与健康人、COPD与哮喘和PRISm与哮喘的VOC分别有9种、9种、5种、5种和7种。五种算法中，随机森林模型在区分COPD与健康人时表现最佳（ROC曲线下面积[AUC] = 0.92 ± 0.01）；支持向量机（SVC）模型在区分PRISm与健康人时AUC为0.78 ± 0.01；逻辑回归在区分哮喘与PRISm（AUC = 0.74 ± 0.02）以及哮喘与COPD（AUC = 0.92 ± 0.01）时表现最佳；随机森林模型在区分哮喘与健康对照时AUC为0.81 ± 0.02。
结论：基于VOC的分类模型可作为区分慢性呼吸系统疾病的新策略。使用便携式微型气相色谱可实现慢性呼吸系统疾病的快速检测，尤其是在群体中可快速识别PRISm个体。
关键词：呼出气；挥发性有机化合物；慢性阻塞性肺疾病；哮喘；保留比率肺功能受损；机器学习；早期检测；生物标志物
（南方医科大学南方医院 黄海伦 龚钊乾 赵海金）
(Tian J, Zhang Q, Peng M, et al. Exhaled volatile organic compounds as novel biomarkers for early detection of COPD, asthma, and PRISm: a cross-sectional study[J]. Respiratory Research, 2025, 26: 173. DOI: 10.1186/s12931-025-03242-5)
(Tian J, Zhang Q, Peng M, et al. Exhaled volatile organic compounds as novel biomarkers for early detection of COPD, asthma, and PRISm: a cross-sectional study[J]. Respiratory Research, 2025, 26: 173. DOI: 10.1186/s12931-025-03242-5)
Abstract
Background: Globally, chronic respiratory diseases have become the third leading cause of death, including chronic obstructive pulmonary disease (COPD) and asthma, and have been threatening human life for a long time. To alleviate the disease burden, it is crucial to develop rapid and convenient screening methods for COPD, preserved ratio impaired spirometry (PRISm), and asthma. Volatile organic compounds (VOCs) in breath can reflect the pathophysiological processes of disease, thereby having the potential to serve as a promising approach for diagnosing respiratory diseases. Can we identify VOC markers in breath with the potential to serve as classification indicators, and further establish learning models for the early detection of COPD, asthma, or PRISm patients?
Methods：This is a cross-sectional study in which exhaled breath samples were collected from 184 patients with COPD, 66 patients with asthma, 72 PRISm individuals, and 45 healthy individuals. From August 2023 to June 2024, the breath samples were analyzed using portable micro gas chromatography (CXBA-Alpha, ChromX Health Co., Ltd.). Potential VOC markers for classification were identified by univariate and multivariate analyses. Subsequently, classification models were established by machine learning algorithms, based on these VOC markers along with baseline characteristics. The sensitivity, specificity, and accuracy of these models were calculated to assess their overall discriminatory performance.
Results：A total of 367 patients were enrolled in our study. We identified nine VOCs distinguishing COPD patients from healthy controls, nine VOCs differentiating the PRISm population from healthy controls, five VOCs separating asthma patients from healthy controls, five VOCs distinguishing COPD patients from asthma patients, and seven VOCs differentiating the PRISm population from asthma patients based on breathomics feature selection. We utilized five algorithms to establish diagnostic models and selected the optimal one among them. The random forest model best distinguished COPD from healthy controls with an area under the receiver operating characteristic curve (AUC) of 0.92 ± 0.01. The support vector classifier (SVC) model was most effective in separating PRISm from healthy controls, achieving an AUC of 0.78 ± 0.01. Logistic regression performed well in discriminating asthma from PRISm (AUC, 0.74 ± 0.02) and COPD (AUC, 0.92 ± 0.01), in contrast, the random forest model differentiated asthma from healthy controls with an AUC of 0.81 ± 0.02.
Conclusion：VOC panel-based classification models have the potential to be a novel strategy for the discrimination of chronic respiratory diseases. Using the portal micro gas chromatography enables swift detection of chronic respiratory disease and, most importantly, facilitates the rapid identification of PRISm individuals within the population.
Background: Globally, chronic respiratory diseases have become the third leading cause of death, including chronic obstructive pulmonary disease (COPD) and asthma, and have been threatening human life for a long time. To alleviate the disease burden, it is crucial to develop rapid and convenient screening methods for COPD, preserved ratio impaired spirometry (PRISm), and asthma. Volatile organic compounds (VOCs) in breath can reflect the pathophysiological processes of disease, thereby having the potential to serve as a promising approach for diagnosing respiratory diseases. Can we identify VOC markers in breath with the potential to serve as classification indicators, and further establish learning models for the early detection of COPD, asthma, or PRISm patients?
Methods：This is a cross-sectional study in which exhaled breath samples were collected from 184 patients with COPD, 66 patients with asthma, 72 PRISm individuals, and 45 healthy individuals. From August 2023 to June 2024, the breath samples were analyzed using portable micro gas chromatography (CXBA-Alpha, ChromX Health Co., Ltd.). Potential VOC markers for classification were identified by univariate and multivariate analyses. Subsequently, classification models were established by machine learning algorithms, based on these VOC markers along with baseline characteristics. The sensitivity, specificity, and accuracy of these models were calculated to assess their overall discriminatory performance.
Results：A total of 367 patients were enrolled in our study. We identified nine VOCs distinguishing COPD patients from healthy controls, nine VOCs differentiating the PRISm population from healthy controls, five VOCs separating asthma patients from healthy controls, five VOCs distinguishing COPD patients from asthma patients, and seven VOCs differentiating the PRISm population from asthma patients based on breathomics feature selection. We utilized five algorithms to establish diagnostic models and selected the optimal one among them. The random forest model best distinguished COPD from healthy controls with an area under the receiver operating characteristic curve (AUC) of 0.92 ± 0.01. The support vector classifier (SVC) model was most effective in separating PRISm from healthy controls, achieving an AUC of 0.78 ± 0.01. Logistic regression performed well in discriminating asthma from PRISm (AUC, 0.74 ± 0.02) and COPD (AUC, 0.92 ± 0.01), in contrast, the random forest model differentiated asthma from healthy controls with an AUC of 0.81 ± 0.02.
Conclusion：VOC panel-based classification models have the potential to be a novel strategy for the discrimination of chronic respiratory diseases. Using the portal micro gas chromatography enables swift detection of chronic respiratory disease and, most importantly, facilitates the rapid identification of PRISm individuals within the population.
上一篇：
没有了
下一篇：
美泊利单抗治疗重度哮喘呈现三种反应模式，临床缓解率存在显著差异
