Huang Jiewei, Zhuang Jiajing, Zheng Huaxian, Yao Ling, Chen Qingquan, Wang Jiaqi, Fan Chunmei
The Clinical Laboratory Center of the Second Affiliated Hospital of Fujian Medical University, Quanzhou, Fujian Province, 362018, People's Republic of China.
The Graduate School of Fujian Medical University, Fuzhou, Fujian Province, 350108, People's Republic of China.
Nat Sci Sleep. 2024 Apr 24;16:413-428. doi: 10.2147/NSS.S453794. eCollection 2024.
Obstructive sleep apnea (OSA) is a common and potentially fatal sleep disorder. The purpose of this study was to construct an objective and easy-to-promote model based on common clinical biochemical indicators and demographic data for OSA screening.
The study collected the clinical data of patients who were referred to the Sleep Medicine Center of the Second Affiliated Hospital of Fujian Medical University from December 1, 2020, to July 31, 2023, including data for demographics, polysomnography (PSG), and 30 biochemical indicators. Univariate and multivariate analyses were performed to compare the differences between groups, and the Boruta method was used to analyze the importance of the predictors. We selected and compared 10 predictors using 4 machine learning algorithms which were "Gaussian Naive Bayes (GNB)", "Support Vector Machine (SVM)", "K Neighbors Classifier (KNN)", and "Logistic Regression (LR)". Finally, the optimal algorithm was selected to construct the final prediction model.
Among all the predictors of OSA, body mass index (BMI) showed the best predictive efficacy with an area under the receiver operating characteristic curve (AUC) = 0.699; among the predictors of biochemical indicators, triglyceride-glucose (TyG) index represented the best predictive performance (AUC = 0.656). The LR algorithm outperformed the 4 established machine learning (ML) algorithms, with an AUC (F1 score) of 0.794 (0.841), 0.777 (0.827), and 0.732 (0.788) in the training, validation, and testing cohorts, respectively.
We have constructed an efficient OSA screening tool. The introduction of biochemical indicators in ML-based prediction models can provide a reference for clinicians in determining whether patients with suspected OSA need PSG.
阻塞性睡眠呼吸暂停（OSA）是一种常见且可能致命的睡眠障碍。本研究的目的是基于常见的临床生化指标和人口统计学数据构建一个客观且易于推广的OSA筛查模型。
本研究收集了2020年12月1日至2023年7月31日转诊至福建医科大学附属第二医院睡眠医学中心的患者的临床数据，包括人口统计学数据、多导睡眠图（PSG）和30项生化指标。进行单因素和多因素分析以比较组间差异，并使用Boruta方法分析预测指标的重要性。我们使用4种机器学习算法，即“高斯朴素贝叶斯（GNB）”、“支持向量机（SVM）”、“K近邻分类器（KNN）”和“逻辑回归（LR）”，选择并比较了10个预测指标。最后，选择最优算法构建最终预测模型。
在OSA的所有预测指标中，体重指数（BMI）显示出最佳预测效能，受试者操作特征曲线下面积（AUC）=0.699；在生化指标预测指标中，甘油三酯-葡萄糖（TyG）指数表现出最佳预测性能（AUC = 0.656）。LR算法优于4种既定的机器学习（ML）算法，在训练、验证和测试队列中的AUC（F1分数）分别为0.794（0.841）、0.777（0.827）和0.732（0.788）。
我们构建了一种高效的OSA筛查工具。在基于机器学习的预测模型中引入生化指标可为临床医生判断疑似OSA患者是否需要进行PSG提供参考。
