皮肤微循环指标对职业工人颈动脉粥样硬化的预警价值研究*

Early warning value of skin microcirculation index for carotid atherosclerosis in occupational workers*

摘要:

目的 探索皮肤微循环指标对职业工人颈动脉粥样硬化（CAS）的预警价值，以期从皮肤微循环角度为CAS预警价值研究提供新的线索和依据。方法 选取2017年2月—2017年9月某重工业集团附属医院进行职业健康体检的324例职业工人作为研究对象，采用套索算法（LASSO）提取常用指标中的特征变量，通过朴素贝叶斯（NB）分类器构建职业人群CAS预警模型，使用受试者工作特征（ROC）曲线筛选对职业人群CAS筛检价值较高的单一皮肤微循环指标，进而建立常用指标+皮肤微循环预警模型，使用ROC曲线和决策曲线分析法（DCA）从曲线下面积（AUC）和净收益2方面分析比较其与初始预警模型的优势。结果 LASSO算法在压缩到第11步CP值最小（CP=28.185），此时纳入的变量有年龄、文化程度、吸烟、饮酒、食盐 程度、体育锻炼、动脉粥样硬化家族史、舒张压、低密度脂蛋白（LDL）、尿酸（UA）和超敏C反应蛋白（hs-CRP），预警模型AUC为0.758（95% CI：0.701，0.782），敏感性为93.3%（95% CI：0.917，0.948），特异性为47.1% （95% CI：0.458，0.486）。根据ROC曲线筛检皮肤微循环指标的检验结果，保留经皮氧分压（tcpO2）、踝肱指数 （ABI）和血流灌注百分比3项指标，并纳入预警模型，建立的常用指标+皮肤微循环预警模型，其AUC为0.835（95% CI：0.795，0.860），敏感性为76.7%（95% CI：0.739，0.796），特异性为88.2%（95% CI：0.846，0.917），DeLong检验表明后者拟合效果优于前者（P?<0.05）。DCA分析结果显示在阈概率为0.2～1.0，常用指标+皮肤微循环预警模型的净收益高于初始预警模型。结论 皮肤微循环指标对于职业人群CAS具有良好的预警价值和临床应用前景。

Abstract:

Objective To Explore the early warning value of skin microcirculation index for CAS of occupational workers, and to provide new clues and basis for CAS early warning value research from the perspective of skin microcirculation. Methods Totally 324 occupational workers were selected as the research object who took occupational health examinations in a certain hospital of the heavy industry group from February to September in 2017. The Lasso algorithm was used to extract the characteristic variables in the common indicators, and the warning model of the professional population CAS was constructed by the NB classifier. A single skin microcirculation index with high value for CAS screening in occupational population was selected by ROC curve and common indicator + skin microcirculation warning model was established, ROC curve and DCA analysis were used to compare its advantages with initial early warning models from both AUC and net income. Results The Lasso algorithm has the smallest CP value (CP = 28.185) after compressed to the 11th step. The variables included at this time are age, education level, smoking status, drinking status, salt level, physical exercise, AS family history, diastolic blood pressure, LDL, UA and Hs-CRP. The early warning model AUC is 0.758 (95% CI: 0.701, 0.782), the sensitivity is 0.933 (95% CI: 0.917, 0.948), and the specificity is 0.471 (95% CI: 0.458, 0.486). The results of screening the skin microcirculation index according to the ROC curve retained the three indicators of tcpO2, ABI and perfusion percentage and were included in the early warning model. The established common indicator + skin microcirculation warning model had AUC of 0.835 (95%CI: 0.795, 0.860), sensitivity of 0.767 (95% CI: 0.739, 0.796), and specificity of 0.882 (95% CI: 0.846, 0.917), the DeLong test showed that the latter fit better than the former (P?