聚类分析在慢性阻塞性肺疾病表型研究中的应用探讨

Phenotyping of Chronic Obstructive Pulmonary Disease by Using Cluster Analysis

目的采用聚类分析的方法对慢性阻塞性肺疾病(COPD)患者综合特征的重要组成部分进行识别及分组,形成亚组(即表型),以探讨聚类分析在COPD患者表型研究中的价值。方法纳入北京同仁医院呼吸科的168例COPD患者,收集人口统计学资料、病程、体重指数(BMI)、吸烟指数、急性加重频率、合并症等。患者于稳定期行肺功能检查、HRCT肺气肿评分,评估呼吸困难程度(MMRC评分),进行COPD评估测试(CAT),测量6分钟步行距离。应用SPSS13.0统计学软件对168例患者进行聚类分析。结果根据慢性阻塞性肺疾病全球策略(GOLD)分级标准,168例患者中GOLD 1级5例,2级75例,3级75例,4级13例。各级患者在性别分布、BMI、吸烟指数、高血压、脑梗死患病率方面差异无统计学意义(P>0.05),但在年龄、病程、呼吸困难评分、6分钟步行距离、BODE评分、CAT评分、冠心病患病率、急性加重频率、肺气肿视觉评分等方面差异具有统计学意义(P<0.05)。按样本聚类的方法可将168例患者分为3类:年轻/轻度组、老年/重度组、老年/中度组。GOLD分级相同的患者可出现在不同的组中,3组之间在年龄、BMI、急性加重频率、呼吸困难评分、CAT评分、合并症的患病率等方面存在差异。相反,不同GOLD分级的患者也可出现在同一组中。结果还发现肺气肿视觉评分也是区分各类特征的重要指标,提示肺气肿是COPD的一个重要表型。结论聚类分析能最大限度地将同质的患者归为一类,将不同质的患者划分到不同的类中,将其用于COPD的表型研究具有一定的临床价值。

Objective To investigate the phenotyping of COPD by cluster analysis and evaluate the value of this method. Methods 168 COPD patients were enrolled from Beijing Tongren Hospital. Demographic and clinical data, such as, sex, age, body mass index （ BMI）, smoking index, course of disease, exacerbation rate, and comorbidities were collected. Pulmonary function test, emphysema scoring by HRCT, dyspnea by MMRC score, COPD assessment test （CAT） score,six-minute walk test were performed for each patient during the stable stage. Cluster analysis was conducted using SPSS 13.0. Results According to the GOLD criteria,5,75,75, and 13 patients were classified into GOLD stage 1,2,3, and 4, respectively. There was no difference among different stages in sex distribution, BMI, smoking index, hypertension, and cerebral infarction incidence（P 〉 O. 05 ）, but the differences in age, disease course, dyspnea score, six-minute walk distance, BODE score, CAT score, coronary heart disease, exacerbation rate, and HRCT emphysema visual score were significant（ P 〈 0.05 ）. By cluster analysis, 168 patients were finally classified into three groups： younger/mild, older/severe, and older/moderate. The patients with the same GOLD stage appeared in different clusters and the patients belonging to different GOLD stages could be in the same cluster. There were significant differences among three groups in age, BMI, exacerbation rate, dyspnea score, CAT score, and comorbidities. The result showed that HRCT emphysema visual score was also an important index to differentiate clusters, suggesting that emphysema was an important phenotype of COPD. Conclusions Cluster analysis can classify homogeneous subjects into the same cluster, and heterogeneous subjects into different clusters. The results suggest that COPD phenotyping by cluster analysis is clinically useful and significant.