基于量表序列相似性的艾滋病患者精神卫生状况聚类分析

Cluster Analysis of Acquired Immunodeficiency Syndrome Patients Mental Status Based on the Sequence Similarity of Scale

目的基于焦虑、抑郁和睡眠障碍量表对艾滋病患者的精神卫生状况进行综合评价。方法将量表的条目作为序列,利用动态时间规整(Dynamic Time Warping,DTW)方法计算患者每个量表得分的相似性,取平均值后得到患者间的相似性。使用层次聚类法进行聚类分析,以Calinski Harabasz Index(CHI)、Davies Bouldin Index(DBI)和轮廓系数(Silhouette Coefficient,SC)3个指标评价聚类质量并确定最优聚类数。根据每个聚类内不同精神状况患者的分布,为各聚类赋予标签,计算F1值评价聚类的准确性。另外使用欧氏距离以及基于单一量表的相似性进行聚类分析,比较二者的聚类质量、准确性、临床解释性。结果基于量表DTW相似性的患者聚类在CHI、DBI和SC 3个聚类质量评价指标上均优于基于欧氏距离的聚类(166.24 vs.72.68、2.91 vs.4.25、0.31 vs.0.16)。使用DTW相似性作为距离测度进行聚类时,利用焦虑、抑郁和睡眠障碍3个量表得分的聚类结果F1值(0.739)高于使用单一量表得分的聚类结果F1值(0.618、0.695、0.693)。结论基于多量表序列相似性的聚类方法在患者聚类方面表现出较好性能,可以对艾滋病患者的精神卫生状况筛查和评估提供帮助。

Objective To comprehensively evaluate the mental health status of acquired immunodeficiency syndrome patients based on the anxiety,depression and sleep disorder scales.Methods All item scores of a scale was regarded as a sequence,and the dynamic time warping(DTW)method was used to calculate the similarity between patients for each scale.The similarity between patients was obtained by averaging the similarities for all scales.The hierarchical clustering method was used for cluster analysis.Three indicators,the Calinski Harabasz Index(CHI),Davies Bouldin Index(DBI)and silhouette coefficient(SC)were used to evaluate the quality of the clusters and to determine the optimal number of clusters.The clusters were labelled according to the distribution of patients with diferent mental conditions within each cluster.F1 score was calculated to evaluate the accuracy of the clusters.In addition,cluster analysis was also performed using Euclidean distance as well as similarity based on a single scale to compare their clustering quality,accuracy,and clinical interpretability.Results Patients clustering based on scale DTW similarity outperformed the clustering based on Euclidean distance on all three clustering quality evaluation metrics,CHI,DBI and SC(166.24 vs.72.68,2.91 vs.4.25,and 0.31 vs.0.16).When clustering using DTW similarity as a distance measure,the F1 value for clustering using three scale scores for anxiety,depression and sleep disorders(0.739)was higher than that for clustering using a single scale score(0.618,0.695,and 0.693).Conclusion Cluster analysis based on the sequence similarity of multiple scales has shown good performance in patient clustering.It can be helpful in screening and assessing the mental health status of acquired immunodeficiency syndrome patients.