Depression Syndrome Typing and Medication Pattern Analysis Through Unsupervised Clustering Combined With Latent Structure Dual Analysis

Abstract

Depression, a most common psychiatric disease, is defined in Traditional Chinese Medicine (TCM) as Yu Syndrome, i.e., depression disorder, or Baihe Disease, i.e., lily bulb disease, a category of emotional disorders treated with lily-based TCM preparations. In TCM, depression is managed through syndrome differentiation and treatment, which is characterized by high efficacy and safety. However, there is no unified standard for the classification of depression syndromes, which leads to a disconnection between the analysis of patients' medication patterns and their actual syndromes and hinders the study of medication patterns specific to particular syndromes. Therefore, this study is focused on investigating the medication patterns of different sub-types of depression patients based on an objective classification system of depression.

Methods 

We searched for and retrieved clinical literature on TCM formulas for depression from relevant databases, including China National Knowledge Infrastructure (CNKI), Wanfang Data, VIP Database, Sinomed, Web of Science, and PubMed. Information on patient symptoms and medication was standardized. Then, the symptoms and the medication frequency of depression patients were statistically analyzed. We used the K-means clustering method combined with implicit structural analysis to objectively categorize depression patients into sub-types. In addition, the main symptoms and core TCM formulas of each sub-type of depression patients were identified. On the basis of objective classification system, we also statistically analyzed the characteristics of herbs used on depression patients, including the 4 basic properties, the 5 flavors, the attributes, the therapeutic efficacy, and the co-occurrence patterns, which may help reveal the medication patterns.

Results 

 A total of 3537 publications and 4434 prescriptions were included in the analysis. By using the K-means algorithm and latent structure analysis methods, patients with depression were categorized into 9 sub-types, with Cluster 6 accounting for the largest proportion. The most common symptoms among depression patients were insomnia and a depressed mood. Medication frequency analysis showed that Radix Bupleuri (Chai Hu), Radix Paeoniae Alba (Bai Shao), Poria (Fu Ling), Rhizoma Chuanxiong (Chuan Xiong), and Radix Curcumae (Yu Jin) were the most commonly used TCM herbs. For the depression sub-types of Clusters 1, 2, and 6, blood-activating and stasis-dissolving herbs were used most often. The depression sub-types of Clusters 3, 4, 5, 8, and 9 were mainly treated with qi-regulating herbs, while the depression sub-type of Cluster 7 was treated with qi-supplementing herbs. Depression patients were mostly treated with herbs that were cold or warm in nature and had sweet, bitter, and pungent flavors. Moreover, treatments for Cluster 1 and Cluster 6 mainly targeted the spleen meridian, while those for Cluster 2, Cluster 3, Cluster 4 and Cluster 5 mainly targeted the heart meridian. The treatments for the other sub-types mainly targeted the liver meridian. The core TCM formulas for the 9 depression sub-types included Zishui Qinggan Decoction, Danzhi Xiaoyao Powder, Huanglian Wendan Tang, Chaihu Guizhi Tang, Modified Xiaoyao Powder, Qinggan Jieyu Tang, Xiaoyao Powder, Xuefu Zhuyu Decoction, and Bazhen Decoction. The most commonly used Chinese herbal medicinal formulas were Gan Cao-Chai Hu, Bai Shao-Chai Hu, and Chen Pi-Chai Hu.

Conclusion 

 Based on machine learning, this study reveals the scientific aspects of TCM typing and syndrome-based treatment. It clarifies the rationale for targeting different symptoms in depression treatment and provides theoretical support for clinicians to make medication prescriptions. It also presents a new perspective for investigating TCM medication patterns.

Keywords: Depression, Syndrome typing, Medication patterns, K-means, Latent structure, Machine learning