Application of Topic Model in the Study of Type 2 Diabetes Treatment Plan
10.6039/j.issn.1001-0408.2017.23.11
- VernacularTitle:主题模型在2型糖尿病治疗方案研究中的应用
- Author:
Fan YANG
;
Haiying LI
;
Shaojie XU
;
Zhaozhong ZOU
;
Yiqin LI
;
Huahong CHEN
- Keywords:
Topic model;
Type 2 diabetes;
Treatment plan;
Complications;
Lab indexes
- From:
China Pharmacy
2017;28(23):3208-3212
- CountryChina
- Language:Chinese
-
Abstract:
OBJECTIVE:To explore the application of topic model in the study of type 2 diabetes treatment plan. METHODS:Clinical data of 650 inpatients with type 2 diabetes in our hospital during Jan. 2012-Jun. 2016 were analyzed retrospectively. The data of clinical diagnosis,lab indexes and clinical drug use were exchanged,summarized and merged by MATLAB R2014a software. Latent Dirichlet allocation and author topic model were adopted to extract the typical topics with topic probability value>0.1,and the topics was described by the complications with cumulative probability value>0.5. RESULTS:A total of 62 complications words,16 abnormal laboratory indexes groups and 20 treatment plans were obtained. A total of 4 typical topics were excavated(cumulative probability values for the first few complications were 0.8786,0.8247,0.8215,0.7536;topic probability value were 0.3364,0.2773,0.2035,0.1176,respectively) and were mainly characterized by peripheral neuropathy,peripheral vascular disease,abnormal lipid metabolism and microvascular lesions;abnormal lab indexes groups met the above characteristics. The complications with high distribution rate included diabetic peripheral neuropathy (0.5787), hypertension (0.3631),atherosclerosis (0.2789),hyperlipidemia (0.4578) and diabetic retinopathy (0.3143);main drugs included Insulin aspart injection,Insulin injection,Methylcobalamin dispersible tablets,etc. CONCLUSIONS:The complications of type 2 diabetes are characterized by peripheral neuropathy,peripheral vascular disease,abnormal lipid metabolism and microvascular lesions. The medication rules with clinical significance can be extracted from the clinical data by topic model.
