Font Size: a A A

The Evolution And Prediction Of Comorbidities Networks In The Elderly Based On Machine Learning

Posted on:2022-03-28Degree:MasterType:Thesis
Country:ChinaCandidate:L TangFull Text:PDF
GTID:2517306524990099Subject:Master of Engineering
Abstract/Summary:PDF Full Text Request
With the aging of the world becoming more and more serious,comorbidities have become more and more common and have become the biggest threat to the lives of the elderly today.Because there may be some internal connections between comorbidities,it will bring many problems to treatment and care.If you can fully understand the rela-tionship between comorbidities,the relationship between patients and diseases and other multiple information,you can carry out overall comorbid care.,May be the key to help solve comorbidities.This article is based on the data on the first page of the medical records of hospi-tals in 21 prefecture-level or above hospitals in Sichuan Province from January 1,2015 to December 31,2019.In order to fully mine the data information and understand the correlation mechanism between diseases,we have done Work as follows:1)Based on exploratory factor analysis,K-Means++clustering algorithm,fuzzy C-means clustering algorithm and hierarchical clustering algorithm to explore comorbidity patterns,and to explore the differences of comorbidities based on different genders accord-ing to gender stratification.In the end,after multiple methods of mutual verification,five multimorbidity patterns were derived,they are skeletal:{Osteoporosis,spondylopathy,arthropathy,depression and thyroid disease};cardiovascular diseases and metabolism:{metabolic disorders,hypertension,atherosclerosis,other blood and hematopoietic organ diseases,cerebrovascular Diseases,gallbladder(biliary tract)and pancreatic diseases,di-abetes,chronic obstructive pulmonary disease and ischemic heart disease,malnutrition,tumors,liver diseases};kidney diseases:{Other diseases of kidney and ureter,prostate hy-perplasia,kidney Failure};Degenerative diseases:{Cataract,choroidal and retinal disor-ders};Mental category:{Organic(including symptomatic)mental disorders,Parkinson's,Alzheimer's disease,skin and subcutaneous tissue other Disease},respectively.2)Based on the association rule mining algorithm,the association analysis of chronic diseases is carried out,the comorbidity pairs are mined under different clusters,and the comorbidity network is constructed according to the analysis results.32 association rules and 33 disease combinations were mined from 7 diseases in cluster 1;6 association rules were mined from 6 diseases in cluster 2,and there were 11 comorbid relationships;clus-ter 3 contained 13 diseases,a total of Mined 103 chronic disease combinations and 79 association rules.3)Based on the graph neural network,a disease prediction model for chronic diseases is proposed.This article is based on the graph attention neural network,the heterogeneous graph attention network and the improved heterogeneous graph attention network model to train and adjust the parameters,and finally found that the improved The effect of the heterogeneous graph attention network is the best,with a precision of 0.33,a recall of 0.42,a micro F1 score of 0.3696,a macro F1 score of 0.3671,and a hamming loss of 0.4547.This study uses hospital medical records data from multiple hospitals in Sichuan Province to explore comorbidity patterns,conduct correlation analysis and construct co-morbidity networks,realize network visualization,train chronic disease prediction models based on graph neural networks,and study chronic diseases There is a certain practical significance.
Keywords/Search Tags:Chronic disease, Multimorbidity Patterns, Multimorbidity correlation, Graph neural network model, Chronic disease prediction model
PDF Full Text Request
Related items