Research On Health Risk Appraisal Of Medical Staff Based On Hierarchical Bayesian Network

Physical examination data is one of the most important data sources and data types in big data on health care.With the arrival of the era of big data,a large number of medical examination records in various industries have been accumulated in the medical system.It is of great significance to conduct health risk appraisal through medical examination data to provide health management solutions.This thesis discussing the analysis method of medical examination record data will provide a more accurate basis for health management.To appraise the health risks of medical examiners and provide health man-agement solutions,a common approach is a risk measurement method based on probabilistic estimation,inferring the abnormal probability of each medical examination item based on known personal information of the medical exam-iner.Bayesian networks explore the correlations among different information variables based on health examination data and make posterior probability infer-ence.Therefore,this thesis assesses the health risks of physical examiners based on a Bayesian network model.In this thesis,the Bayesian network for characterizing medical examination data is divided into two levels,"basic information layer" and "medical examina-tion item layer".A method for learning the hierarchical Bayesian network from medical examination data with missing values is designed.First,the connections within the basic information layer are learned based on the semi-naive Bayes algorithm(TAN),and the directions of inter-layer connections are constrained(called hierarchical TAN-type constraint,HTC).Then,the overall structure of the network(including connections within the medical examination item layer and connections between the two layers)is learned using the ensemble method(Bagging),i.e.,bootstrap is used to obtain the confidence level of node connec-tions,where the basic learning algorithm is the structure EM algorithm.To further improve the generalization performance of the model,random subspace is used in Bagging to filter variables randomly.Finally,the health risk appraisal is based on inferences from the hierarchical Bayesian network to provide a more accurate analysis of the physical examination items for the examiners.Through the spread of COVID-19,we have become more aware that medical personnel play an important role in ensuring social safty and personal health,but are facing huge occupational health risks.Therefore,it has important relevance to analyze the medical staff's medical examination data and conduct health checks based on relevant risk factors.In this thesis,the hierarchical Bayesian network and its structural learning algorithm are used to analyze the medical staff's phys-ical examination data of a triple-A hospital in China for seven consecutive years.Empirical evidence shows that the hierarchical Bayesian network structure shows the high-risk disease characteristics and trends of the hospital's medical staff,and the predictive performance of the model has some advantages over SEM al-gorithm,which can provide a basis for future health management of the medical staff.