Research On Personalized Clinical Pathway Recommendation Based On Hypergraph And Pre-training Strategy

A clinical pathway,also known as care pathway,integrated care pathway,critical pathway,or care map,is one of the main tools used to manage the quality in healthcare concerning the standardization of care processes,which aims to reduces the variability in clinical practice and improves outcomes Therefore,many countries have proposed their own national standard clinical pathways to guide hospitals in the management of patient visits.However,the items in the national standard clinical pathways are usually too general to be used directly,and the treatment process varies from person to person because of the different physical quality,morbidity symptoms,and drug resistance of patients,so how to deal with the individualization of clinical pathways is an important issue.In order to solve the problem of personalized clinical pathways,this thesis proposes a model of personalized clinical pathway recommendation based on hypergraphs,with the main purpose of obtaining treatment experience from treatment datasets.The model first uses the hypergraph model to characterize the clinical items,and then further uses the self-attention mechanism for training,which can effectively capture the importance of the current clinical items to the next recommended clinical items,and recommends personalize clinical item for the patients based on this.In order to further combine the guiding role of national standard clinical pathways,this thesis proposes a pre-training strategy for personalized clinical pathway recommendation,which is pre-trained on national standard clinical pathways and then finetuned on the real world clinical dataset.Specifically,this thesis introduces national standard clinical pathways for corresponding disease types,and after serialization,pre-training is performed on the processed national standard clinical pathways for two pre-training tasks,which can not only accelerate the convergence speed of the clinical dataset,but also improve the accuracy of the recommended clinical items to patients.In this way,both empirical knowledge from the real-world treatment dataset and guidance from the national standard clinical pathways can be obtained,leading to more accurate recommendations of personalized clinical pathways to patients.The thesis conducts experiments on personalized clinical pathway recommendations on the coronary heart disease clinical datasets of local hospitals and publicly available medical datasets,and is able to recommend more accurate clinical pathways to patients than some deep learning sequence models.In addition,the national standard clinical pathways for the corresponding diseases proposed by the National Health Planning Commission are referred to,and the medical knowledge is combined with the relevant medical knowledge to learn the representations of the clinical items,and the accuracy of the personalized clinical pathways and the convergence speed of the model can be further improved by pre-training the national standard clinical pathways.The personalized clinical pathway recommendation model proposed in this thesis can provide patients with more accurate and personalized recommendations for clinical items,which can assist doctors in decision making and improve the quality of medical services and save medical costs.