Research On Precision Medicine For Sepsis Based On Process Mining

At present,information systems have been fully popularized in the medical field.Since many medical pathways exist in the vast medical database,how to obtain effective treatment pathways from them has become a current research hotspot.Sepsis is characterized by rapid morbidity and high mortality.It is difficult to develop a specific medical plan through clinical pathways because of the diverse etiology and the different degrees of severity and urgency among cases.The rapid development of process mining has led to new solutions for sepsis treatment.Using process mining technology to mine and analyze medical logs of sepsis can propose reasonable and precise medical treatment plans based on case characteristics.The following challenges are faced in discovering treatment pathways for sepsis using process mining technology:(1)medical logs are stored in different data tables in the form of relational data,not the form of "case-event" hierarchy required for process discovery;(2)existing process discovery algorithms treat multiple business flow patterns in a log as one business flow pattern,causing the same events between different business flow patterns to interfere with each other.It can result in a complex,unreadable and meaningless process model;(3)traditional process mining techniques focus on discovering process models,using them to replay the process and judge the rationality of the process.However,implementing treatment pathway recommendations requires a combination of other techniques.To address the above challenges,the following work is carried out in this paper.(1)The NCSV2XES algorithm was proposed to convert multiple data tables of relational databases into hierarchical tag language event log files by artificially calibrating the case ID,activity name,occurrence time and case characteristic labels in existing relational logs.The XES2Train algorithm was used to obtain treatment event sequences and case characteristic label groups from the generated event log files as training datasets for the GRU-Attention neural network classification model.(2)Before mining the process model,the cases in the existing log files were clustered using the trace clustering algorithm LMTC,and the treatment plans in each cluster had similar attributes and pathway characteristics.The process model corresponding to each cluster is mined separately by combining the log noise processing algorithm ABPD and the process discovery algorithm NLIM.The obtained multiple process models corresponded to multiple treatment plans for sepsis case characteristics,and this approach avoided as much as possible the confusion of the model by adding the same events to multiple plans.(3)Using a process tree as a process model and optimizing the cyclic structure in the process tree to a specified sequential structure made the model more applicable to medical pathway recommendation.The treatment pathways were derived from the process tree model.The case characteristic labels corresponding to the treatment pathways were determined by the GRU-Attention neural network classification model.(4)Design a process mining based accurate classification treatment system for sepsis.By importing medical logs,the system analyzed the medical logs of sepsis cases and generated a recommendation model.Users input the case feature labels of sepsis patients,and the system will output one or more recommendations for treatment pathways based on the recommendation model.