Adverse Event Warning And Disposal Scheme Recommendation System Based On Collaborative Filtering And Knowledge Graph

With the continuous improvement of medical safety and medical efficiency,traditional medical adverse event management has been far from meeting the needs of medical staff and patients.In the future development trend,whether it can give early warning of possible adverse events in advance to effectively reduce the incidence of adverse medical events,whether it can assist medical staff to properly handle the events after the occurrence of adverse events,prevent serious consequences,and increase the proportion of adverse events with "no obvious adverse consequences".They are two very important indexes to judge the medical quality of a hospital.Therefore,the main research content of this thesis is to discuss how to reduce the incidence of adverse events in hospitals and improve the treatment effect of adverse events by constructing algorithms and writing systems.Specifically,the work content includes the following four parts:(1)The model of early warning of adverse events on admission was proposedIn view of the problems of waste of resources,low efficiency,poor prevention effect and high incidence of adverse events existing in the traditional adverse event prevention method that can only monitor patients after admission,An Early Warning Of Adverse Events On Admission model is proposed.There are two methods to implement this model.The adverse event warning algorithm Event＿Warring was constructed based on collaborative filtering and knowledge graph.The algorithm firstly constructs two kinds of data sets,namely patient basic information score vector or patient basic information knowledge map.The newly admitted patients will be substituted into the data set,and the responsible person information set and patient information set similar to their information can be calculated by superimposing score vector dimension or adding constraint condition successively.Then according to these two sets,the set of adverse events that the patient may experience is calculated respectively.Then the two adverse event sets were tested according to different fusion ratios to find the optimal value;Finally,the program was written according to the algorithm and used online in the department.The results show that the adverse event warning algorithms based on the two ideas both have good warning effects and can effectively reduce the incidence of adverse events in the departments.(2)Recommended methods of adverse event disposal programs are proposedIn view of the traditional problem that patients can only rely on the personal experience of medical staff to solve problems after adverse events,the disposal plan cannot be shared,the adverse event treatment effect is poor,and the proportion of "no obvious adverse consequences" events is low,A Recommended Adverse Event Disposal Plan is proposed.There are two methods to realize this mode,namely,constructing an Adverse Event Disposal Plan recommendation algorithm Event＿recommend based on collaborative filtering and knowledge graph respectively.The algorithm firstly construct adverse events or adverse events knowledge score vector map two data sets,and then to the adverse events information into data set calculation in order to get similar set of adverse events,then the collection with the patient data sets for joint calculation again,to filter to the adverse events collection,the collection of the screening results as the final recommendation;Finally,the program was written according to the algorithm and used online in the department.The results show that the recommendation algorithms of adverse event disposal schemes constructed based on the two ideas both have good recommendation effects and both effectively improve the proportion of "no obvious adverse consequences" events in the departments used.(3)a strategy to increase the computing dimension of the early warning algorithm and recommendation algorithm is proposedPilot department in the hospital for algorithm used in the process of a drop in the incidence of adverse events not below 6.6%,for the early warning algorithm added feature is proposed to calculate dimensions,the method of characteristics of this method by increasing patient and responsible persons of triples,extended dimension calculation,the patient and responsible individual information added to the algorithm of iterative computation,This method further improves the effect of the early warning algorithm.In pilot unit in the process of using the ’no significant adverse consequences of events accounted for more than 84%cannot ascend to this problem,put forward the disposal scheme as the core of the improved method,the method in addition to considering the similarity of patients with adverse events,added/disposal scheme used for the calculation of dimension,and use it as a core,The accuracy and safety of the recommendation results are further improved.(4)an early warning and recommendation algorithm construction model combining collaborative filtering and knowledge graph is proposedThe adverse event warning and recommendation algorithm constructed based on a single idea has low accuracy,recall rate and F1 value,and the treatment effects of different types of adverse events are quite different,which cannot be used in all departments of the hospital.Collaborative filtering and knowledge graph are combined to construct a new adverse event warning algorithm and adverse event disposal scheme recommendation algorithm,and the two kinds of algorithms are fused in two ways.The fusion logic of the early warning algorithm is as follows: the algorithm based on collaborative filtering is used to process the data of patients,and the algorithm based on knowledge graph is used to process the data of responsible persons.In Method 1,the results of the two processes are directly fused.In method 2,the recommendation set of the knowledge graph was first substituted back into the Patient score data set(Patient)for secondary screening,and then the screened results were fused with those based on the collaborative filtering algorithm.The fusion method of the recommendation algorithm of the disposal plan is as follows: Method 1 first uses the algorithm of the two ideas to process the incoming adverse event data respectively;Directly fuse the recommendation results of the two algorithms;Method 2: First determine the categories of adverse events to be treated,then select different computing strategies according to different categories,and finally fuse the results.The experimental results show that the algorithm constructed after the fusion of the two ideas not only effectively improves the accuracy,recall rate and F1 value,but also well realizes the ability to deal with all kinds of adverse events with high quality,which lays a foundation for the subsequent online use of the system in the whole hospital.