Research On Medical Treatment Behavior Prediction And Recommendation Model Based On Medical Insurance Data

In China,with the continuous development of national medical insurance policies,more and more people have joined the medical insurance.Predicting the future medical treatment behaviors of patients from historical medical insurance data is an important research hotspot,especially the prediction of hospitalization behavior because it is related to the allocation of government medical funds.Medical insurance data consist of medical visit sequences over time,where each visit contains information about patient,conditions,treatments,outcomes and so on.The most important challenge of this issue is how to correctly model such temporal and high dimensional data to significantly improve the prediction performance.Existing work mainly solves this problem by Recurrent Neural Networks(RNNs).However,the predictive power of RNNs drops significantly when the length of the patient visit sequences is large.At the same time,the method ignores the influence of the time interval within visit sequences on the modeling.For example,in a medical record,a patient has a small number of medical records for a long period of time or has a lot of medical records in a short period of time.In addition,this model not only does not have a good interpretability of the results,but also ignores the specific information of certain different groups.Moreover,whether the prediction results of the model are meaningful still require further verification.Therefore,the above problems are deeply studied:1.We propose an Attention and Time adjustment factors based Bidirectional LSTM hospitalization behavior prediction model(ATB-LSTM).The model uses a hidden layer to preserve the impact state of medical visit sequences at different time on future prediction,and introduces the attention mechanism and the time adjustment factor to jointly determine the strength of the hidden state at different moments,which significantly improves the predictive performance of the model.2.A grouping prediction model of medical treatment behavior based on tensor CP decomposition is proposed.This model is an improvement of ATB-LSTM model.It mainly includes two parts:Similarity grouping and Grouping prediction.For Similarity grouping,the model firstly constructs a heterogeneous information network basic on medical data,and uses a tensor CANDECOMP/PARAFAC(CP)decomposition method to achieve similarity grouping.In terms of grouping prediction,on the basis of bidirectional LSTM framework,an attention mechanism and time adjustment factor are introduced to achieve multi-task medical treatment behavior prediction,namely disease prediction and hospitalization behavior prediction.3.We propose a recommendation model for medical treatment migration based on two-layer CNN framework.The model firstly uses a BW-SMOTE algoritithm to unbalance the medical insurance data and reduce the imbalance of data distribution.Then,according to the prediction results obtained from the previous two parts of the study,that is,the situation of medical migration occurred,through therst layer of CNN similarity learning method,the insured who is similar to the test patient's medical treatment behavior can be found with the medical treatment within the region and outside the region.Through the second layer of CNN framework to learn the evaluation indicators of similar groups,and finally by comparing the evaluation indicators,it is concluded whether the test patients are worthy of medical treatment migration.Finally,the proposed models is evaluated by using real datasets.The experiment demonstrates that the prediction model proposed in this paper is more accurate than other prediction methods.Through the above research,this paper achieves the prediction of the patient's future medical treatment behavior.Medical institutions and government departments can make corresponding adjustment strategies based on the predictive results to maximize the value of medical resources.