Research On Key Drug Discovery And Mortality Risk Prediction For Electronic Medical Records

Electronic medical records are an electronic record system that can collect,store,manage,and share patient medical information to improve patient care.It can help doctors more effectively manage and access patient information,improve the efficiency of medical services,reduce medical costs,and improve the security and accessibility of medical information.In addition,the rich content contained in electronic medical records also has great secondary utilization value.Through data analysis of electronic medical records,useful information in the patient’s diagnosis and treatment process can be effectively mined,laying the foundation for further implementation and improvement of clinical decision support systems.Starting from the secondary utilization of electronic medical records,this thesis first discovers the key drugs contained in the electronic medical records to provide theoretical guidance for the selection of clinical drugs in the patient’s diagnosis and treatment process.Secondly,by analyzing the patient’s diagnosis,medication and surgery information,the death risk of the patient is predicted,so that more effective treatment measures can be taken to improve the survival rate of the patient.First,the thesis models the medication data in the electronic medical records as a complex network,and selects the key drugs in the treatment process by analyzing the degree and closeness centrality of the drug nodes.To make the results more intuitive,the analysis results of each index are visualized.However,the commonly used methods have certain limitations,so the thesis proposes a key node discovery method based on decision matrix.The experimental results show that the method proposed in this thesis can more effectively mine the key node information than the commonly used methods.Secondly,the thesis extracts the patient’s diagnosis,medication and surgery information when visiting the electronic medical records,and uses the information of the three aspects to predict the patient’s death risk.Because the existing death risk prediction research mostly limits the use of data to a single aspect,and requires specific and laborious data preprocessing work,with poor portability,this thesis proposes an easily expandable two-layer attention mechanism neural network to predict the patient’s mortality rate.The first layer attention network uses three independent LSTMs to respectively extract the attention features of the patient’s single admission diagnosis,medication and surgery,and the three are summarized and input into the second layer attention network using Transformer Encoder.Finally,the full connection layer is used to obtain the death rate prediction result of the patient.The experiment uses the large public electronic medical records database MIMIC-III,and through the comparison experiment and ablation experiment with the baseline method,the feasibility of the model is proved,and it can consider the information of the patient’s diagnosis,medication and surgery comprehensively,with good practicality and scalability.