Research On Machine Learning For Chronic Kidney Disease Related Clinical Decision Support System

With the development of technology in China and the improvement of the medical security system,clinical decision support systems based on artificial intelligence and big data are gradually becoming an important tool to meet people’s health needs.Based on the actual needs from Tongde Hospital of Zhejiang Province,a collaborating institution,this paper focuses on the prediction of chronic kidney disease and identification of adverse drug reactions.In the first study,electronic medical records of 2213 patients were collected from Tongde Hospital of Zhejiang Province.Among them,639 patients subsequently developed chronic kidney disease.Then,we proposed a MD-BERT-LGBM method to establish a predictive model and compared it with other existing machine learning methods.Finally,we obtained the accuracy rate(83.06%),recall rate(73.17%),precision rate(63.93%),and area under the receiver operating characteristic curve(88.34%)of the model by a ten-fold cross-validation evaluation method,and we found that all of these results were better than other machine learning methods used for comparison.In the second study,355 papers that describe the relationship between drugs and chronic kidney disease were collected from the MEDLINE.Among them,87 papers mentioned drugs that had adverse effects on patients with chronic kidney disease.Then,this study adopted a strategy that expanded the training dataset and model the dataset with convolutional neural networks and bidirectional encoder representations from transformers respectively.The results show that the convolutional neural network performs better,with the accuracy rate(84.82%),recall rate(60.87%),precision rate of73.64%,and area under the receiver operating characteristic curve(88.38%).Finally,we also developed a graphical user interface for chronic kidney disease prediction and adverse drug reaction determination through a software tool that can be easy for physicians to use.The chronic kidney disease prediction model established in this paper can accurately predict an individual’s risk of developing chronic kidney disease in the future,and assist physicians in preventing and controlling the occurrence of chronic kidney disease.The adverse drug reaction identification model can provide early warning to physicians to avoid the use of drugs that are likely to cause adverse reactions on people who already had or may have chronic kidney disease in the future.Overall,the work in this paper has tremendous positive implications for the current problem of chronic kidney disease prevention.