Study Of Data-Driven Knowledge Discovery System In The Prognosis Of Heart Failure

Heart failure(HF)is a life-threatening chronic disease.The prognosis of HF is more complex and there are many factors that may cause the high re-hospitalization rate,high mortality rate and large medical burden.In recent years,the rapid development of high-tech technologies such as big data,machine learning and deep learning has brought new opportunities and challenges to the prediction and analysis of HF prognosis risks.The purpose of this study is to establish a predictive model for the prognosis of patients with HF and find out the rules for the prognosis of patients by using Explainable Artificial Intelligence.The study mainly uses artificial intelligence technology to establish the predictive model,discover medical knowledge and build the medical knowledge discovery system for patients with HF.This will provide decision information for clinicians,assist doctors in treating patients,reduce the risk of rehospitalization and mortality of patients with HF,and improve the medical resource allocation ability.The specific research contents are as follows:1.First,the data of patients hospitalized for HF in a domestic hospital were retrieved,and then the in-hospital clinical indicators of these patients were statistically analyzed.Machine learning technologies were used to establish logistic regression,support vector machine,random forest,artificial neural network and extreme gradient boosting models respectively to predict the prognosis of patients with HF.The main outcome is the in-hospital mortality within one year,and the secondary outcome is the all-cause readmission within one year and the use of positive inotropic drugs.Through the analysis and explanation of the prognosis risk model of patients with HF,the highrisk factors affecting the prognosis were found and the mortality risk tree of patients with HF is established according to the main outcome.The feasibility of the AI-based prognosis predictive model of patients with HF is verified either.2.Based on the data of patients hospitalized for heart failure in a domestic hospital,a machine learning method called Rule Fit including gradient bosting method and logistic regression was used to predict the 1-year in-hospital mortality of patients with HF,and medical rules could be extracted from the model by using Explainable Artificial Intelligence.Combining the knowledge of medical experts and statistical methods,a medical rule verification system was established to verify the medical and statistical validity of the extracted rules.3.Based on the data of patients hospitalized for heart failure in a domestic hospital,a Rule Net model based on the combination of machine learning and deep learning was established.The model first extracted the coarse-grained medical rules from the gradient boosting decision trees.The rules were embedded into the deep learning model based on attention mechanism.Rule Net could predict the prognosis risk of patients with HF through the end-to-end optimization method based on gradient descent,and learn the global importance information of rule features through the attention mechanism which is an Explainable Artificial Intelligence method to screen medical rules.4.Based on the data of patients hospitalized for heart failure in a domestic hospital,a data-driven medical knowledge discovery framework for prognosis risk of heart failure was established.The system mainly includes four parts: data generator,medical knowledge mining,medical knowledge evaluation and medical knowledge application.The data generator could select and preprocess the features required for modeling by using statistical methods and medical inclusion and exclusion criteria.Medical rule mining system applied the Rule Net model,which combines machine learning and deep learning,to predict the prognosis risk of patients with HF and extract medical rules based on the Explainable Artificial Intelligence method.Medical knowledge evaluation used case-control validation method based on propensity score and risk analysis method to validate the medical rules extracted from Rule Net automatically,and further confirmation of medical experts was also included to generate medical knowledge.The medical knowledge application added the medical knowledge to the traditional predictive model to improve the prediction ability,and a prognosis scale model of patients with HF based on medical knowledge was established either.In summary,the innovation of this paper mainly include: First of all,by combining machine learning and deep learning technology,the Rule Net model was proposed to predict the prognosis risk of patients with HF accurately,and relevant risk rules and protection rules were mined based on Explainable Artificial Intelligence method.Secondly,using the idea of case-control group and risk analysis,the automatic medical knowledge evaluation algorithm based on propensity score was proposed.An automatic medical knowledge evaluation system was established to optimize the randomized clinical trials,reduce workload of doctors and strengthen the reliability of knowledge-based medical rules obtained from medical knowledge mining.After the mining and evaluation of medical knowledge,a medical knowledge base was established for the prognosis risk of HF patients,and an accurate prediction scale model was established based on the knowledge base.Finally,an automatic prognosis risk knowledge discovery system for patients with HF was established by using the in-hospital data.The system includes data generator,medical knowledge mining,medical knowledge evaluation and medical knowledge application,so as to provide doctors with certain auxiliary decision-making suggestions and scientific research inspiration,and further promote the development of HF prognosis research.