Diabetic Cardiovascular Disease Risk Assessment Based On Causal Stability Learning

Diabetes is a common chronic metabolic disease that causes a variety of complications that pose a serious threat to human life and health,including diabetic nephropathy and diabetic cardiovascular disease,the prevention and treatment of which has become a major social challenge.Diabetic cardiovascular disease is a major cause of disability and death in diabetic patients,but it is also a preventable and controllable disease.Early intervention and treatment of diabetic cardiovascular disease can effectively control the progression of the disease and improve the survival rate of diabetic patients.Therefore,it is of great research significance and clinical value to seek an effective risk assessment method for early prevention and treatment of the diabetic cardiovascular disease.Much of the current research on diabetic cardiovascular disease is based on a statistical approach that has made some progress in correlating risk features such as patient’s age and cholesterol with the risk of the target disease.However,this approach,which treats individual features as the same type of risk profile as metabolic features,lacks attention to individual features and ignores the causal relationship between risk features and disease risk,as well as the important disease background information of diabetes.In response,a new deep learning network model,a causally stable learning network based on individual feature interactions,is proposed for the diabetic cardiovascular disease risk assessment task based on a medical dataset collected from a tertiary hospital in Shanghai.As a result,the research in this paper can be divided into three main sections as follows:（1）An importance score-based metabolic feature selection method for diabetes is proposed for feature analysis and selection of diabetes datasets,thereby reducing the impact of irrelevant or redundant metabolic features on the diabetic cardiovascular disease risk assessment task.The method combines model-related and model-independent feature selection methods for feature importance scoring,followed by ranking using the scores,and finally feature selection through a common focus strategy.The experimental results show that the feature selection method proposed in this paper exhibits high accuracy and good model generalisation in the diabetic cardiovascular disease risk assessment task,providing support for achieving the research tasks in this paper.（2）A causally stable learning model based on individual feature interactions is proposed for diabetic cardiovascular disease risk assessment.The model takes into account the characteristics of diabetic disease and uses the patient’s long-term medical visit data as model input,with the output being an assessment of the patient’s risk of diabetic cardiovascular disease.To reduce the impact of comorbidity differences in the diabetes dataset and to enhance model stability,the model firstly uses a Causal and Time-aware Long Short-Term Memory（Causal-aware TLSTM）network to learn disease risk information from patients’ long-term metabolic features.Secondly,to emphasize the clinical significance of individual features,the model interacts individual features with the disease risk information obtained in the Causal-aware TLSTM based on an attention mechanism for feature interaction,resulting in a more complete and accurate disease risk representation with more information.Finally,the model uses a fully connected network for disease risk assessment.The experimental results show that the model proposed in this paper performs better in the diabetic cardiovascular disease risk assessment task and consistently outperforms the comparison model with experimental assessment metrics up to 94.33%,89.84%,93.33% and 93.90% in model Accuracy,Recall,F1-Score and area under the receiver operating characteristic curve（AUC）,respectively.（3）This paper designs and constructs a diabetic cardiovascular disease risk assessment aid system.The system is designed using the idea of separation of front and back ends,combining the feature selection method and the disease risk assessment model proposed in this paper,and is built according to the software development steps of requirements analysis,use case analysis and system design.The system is divided into an administrator side and a doctor side,with the administrator side mainly managing system data and user information,and the doctor side mainly managing patient consultation information and diabetic cardiovascular disease risk assessment tasks.