Analysis And Prediction Of Risk Factors For Anxiety And Depression After Ischemic Stroke Based On Machine Learning And Complex Network Models

Objective: Ischemic stroke,with high morbidity and disability rate,is regarded as one of the important diseases endangering the health of middle-aged and elderly people.Several studies indicated that ischemic stroke not only causes physical disorders,but also emotional disorders(the probability of occurrence is about 20%-50%),and brings a great burden to patients and their families,in which depression and anxiety are the most common post-stroke mood disorders.There is currently no effective predictive analysis method and the diagnosis of post-stroke anxiety and depression mainly relies on related scales.However,due to the complex and time-consuming evaluation process and the active cooperation of patients and their families,there are limitations in actual clinical practice.The development of artificial intelligence and big data technology has brought dawn to the medical field,such as Thin-Prep imaging system,robotic minimally invasive surgery and other application achievements have provided a lot of convenience for heavy clinical work.As the core of artificial intelligence applications,machine learning and complex network models play an important role.Therefore,this paper intends to utilize a combination of medical big data and artificial intelligence methods to predict post-stroke anxiety and depression through machine learning models,and establish a complex network model to analyze the risk factors for poststroke anxiety and depression to optimize the model.This model can effectively predict the state of anxiety and depression after ischemic stroke,and provide a strong basis and scientific guidance for clinical evaluation and intervention.Methods: A total of 416 ischemic stroke patients who met the inclusion and exclusion criteria admitted to the Department of Neurology,The First Affiliated Hospital of China Medical University from May 2017 to December 2021 were consecutively enrolled.Demographic data such as age,and medical history data such as smoking history,drinking history,fasting tests such as blood routine,liver function,kidney function,and other test indicators in the morning after admission were collected;brain magnetic resonance or brain CT images of patients were collected.The NIHSS score was used to assess the severity of stroke within 24 hours of admission.In order to better verify the accuracy of the machine learning model,the hospital anxiety and depression scale-anxiety subscale(HADS-A),Hamilton Anxiety Scale(HAMA),and Anxiety Selfrating Scale(SAS)were used to evaluate the patient’s anxiety state on the 10th-14 th day of onset.The degree of depression was assessed with the Hamilton Depression Scale(HAMD),and the patients were marked with post-stroke anxiety or depression according to each scale,based on random forests,support vector machines,decision trees,stochastic gradient descent,and multilayer perceptron.The five machine learning models were trained under supervised supervision,and indicators such as Area Under Curve(AUC),Euclidean Distance(ED),Root Mean Square Error(RMSE),and Mean Square Error(MSE)were selected to evaluate the prediction effect of the five learning models.Then,the risk factors of post-stroke anxiety and depression were evaluated based on the complex network model and compared with traditional statistical methods.Finally,we sort out the important risk factors that are suitable for input into the machine learning prediction model and verify through traditional statistical methods,which can finally verify and optimize the machine learning prediction model.Results:1.Experimental results based on machine learning prediction model(1)In the qualitative analysis of PSA based on 27 risk factors,five common machine learning prediction models showed that the random forest model had the highest accuracy,and the AUC results based on HADS-A,HAMA,and SAS scales were 0.71,0.67,and 0.65,respectively;(2)In the quantitative analysis of PSA based on 27 risk factors,the random forest model based on the HADS-A scale had the smallest average Euclidean Distance,Root Mean Square Error,and Mean Square Error(14.4879,2.2486,and 25.2953,respectively),The mean Euclidean Distance,Root Mean Square Error,and Mean Square Error of the random forest model based on the HAMA scale are the smallest(20.3966,3.1670,and 50.2419,respectively),and the mean Euclidean distance of the random forest model based on the SAS scale,the root mean square error and the Mean Square Error are the smallest(35.4597,5.5170 and 173.7589,respectively);(3)In the qualitative analysis of PSD based on 27 risk factors,the random forest model has the highest accuracy,and the AUC result is about 0.69;(4)In the quantitative analysis of PSD based on 27 risk factors,the random forest model had the smallest average Euclidean distance(13.5859),the smallest average Mean Square Error(22.5611),and the smallest average Root Mean Square Error(2.1074).2.Analysis of post-stroke anxiety and depression based on the complex network modela.Analysis of risk factors of post-anxiety and depression(1)The risk factors analyzed by the complex network model are consistent with the results obtained by traditional statistical methods.What’s more,in the complex network model of post-stroke anxiety and depression,the risk factors leading to post-stroke anxiety are hypertension history,drinking history,non-mild stroke(NIHSS score > 4),diabetes history and low level of high-density lipoprotein;(2)The risk factors analyzed by the complex network model are consistent with the results obtained by traditional statistical methods.What’s more,in the complex network model of post-stroke anxiety and depression,the risk factors leading to post-stroke depression are non-mild stroke(NIHSS score > 4),diabetes history,drinking history,and hypertension history in turn;(3)Based on the complex network model,the risk factors of post-stroke anxiety and depression are calculated as hypertension,drinking history,non-mild stroke(NIHSS score > 4),diabetes history,C-reactive protein,and high-density lipoprotein,fasting glucose,and triglyceride;b.Optimize the machine learning models via the analyzed vital risk factors of post-stroke anxiety and depression from the complex network theory.(1)The prediction results of the machine learning model of post-stroke anxiety based on the important risk factors of complex network analysis show that the random forest model has the best effect.The AUC value based on the HADSA scale was 0.75,the AUC value based on the HAMA scale was 0.81,and the AUC value based on the SAS scale was 0.72.(2)The prediction results of the machine learning model of post-stroke depression based on the important risk factors of complex network analysis showed that the random forest model had the best effect,and the AUC was 0.70.(3)Whether it is based on all the factors or the important risk factors analyzed by the complex network model,6)-fold cross-validation is performed on the machine model.With the increase of the 6)value,the prediction accuracy of post-stroke anxiety and depression improves.Conclusion:1.Machine learning models can be used to predict post-stroke anxiety or depression.Among the 5 machine learning models,the random forest model has the best prediction effect and is better than support vector machines,decision trees,stochastic gradient descent,and multilayer perceptron;2.History of hypertension,drinking history,non-mild stroke(NIHSS score > 4),diabetes history,and serum level of high-density lipoprotein are risk factors for anxiety in patients with acute ischemic stroke;3.Non-mild stroke(NIHSS score > 4),diabetes history,drinking history,and hypertension history are risk factors for depression in patients with ischemic stroke;4.The complex network model is capable of analyzing the risk factors of poststroke anxiety and depression and quantify the degree of risk.According to the computed risk factors,the machine learning model can be optimized to improve its prediction accuracy.