Machine Learning Based PSD For AIS Patients Risk Prediction Model Study

Objective: To describe the occurrence of PSD in patients with AIS 1 month after onset,construct a risk prediction model based on machine learning algorithms,and select the best model and analyze the key factors of modeling in order to provide a reference for medical personnel to identify and prevent PSD as early as possible.Methods:(1)Searched major Chinese and foreign databases,identified data extraction entries through literature review and group discussion;(2)Collected 490patients’ demographic,disease-related and psychosocial data,then followed up patients 1month after onset using PHQ-9,collated data and used SPSS25.0 to perform descriptive statistics,and then performed data preprocessing;(3)Based on the Python 3.8programming language,the data were randomly divided into training and testing data in the ratio of 8:2,then five single models and four composite models were constructed using two variable screening methods(no screening and LASSO screening),and the test data were used for validation,finally analyzed the key factors of modeling;(4)Compared the nine models through thier accuracy,precision,recall,F1 and AUC,then selected the best model based on AUC.Results:(1)The prevalence of PSD in AIS patients was 34.9%,including mild depression(25.7%),moderate depression(7.8%),and moderate to severe depression(1.4%);(2)five single models and four composite models were constructed based on no screening variables and LASSO regression screening.SSRS score,NIHSS score,family history of stroke,hs-CRP,history of hyperlipidemia,and BMI were the key factors for the models.(3)The LASSO＿DNN model with its AUC(0.8589)has the best predictive performance among all models for PSD risk prediction in AIS patients.Conclusion: The incidence of PSD in AIS patients was 34.9%;SSRS score,NIHSS score,family history of stroke,hs-CRP,history of hyperlipidemia,and BMI were considered as key factors for model construction;among the nine prediction models,the composite model LASSO＿DNN was the best model for PSD risk prediction in patients with AIS,it was conducive to improving the objectivity and accuracy of assessment results,helping medical personnel to identify patients with depressive tendencies in a timely manner,and to make focused interventions and develop individualized nursing intervention strategies,thereby improving patients’ quality of life and enhancing the overall quality of care.