A Study Based On Logistic Regression And XGboost Machine Learning Model For Predicting Stroke-Related Pneumonia

Objective:To investigate the incidence of stroke associated pneumonia(SAP)in stroke patients;To determine the influencing factors of SAP in stroke patients;Based on e Xtreme Gradient Boosting(XGboost)algorithm and Logistic regression,a prediction model for the risk of stroke associated pneumonia was constructed,and the prediction effects of the two models were compared to provide a basis for the risk prediction of stroke associated pneumonia.Methods:On the basis of literature review and analysis of the influencing factors of stroke patients with SAP,the data of 2634 stroke patients hospitalized in the department of neurology of a third level hospital in Chengde City from January 1,2020 to December 31,2021 were collected retrospectively;SPSS25.0 was used for single factor analysis and multi factor analysis;Based on XGbost method and Logistic regression,a prediction model for the risk of stroke associated pneumonia was constructed.The model evaluation indicators were accuracy,accuracy,sensitivity,f1,area under the curve(AUC)and Yodon index.Results:1.2634 eligible patients were finally included,with an average age of(64.31±11.86)years,1712(65%)males and 922(35%)females;The results of univariate analysis showed that the influencing factors of SAP in stroke patients were age,length of hospitalization,stroke type,long-term bedridden,smoking,dysphagia,pulmonary history,NIHSS score,Glasgow score,antibiotics,dehydrating agent or antacid,gastric tube was retained,days of gastric tube retention,laboratory examination,muscle strength grading,daily living ability score,and stroke occurred for the first time.2.Multivariate analysis showed that age(18～59 years old)(OR=2.705,CI=1.207～6.063,P=0.016),dysphagia(OR=0.355,CI=0.133～0.946,P=0.038),antibiotics(OR=0.295,CI=0.132～0.661,P=0.003),NIHSS score(0～4)(OR=23.961,CI=5.227～109.831,P=0.000),NIHSS score(5～9)(OR=7.605,CI=2.224～26.008,P=0.001)NIHSS score(10～14)(OR=4.458,CI= 1.827～10.874,P=0.001)is an independent influencing factor for SAP occurrence.3.In the XGboost model,the top five variables in the order of importance of influencing factors are nasal feeding,dysphagia,increased neutrophil/lymphocyte ratio,long-term bed rest and NIHSS score.Evaluation indexes of the two models,accuracy(XGboost=0.93 ± 0.00,Logistic=0.92 ± 0.01);Accuracy(XGboost=0.12 ± 0.06,Logistic=0.19 ± 0.03);Sensitivity(XGboost=0.48 ± 0.12,Logistic=0.32 ± 0.19);f1(XGboost=0.27 ± 0.05,Logistic =0.17± 0.09);AUC(XGboost=0.91 ± 0.02,Logistic=0.87 ± 0.04);Jordan index(XGboost=0.42 ± 0.12,Logistic=0.26 ± 0.19).The prediction effect of XGboost model is better than that of Logistic regression model.Conclusion:The prediction model of SAP after stroke based on XGboost algorithm has stronger recognition of SAP occurrence,and the prediction effect is better than that of Logistic regression model,and the XGboost model can identify risk factors that are easy to ignore better than the logistic regression model.The prediction model of SAP after stroke constructed by XGboost algorithm showed whether nasal feeding,dysphagia,neutrophil/lymphocyte ratio was increased,whether bed rest was long-term,and NIHSS score were the influencing factors of pneumonia after stroke,suggesting that these five points should be paid attention to in the clinical nursing process,and targeted interventions should be formulated based on such influencing factors.