Prediction Of Weaning In Mechanically Ventilated Patients And Analysis Of Its Influencing Factors

The field of critical care medicine is developing with the progress of society.Mechanical ventilation is an important means of life support for clinical severe patients.After the patient’s underlying disease is treated and improved,it is necessary to slowly achieve spontaneous breathing and then be released from the ventilator.The traditional weaning method is mainly based on the doctor’s clinical experience to make weaning judgment,lacking theoretical basis and relevant quantitative indicators.At present,the improved planned weaning methods are mostly based on spontaneous breathing test and single indicator to determine weaning.Although weaning is more timely,the failure rate of weaning is still high,which will have adverse effects on patients,their families and the hospital.Therefore,how to help patients with mechanical ventilation achieve safe and timely weaning has become a major problem for critical care physicians in their work.In view of the above-mentioned problems,this thesis introduces the machine learning method into the weaning prediction research of mechanically ventilated patients,aiming to assist clinicians to improve the success rate of weaning.In this thesis,628 patients with mechanical ventilation who meet the basic conditions of weaning in ICU are selected as the research object,and the weaning prediction model is established to explore the important factors affecting the weaning of patients.In the data preprocessing part,it includes cleaning the messy original data,identifying and processing missing values,data binning,and feature encoding.In the feature selection part,three methods of univariate selection,recursive feature elimination,and LASSO are used to filter variables to improve the quality of the data.In the model building part,based on the three feature subsets,the weaning prediction model is constructed by using machine learning algorithms including Logistic Regression,SVM,Random Forest,Adaboost and Ada-LR and the accuracy,sensitivity,specificity and AUC are used as the metrics of the model for comprehensive evaluation.The results show that Ada-LR combination model based on recursive feature elimination feature subset has the best prediction performance,and accuracy,sensitivity,specificity,and AUC reached 0.81,0.83,0.81,and 0.90,respectively.At the same time,based on the literature method and the common factors of the three feature subsets,it is found that the APACHE II score,time of mechanical ventilation and age,etc.are important factors affecting weaning.The conclusion shows that the Ada-LR combination model based on recursive feature elimination feature subset has good predictive value,which can provide a theoretical basis for weaning prediction in clinical field and assist clinicians in weaning.