Research on data-driven predictive model for extubation in intensive care unit
Objective To develop and evaluate a data-driven prediction model for extubation in intensive care unit(ICU).Methods Data was collected from ICU patients undergoing invasive mechanical ventilation.The data included non-sequential data,real-time monitored low-frequency vital signs time-series data,and high-frequency mechanical ventilation time-series data.Prediction models were constructed using algorithms such as decision trees,naive Bayes,support vector machines,ensemble classification,generalized linear models,and neural networks.The performance of these models was then evaluated.Results A total of 204 patients were included,in which 122 patients were successfully extubated and 82 failed extubation.High-frequency mechanical ventilation time-series data was available for 163 patients,which was used to construct a deep neural network model.This model demonstrated the best performance in predicting extubation success.The accuracy was 94.2％.The area under ROC curve(AUC)was 0.81.The sensitivity was 100％,and the specificity was 80％.Other models showed comparatively inferior performance.The accuracy of the classification ensemble model was 75％,the AUC value was 0.76,the sensitivity was 83.3％,and the specificity was 62.5％.Conclusions This study reveals that the deep neural network model,built with high-frequency time-series data,exhibits outstanding performance in predicting extubation success in ICU patients.It significantly surpasses other models without such data.
NETL
NSTL
万方数据
