Skip to main content

Table 3 Cross-validation and ROC curve analysis of the two predictive models

From: Study of multiparameter respiratory pattern complexity in surgical critically ill patients during weaning trials

Models AUC (SE) Threshold AUC Specificity Sensitivity
    95% CI   
Model 1 0.831 (0.14) 0.402 0.73-0.92 0.895 0.858
(RSBI, P0.1 and RSBI* P0.1)      
Model 2 0.916 (0.006) 0.296 0.76-0.98 0.967 0.886
(RSBI, RSBI* P0.1, SampEn, LLE)      
  1. Mean values of areas under the curve (AUC) with standard errors (SE), 95% confidence intervals (CI) and best threshold among regression coefficients that managed to classify groups with different weaning outcome with the best combined sensitivity and specificity, were computed in Matlab. Three indices (RSBI, P0.1 and RSBI* P0.1) were included in the model 1 and four indices (RSBI, RSBI* P0.1, SampEn and largest lyapunov exponents of mean inspiratory flow time series) were selected in model 2 respectively, which was proven to discriminate more accurately patients with different weaning outcome.