Early Prediction of Organ Failures in Patients with Acute Pancreatitis Using Text Mining
Table 2
The performance of early predicting organ failures in AP using our proposed model.
Model
Performance
Respiratory failure
Circulatory failure
Renal failure
Model 1
Training
Accuracy
0.727 (0.033)
0.668 (0.034)
0.786 (0.048)
Recall
0.762 (0.023)
0.680 (0.036)
0.758 (0.033)
Precision
0.446 (0.039)
0.374 (0.040)
0.141 (0.039)
Specificity
0.776 (0.044)
0.727 (0.026)
0.832 (0.015)
Testing
Accuracy
0.646 (0.027)
0.684 (0.022)
0.829 (0.046)
Recall
0.649 (0.034)
0.664 (0.023)
0.681 (0.036)
Precision
0.338 (0.013)
0.374 (0.013)
0.169 (0.040)
Specificity
0.692 (0.047)
0.718 (0.027)
0.842 (0.020)
Model 2
Training
Accuracy
0.744 (0.039)
0.820 (0.013)
0.871 (0.007)
Recall
0.866 (0.011)
0.769 (0.035)
0.905 (0.036)
Precision
0.482 (0.020)
0.584 (0.021)
0.264 (0.031)
Specificity
0.789 (0.025)
0.844 (0.034)
0.892 (0.036)
Testing
Accuracy
0.659 (0.030)
0.752 (0.033)
0.823 (0.042)
Recall
0.619 (0.025)
0.723 (0.020)
0.519 (0.022)
Precision
0.342 (0.048)
0.457 (0.008)
0.136 (0.029)
Specificity
0.683 (0.022)
0.783 (0.050)
0.746 (0.020)
Model 3
Training
Accuracy
0.659 (0.008)
0.820 (0.029)
0.803 (0.027)
Recall
0.850 (0.022)
0.784 (0.035)
0.894 (0.013)
Precision
0.402 (0.014)
0.582 (0.034)
0.194 (0.019)
Specificity
0.788 (0.045)
0.844 (0.015)
0.892 (0.021)
Testing
Accuracy
0.687 (0.019)
0.771 (0.035)
0.789 (0.024)
Recall
0.664 (0.014)
0.732 (0.031)
0.708 (0.017)
Precision
0.377 (0.012)
0.485 (0.010)
0.143 (0.043)
Specificity
0.727 (0.021)
0.808 (0.046)
0.831 (0.035)
Model 4
Training
Accuracy
0.772 (0.047)
0.784 (0.036)
0.838 (0.032)
Recall
0.738 (0.044)
0.792 (0.044)
0.825 (0.024)
Precision
0.371 (0.017)
0.480 (0.016)
0.214 (0.011)
Specificity
0.814 (0.043)
0.831 (0.018)
0.873 (0.019)
Testing
Accuracy
0.689 (0.030)
0.687 (0.011)
0.655 (0.044)
Recall
0.543 (0.010)
0.688 (0.016)
0.854 (0.026)
Precision
0.228 (0.033)
0.374 (0.026)
0.105 (0.041)
Specificity
0.649 (0.038)
0.720 (0.014)
0.768 (0.023)
Model 1: input of structured features only. Model 2: input of text features only. Model 3: input of mixed features combined with attention mechanism. Model 4: input of mixed features and use of attention mechanism with multitask. Standard deviation is in the bracket.