Joeky Senders

150 Chapter 8 performance was less consistent compared to the other models as denoted by the larger standard deviations. Lastly, ClinicalTextMiner was the only model that reached the AUC performance threshold of >0.99, with 600 training examples. TABLE 3. Incremental model performance according to the area under the receiver operating characteristic curve. Sample size Bootstrapped AUC Deep learning-based models Regression-based models ClinicalTextMiner CNN Lasso regression Logistic regression 25 0.794±0.052 0.751±0.085 0.698±0.106 0.507±0.029 50 0.901±0.038 0.835±0.062 0.850±0.070 0.583±0.109 75 0.943±0.023 0.885±0.052 0.904±0.033 0.715±0.096 100 0.958±0.016 0.904±0.038 0.929±0.022 0.785±0.078 150 0.977±0.008 0.925±0.028 0.949±0.014 0.890±0.034 200 0.983±0.005 0.936±0.024 0.959±0.010 0.918±0.019 250 0.986±0.003 0.943±0.024 0.965±0.009 0.929±0.014 300 0.987±0.002 0.945±0.020 0.968±0.008 0.944±0.011 400 0.988±0.002 0.954±0.015 0.973±0.006 0.950±0.009 500 0.989±0.002 0.956±0.015 0.977±0.004 0.961±0.009 600 0.990±0.001 0.956±0.014 0.981±0.003 0.969±0.006 800 0.991±0.001 0.962±0.012 0.983±0.003 0.974±0.005 1000 0.991±0.001 0.963±0.013 0.985±0.003 0.979±0.003 1200 0.991±0.001 0.963±0.013 0.986±0.002 0.981±0.002 1500 0.992±0.001 0.965±0.012 0.988±0.002 0.984±0.002 1800 0.992±0.001 0.965±0.013 0.989±0.002 0.985±0.002 2100 0.992±0.001 0.969±0.011 0.989±0.002 0.986±0.001 2500 0.992±0.001 0.964±0.013 0.990±0.001 0.988±0.001 3000 0.992±0.001 0.966±0.013 0.990±0.001 0.989±0.001 Abbreviations: AUC=area under the receiver operating characteristic curve; CNN=convolutional neural networks