Table 2 ICU phenotypes used in the benchmark data set along with their prevalence and the per-phenotype classification performance of the best LSTM network.

From: Multitask learning and benchmarking with clinical time series data

Phenotype Type Prevalence AUC-ROC
Train Test
Acute and unspecified renal failure acute 0.214 0.212 0.806
Acute cerebrovascular disease acute 0.075 0.066 0.909
Acute myocardial infarction acute 0.103 0.108 0.776
Cardiac dysrhythmias mixed 0.321 0.323 0.687
Chronic kidney disease chronic 0.134 0.132 0.771
Chronic obstructive pulmonary disease chronic 0.131 0.126 0.695
Complications of surgical/medical care acute 0.207 0.213 0.724
Conduction disorders mixed 0.072 0.071 0.737
Congestive heart failure; nonhypertensive mixed 0.268 0.268 0.763
Coronary atherosclerosis and related chronic 0.322 0.331 0.797
Diabetes mellitus with complications mixed 0.095 0.094 0.872
Diabetes mellitus without complication chronic 0.193 0.192 0.797
Disorders of lipid metabolism chronic 0.291 0.289 0.728
Essential hypertension chronic 0.419 0.423 0.683
Fluid and electrolyte disorders acute 0.269 0.265 0.739
Gastrointestinal hemorrhage acute 0.072 0.079 0.751
Hypertension with complications chronic 0.133 0.130 0.750
Other liver diseases mixed 0.089 0.089 0.778
Other lower respiratory disease acute 0.051 0.057 0.694
Other upper respiratory disease acute 0.040 0.043 0.785
Pleurisy; pneumothorax; pulmonary collapse acute 0.087 0.091 0.709
Pneumonia acute 0.139 0.135 0.809
Respiratory failure; insufficiency; arrest acute 0.181 0.177 0.907
Septicemia (except in labor) acute 0.143 0.139 0.854
Shock acute 0.078 0.082 0.892
All acute diseases (macro-averaged)     0.796
All mixed (macro-averaged)     0.768
All chronic diseases (macro-averaged)     0.746
All diseases (macro-averaged)     0.776