Skip to main content

Table 3 The visit-level precision @k of diagnosis prediction task

From: Incorporating medical code descriptions for diagnosis prediction in healthcare

Dataset@kMLPMLP +RNNRNN +RNNaRNNa+DipoleDipole +RETAINRETAIN +GRAMGRAM +
MIMIC-III50.69390.71240.66160.71600.65040.70830.65990.70740.68350.7167 0.68850.7132
 100.64410.66030.61450.65650.60210.65270.61160.65390.63610.6623 0.64240.6596
 150.68120.6926 0.65460.69060.64120.68560.65240.69030.67770.69180.68280.6918
 200.74200.7544 0.71990.75110.71090.74550.71590.74830.74030.75010.74340.7513
 250.79390.8070 0.77550.80190.76970.80090.77230.80200.79120.80100.79410.8028
 300.83570.84600.81860.84560.81420.84450.81690.84530.83350.84450.83770.8468
Heart failure50.44510.49470.48900.51720.49760.51030.49640.51110.37510.51400.53410.5365
 100.61220.62060.65850.68790.66750.68170.66890.68290.53780.68280.71230.7159
 150.69960.70600.74360.76830.74960.76310.75140.76480.63720.76130.79010.7939
 200.76060.76430.80060.82130.80500.81740.80700.81670.70880.81430.84020.8442
 250.81000.81400.84250.85930.84530.85600.84760.85570.76550.85330.87610.8789
 300.84770.85110.87430.88790.87700.88570.87850.88460.81020.88260.90250.9047
  1. denotes the highest precision among all the approaches on the same k