Skip to main content

Table 4 The code-level accuracy @k of diagnosis prediction task

From: Incorporating medical code descriptions for diagnosis prediction in healthcare

Dataset@kMLPMLP +RNNRNN +RNNaRNNa+DipoleDipole +RETAINRETAIN +GRAMGRAM +
MIMIC-III50.31040.31810.29520.31930.29100.31620.29410.31550.30560.3198 0.30720.3183
 100.50400.51380.47960.51110.46930.50850.47670.50860.49800.5160 0.50030.5138
 150.62860.63520.60190.63350.58890.62900.59710.63250.62580.6360 0.62670.6348
 200.71140.7239 0.68940.71980.68220.71440.68450.71680.71290.72020.71300.7196
 250.77540.7852 0.75450.78040.74910.77850.75010.77950.77350.78060.77280.7794
 300.82140.8294 0.80400.82790.79870.82690.79900.82800.81980.82860.82200.8283
Heart failure50.45800.51320.55990.59600.56990.58820.56870.58680.40850.58080.61520.6227
 100.62660.64120.68350.71690.69200.71090.69530.71050.54600.70420.73930.7455
 150.71240.72540.76030.78760.76450.78450.77020.78410.65120.77650.80880.8130
 200.77170.78270.81320.83550.81530.83340.82090.83070.71620.82610.85440.8580
 250.82060.82830.85160.86980.85320.86730.85800.86550.76840.86220.88720.8902
 300.85720.86350.88120.89580.88250.89430.88600.89230.81000.88990.91130.9134
  1. denotes the highest accuracy among all the approaches on the same k