ClinicalNLP/Clinical_prediction/CATNet

• CATNet
• • Requirements
  • Data
  • Main Entrance
  • • Script
    • • Training
      • Test
• Test results
• • For medication prediction:
  • For diagnosis prediction:
  • For labtest prediction
  • For procdure prediction

CATNet

This repository provides the code for "CATNet: Cross-event Attention-based Time-aware Network for Medical Event Prediction".

Requirements

pytorch == 1.4.0
sklearn == 0.19.1

Details please find in 'requirment.txt'

Data

The project contains prepared data of MIMIC_III dataset and eICU dataset in data folder.

Main Entrance

train_eval.py
contains training and evaluation codes.

Script

Training

python train_eval.py [--predDiag/--predProc/--predLabtest]

Default setting is 'predMedic'.

Test

python train_eval.py --test_mode 1 [--predDiag/--predProc/--predLabtest] --best_parameters_file "the saved model path"

Test results

For medication prediction:

saved_path:"CATNet_MIMIC_result/tran_Transformer_AllCodeLevelattentionFusion_Transmain_selfatt_gated_mimic_data_L1_lr0.000100_dp0.200000_vs256_hs256_focal2.00"

tran_Transformer_AllCodeLevelattentionFusion_Transmain_selfatt_gated_mimic_data_L1_lr0.000100_dp0.200000_vs256_hs256_focal2.00
precision10:0.114467+0.000619
precision20:0.195408+0.000875
precision30:0.258051+0.001097
precision40:0.310316+0.001141
precision50:0.352882+0.000875
f1_10:0.148607+0.000923
f1_20:0.249582+0.001154
f1_30:0.326447+0.001434
f1_40:0.390639+0.001489
f1_50:0.442654+0.001117
recall10:0.211769+0.001691
reacall20:0.345318+0.001788
recall30:0.444174+0.002121
racall40:0.527067+0.002211
racall50:0.593686+0.001686
auc:0.863677+0.000693	
aupr:0.410611+0.001777

[0.00061919 0.00087473 0.00109741 0.0011408  0.00087454 0.00169134
0.00178817 0.00212061 0.00221059 0.00168617 0.00092271 0.00115434
0.00143359 0.00148909 0.00111718 0.00069308 0.00177707]

For diagnosis prediction:

saved_path:"CATNet_MIMIC_result/tran_Transformer_AllCodeLevelattentionFusion_Transmain_selfatt_gated_mimic_data_L1_lr0.000100_dp0.200000_vs256_hs256_focal2.00"

tran_Transformer_AllCodeLevelattentionFusion_Transmain_selfatt_gated_mimic_data_L1_lr0.000100_dp0.200000_vs256_hs256_focal2.00
precision10:0.096805+0.000534
precision20:0.139051+0.000734
precision30:0.164618+0.000503
precision40:0.183265+0.000557
precision50:0.197467+0.000508
f1_10:0.154968+0.000831
f1_20:0.220880+0.001150
f1_30:0.260609+0.000789
f1_40:0.289587+0.000902
f1_50:0.311674+0.000852
recall10:0.388228+0.002101
reacall20:0.536745+0.002808
recall30:0.625143+0.001978
racall40:0.689749+0.002622
racall50:0.739193+0.002828
auc:0.952150+0.000144	aupr:0.418250+0.001938

[0.00053441 0.00073434 0.00050334 0.00055655 0.00050795 0.00210088
0.00280844 0.00197754 0.00262237 0.00282807 0.00083067 0.00114963
0.00078873 0.00090183 0.00085226 0.00014443 0.00193821]

For labtest prediction

saved_path:"CATNet_MIMIC_result/tran_Transformer_AllCodeLevelattentionFusion_Transmain_selfatt_gated_mimic_data_L1_lr0.000100_dp0.200000_vs256_hs256_focal2.00"

tran_Transformer_AllCodeLevelattentionFusion_Transmain_selfatt_gated_mimic_data_L1_lr0.000100_dp0.200000_vs256_hs256_focal2.00
precision10:0.198717+0.000042
precision20:0.395544+0.000067
precision30:0.567114+0.000585
precision40:0.715342+0.001114
precision50:0.845559+0.001528
f1_10:0.186293+0.000040
f1_20:0.368800+0.000089
f1_30:0.519662+0.000651
f1_40:0.644931+0.001015
f1_50:0.753113+0.001291
recall10:0.175331+0.000039
reacall20:0.345443+0.000108
recall30:0.479539+0.000714
racall40:0.587140+0.001126
racall50:0.678890+0.001160
auc:0.973918+0.000215	aupr:0.857782+0.001112

[4.17566790e-05 6.73908190e-05 5.85298274e-04 1.11416471e-03
1.52773168e-03 3.86635917e-05 1.08023969e-04 7.13885396e-04
1.12570396e-03 1.15988271e-03 4.01739318e-05 8.88409114e-05
6.51455304e-04 1.01540774e-03 1.29143400e-03 2.14930835e-04
1.11191166e-03]

For procdure prediction

saved_path="CATNet_MIMIC_result/tran_Transformer_AllCodeLevelattentionFusion_Transmain_selfatt_gated_mimic_data_L1_lr0.000100_dp0.200000_vs256_hs256_focal2.00"

tran_Transformer_AllCodeLevelattentionFusion_Transmain_selfatt_gated_mimic_data_L1_lr0.000100_dp0.200000_vs256_hs256_focal2.00
precision10:0.043221+0.000236
precision20:0.053849+0.000340
precision30:0.060217+0.000343
precision40:0.063566+0.000338
precision50:0.065849+0.000248
f1_10:0.080424+0.000442
f1_20:0.100041+0.000643
f1_30:0.111790+0.000639
f1_40:0.117971+0.000617
f1_50:0.122217+0.000470
recall10:0.577742+0.005386
reacall20:0.703553+0.006233
recall30:0.778776+0.005240
racall40:0.818530+0.003752
racall50:0.848779+0.004625
auc:0.948465+0.000816	aupr:0.299418+0.001905

[0.0002361  0.00034017 0.0003431  0.00033816 0.00024796 0.0053862
 0.00623337 0.00523982 0.00375247 0.00462533 0.00044201 0.00064269
 0.00063925 0.00061666 0.00047033 0.00081647 0.00190467]