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

@k

MLP

MLP +

RNN

RNN +

RNNa

RNNa+

Dipole

Dipole +

RETAIN

RETAIN +

GRAM

GRAM +

MIMIC-III

5

0.6939

0.7124

0.6616

0.7160

0.6504

0.7083

0.6599

0.7074

0.6835

0.7167

0.6885

0.7132

 

10

0.6441

0.6603

0.6145

0.6565

0.6021

0.6527

0.6116

0.6539

0.6361

0.6623

0.6424

0.6596

 

15

0.6812

0.6926

0.6546

0.6906

0.6412

0.6856

0.6524

0.6903

0.6777

0.6918

0.6828

0.6918

 

20

0.7420

0.7544

0.7199

0.7511

0.7109

0.7455

0.7159

0.7483

0.7403

0.7501

0.7434

0.7513

 

25

0.7939

0.8070

0.7755

0.8019

0.7697

0.8009

0.7723

0.8020

0.7912

0.8010

0.7941

0.8028

 

30

0.8357

0.8460

0.8186

0.8456

0.8142

0.8445

0.8169

0.8453

0.8335

0.8445

0.8377

0.8468

Heart failure

5

0.4451

0.4947

0.4890

0.5172

0.4976

0.5103

0.4964

0.5111

0.3751

0.5140

0.5341

0.5365

 

10

0.6122

0.6206

0.6585

0.6879

0.6675

0.6817

0.6689

0.6829

0.5378

0.6828

0.7123

0.7159

 

15

0.6996

0.7060

0.7436

0.7683

0.7496

0.7631

0.7514

0.7648

0.6372

0.7613

0.7901

0.7939

 

20

0.7606

0.7643

0.8006

0.8213

0.8050

0.8174

0.8070

0.8167

0.7088

0.8143

0.8402

0.8442

 

25

0.8100

0.8140

0.8425

0.8593

0.8453

0.8560

0.8476

0.8557

0.7655

0.8533

0.8761

0.8789

 

30

0.8477

0.8511

0.8743

0.8879

0.8770

0.8857

0.8785

0.8846

0.8102

0.8826

0.9025

0.9047

  1. denotes the highest precision among all the approaches on the same k