Table 3 Characteristics of the human-computer test set

From: A phenotype-based AI pipeline outperforms human experts in differentially diagnosing rare diseases using EHRs

 

Rare diseasesa

 Number of cases

75

 Number of departments

5

 Number of diseases

16

 Age, years: median (average)

29 (31.6)

 Female (%)

36 (48%)

Department: pediatrics

 

 Number of diseases

3

 Number of cases

15

 Age, years: median (average)

7.42 (7.40)

 Number of physicians

10

Department: neurology

 

 Number of diseases

3

 Number of cases

15

 Age, years: median (average)

51 (49.1)

 Number of physicians

10

Department: renal

 

 Number of diseases

3

 Number of cases

15

 Age, years: median (average)

24 (24.4)

 Number of physicians

10

Department: cardiology

 

 Number of diseases

4

 Number of cases

15

 Age, years: median (average)

40 (39.1)

 Number of physicians

10

Department: hematology

 

 Number of diseases

3

 Number of cases

15

 Age, years: median (average)

46 (38)

 Number of physicians

10

Physicians

 

 Total Number of physicians

50

 Experience, years: median (average)

11 (12.9)

 Number of diagnoses per case: median (average)

3 (2.67)

  1. aThe 16 rare diseases were: Prader-Willi syndrome (PWS), Hepatolenticular degeneration (HD), McCune-Albright syndrome (MAS), Multiple system atrophy (MSA), Amyotrophic lateral sclerosis (ALS), Generalized myasthenia gravis (GMG), Alport syndrome (AS), Fabry disease (FD), Gitelman syndrome (GS), Marfan syndrome (MFS), Arrhythmogenic right ventricular dysplasia/cardiomyopathy (ARVD/C), Brugada syndrome (BS), Restrictive cardiomyopathy (RCM), POEMS syndrome (POS), Paroxysmal nocturnal hemoglobinuria (PNH), and Niemann-Pick disease (NPD). More details of rare diseases are provided in Supplementary Table 4.