Skip to main content

Table 5 Results on rare disease identification (Text-to-ORDO) from MIMIC-III discharge summaries

From: Ontology-driven and weakly supervised rare disease identification from clinical notes

 

validation (n=64+/400)

test (n=82+/673)

Text to ORDO

P

R

\(F_{1}\)

P

R

\(F_{1}\)

SemEHR [15]

18.7

95.3

31.3

13.9

92.7

24.1

+ rules

53.9

75.0

62.7

49.0

86.6

62.6

+ WS (rules+BERT)

67.6

75.0

71.1

64.7

80.5

71.7

+ SS (anns+BERT)

-

-

-

73.3

80.5

76.7

  1. The column statistics (n=\(N_+\)+/N) shows number of positive data \(N_+\) and all samples N in the dataset. WS, weak supervision; SS, strong supervision; anns, annotations. BlueBERT-base (PubMed+MIMIC-III) was used as the BERT model. The best scores, either or not considering strong supervision (SS), are bolded