Fig. 4
From: Ontology-driven and weakly supervised rare disease identification from clinical notes
Number of rare disease patient stays from MIMIC-III (n=59,652): ICD (code-based) vs. NLP (text-based, with weak supervision), for 10 selected diseases. Admissions are split into those only identified through links from ICD-9 codes (in black), those only identified from free texts with weak supervision (NLP, in white), and the intersection of cases from both ICD-9 and NLP (in grey). The percentage after each horizontal bar shows the accuracy of NLP based on the manual assessment of the identified cases