Towards Temporal Segmentation of Patient History in Discharge Letters
Galia Angelova, Svetla Boytcheva · 2011
This paper reports about ongoing work in automatic identification of temporal markers and segmentation of patient histories into episodes. We discuss the discourse structure of the Anamneses in Bulgarian hospital discharge letters and present experiments with a corpus of 1,375 anonymised discharge letters of patients with endocrine and metabolic diseases. Our IE prototype discovers 32,445 key terms in the corpus, among them more that 7,000 occurrences of drug names and about 7,500 occurrences of diagnoses. The temporal markers occur 8,248 times usually paired with tokens pointing the direction of time “forward” or “backwards”. Temporal markers are identified with precision 84%, recall 57% and f-measure 67.9%.