Rule-based extraction of family history information from clinical notes

Abstract

One of the features of Electronic Health Records (EHR) is to store the patient clinical data. Despite the efforts to structure all this data, clinical reports and notes containing essential information about each patient are still stored in free text. Some of this information refers to the family's health history and may be highly relevance for diagnosis and prognosis. We proposed two methodologies to unify this knowledge and extract family history information from clinical notes using rule-based techniques in natural language processing (NLP). With these methods, we intend to collect the family members mentioned in the text as well as associations to diseases and living status. The proposed methods were evaluated into two stages, demonstrating F-scores of 0.72 and 0.74 for the discovery of family members and observations, and 0.62 and 0.52 for the detection of the family relations with the observations, and their living status. Our methodologies raised new strategies to automatically annotate large amounts of EHRs, facilitating the detection of comorbidities within family relations.