Identification and extraction of family history information from clinical reports

Many clinical reports contain family history, which is valuable information for clinical decision support and research. We developed a simple natural language processing algorithm to identify and extract family histories. The algorithm was tested on a set of discharge summaries and outpatient clinic notes. The precision and recall of extracting all diagnoses were 85.12% and 86.93%, respectively. The precision and recall of differentiating family history from patient history diagnoses were 96.30% and 92.86%, respectively. Both the precision and recall of exact family member assignment were 92.31%.