A Large Scale, Cross-disease Family Health History Data Set

Introduction: A family health history data set need to be evaluated before applying to the study of genetic diseases, genetic counseling, and epidemiological studies. We have obtained a large scale, cross-disease family health history data set (FhhDS) from electronic discharge summaries at Columbia Presbyterian Medical Center by using a pattern matching parser we have developed 1. Currently, FhhDS contains 22,292 patients' family health history. Here we have evaluated FhhDS by examining the scale, the coverage, and the completeness of the data set. Background: The Columbia Presbyterian Medical Center (CPMC) has the Columbia Clinical Repository that contains many electronic discharge summaries 2. All electronic discharge summaries are inpatient and obtained by physician dictation. We have developed and evaluated a pattern matching parser-Family Health History Extraction Processor (FhhEP) that extracts family health history from narrative electronic discharge summaries and builds a database-ready output 1. The output is the health history findings containing ten attributes: unique identification number of the discharge summary, family member, number of the family members, disease, trait, presence or absence of the disease, certainty, death condition, date of developing disease or death, and location of the disease. From the unique identification number of the discharge summary, we can retrieve the patient's MRN and discharge date. Five physicians have evaluated the natural language processing parser. The majority physician opinion is the reference standard. Sensitivity is the number of positive answers in common between the parser findings and the reference standard divided by the number of positive answers in the reference standard. Precision is the number of positive answers in common between the parser findings and the reference standard divided by the total number of parser findings. The sensitivities in the presence of all ten attributes, five attributes (family member, number of family member, disease, trait, and presence of disease), four attributes (family member, disease, trait, presence of disease), and three attributes (disease, trait, presence of disease) are 75.6%, 85.3%, 85.7%, and 93.0%, respectively. The precision in the presence of all the attributes, five attributes (family member, number of family member, disease, trait, and presence of disease), four attributes (family member, disease, trait, presence of disease), and three attributes (disease, trait, presence of disease) are 72.0%, 72.3%, 72.3%, and 79.5%, respectively 1. Methods: We obtained the FhhDS by running the FhhEP through all the electronic discharge summaries at CPMC, one year at a time, from 1992 to 1998. We measured the …