Automatic text analysis of medical records

Clinical narratives stored in a computerized medical record (PROMED) were automatically analyzed by a computer program (LOGSTORY). No medical knowledge was built into LOGSTORY prior to the analysis. The clinical sample consisted of 5,041 patients and 14,323 diagnoses. The present study concerns 375 diagnoses from general practice and 160 diagnoses from occupational medicine. LOGSTORY reproduced the symptoms, signs, laboratory investigations, anatomy and etiology of diabetes mellitus, obesity and lung diseases. Similarities and differences between the clinical states were automatically recognized and quantified. The extraction of knowledge from the clinical narratives required a problem-oriented medical record. PROMED-LOGSTORY may be useful for self-evaluation, peer review, quality control and research.