Abstract
The 9 th revision of the International Classification of Diseases (as modified in the United States, also known as ICD-9-CM) is one of the most widely used classification system for clinical problems and procedures. Electronic health record systems usually require physicians to encode diagnoses or medical procedures through appropriate ICD-9-CM codes. Since ICD-9-CM is constituted by more than 13,000 diagnosis and 3,500 procedure codes, clearly the coding procedure is a non-trivial task. In this paper we present a coding support system for assisting physicians in the assignment of ICD-9-CM codes. The proposed system is based on tools for natural language processing and text-searches on knowledge bases. It analyzes a short medical text and then returns the list of ICD-9-CM codes that are most likely to be associated with the given text. A set of numerical experiments were carried out in order to assess the reliability and quality of the proposed support system. The results of the experiments show that the support system has an accuracy of 92%, largely improving physicians' average performances.