First Approaches on Spanish Medical Record Classification Using Diagnostic Term to Class Transduction

This paper presents an application of finitestate transducers to the domain of medicine. The objective is to assign disease codes to each Diagnostic Term in the medical records generated by the Basque Health Hospital System. As a starting point, a set of manually coded medical records were collected in order to code new medical records on the basis of this set of positive samples. Since the texts are written in natural language by doctors, the same Diagnostic Term might show alternative forms. Hence, trying to code a new medical record by exact matching the samples in the set is not always feasible due to sparsity of data. In an attempt to increase the coverage of the data, our work centered on applying a set of finite-state transducers that helped the matching process between the positive samples and a set of new entries. That is, these transducers allowed not only exact matching but also approximate matching. While there are related works in languages such as English, this work presents the first results on automatic assignment of disease codes to medical records written in Spanish.