Automatic allergy classification based on Russian unstructured medical texts

Most of the medical data in hospital information systems databases are stored in an unstructured form. Techniques for processing unstructured records are widely presented in scientific papers focused on English data. This paper proposes a method for intellectual analysis of unstructured allergy anamnesis in Russian in order to identify the presence and type of allergy and intolerance of a patient. The method is based on machine learning algorithms and uses international standards for the exchange of medical data and terminology standards, such as FHIR and SNOMED CT. As a result of the experiment, about 12 thousand medical records were processed. F-measure for the developed classification models ranged from 0.93 to 0.96. The models showed high values of metrics for evaluating the effectiveness of the models. In the future, structured data can be used in models for predicting medical risks. Further development of methods for structuring medical texts will ensure the interoperability of medical data.