Using Natural Language Processing to Classify Free-text Laboratory Results into Standardized Diagnostic Categories

The use of natural language processing (NLP) for classification of free-text lab results into standardized diagnostic categories to organize unstructured clinical data is studied here. A collection of 6,800 lab reports was employed to train and test transformer-based models like BERT and BioBERT on ten diagnostic categories like infections, thyroid disease, and diabetes. The highest performing model yielded an average confidence score of 0.920. Moderately severe cases constituted 39.96% of the data, while high severity cases comprised 20.32%. The most frequent classes were liver dysfunction and diabetes, with the lowest being vitamin deficiency. The analysis proved to have a balanced distribution between classes. These findings underscore the competence of transformer-based NLP models in clinical classification and imply that these models, particularly when paired with domain-specific knowledge graphs and ontologies, have the potential to boost diagnostic precision and facilitate efficient clinical decision-making.