Machine Learning in Clinical Text Classification: Specialty Identification and COVID-19 Risk

A report from the World Health Organization reveals that many people lack access to good healthcare services. Primary health care is often inaccessible, not only in developing countries, but also in developed nations like the United States. The lack of sufficient primary care physicians is one of the chief factors contributing to healthcare inaccessibility. Prior research has attempted to address the issue by examining patient symptoms and transcripts through the use of machine learning algorithms, but because numerous illnesses can produce identical symptoms, these efforts have struggled to correctly diagnose and guide patients. We sought to increase the access to healthcare services by utilizing a machine learning system to guide a patient to the appropriate specialist based on the symptoms indicated in their transcripts. In this study, we developed and evaluated an algorithm-based solution that would give the public credible, data-driven, and personalized information about their symptoms, enabling patients and their doctors to make better-educated decisions based on statistics and text transcripts. To do so, we built three models: (1) a transcript model, which uses clinical transcripts to predict the appropriate medical specialist; (2) a keyword model, which uses keyword extraction to reduce noise and isolate the symptoms from the clinical transcripts, and then uses these keywords to predict the appropriate medical specialist; and (3) a COVID-19 risk detection model, which predicts the COVID-19 risk of a patient, something that has not been fully investigated in this field of research.