Hierarchical Prescription Pattern Analysis with Symptom Labels.

Identifying the prescription patterns would be a useful and interesting goal from multiple perspectives. Firstly, the identified patterns could expand the horizon of the medical practice knowledge. Secondly, the identified prescription patterns can be evaluated by subject-matter experts to label some of the patterns as anomaly calling for further investigation, i.e., prescription costs for insurance companies. Recently, the Health Insurance Review & Assessment Service (HIRA), South Korea, released a dataset on about six millions prescriptions on sampled population over three years. This paper presents the statistical modeling details of Tag Hierarchical Topic Models (Tag-HTM) and the application of Tag-HTM to the HIRA dataset. The application of Tag-HTM revealed a hierarchical structure of medicine-symptom distributions, which would be a new information to medical practitioners given that previous disease classification was mainly done by the anatomical and the disease cause aspects. Also, Tag-HTM was able to isolate the prescription patterns with higher medical costs as a branch of hierarchical clustering, and this cluster would be a prescription collection of interests to subject-matter experts in the insurance companies.