Uncovering a Standard of Care using Frequent Itemsets
In this project, I wanted to identify frequently-seen sequences of procedures for patients with certain conditions. This could help to identify established standards of care, starting directly from the data. Such information would be useful to identify suspect providers who deviate from these standards.
View Project on GitHub »
Using Topic Models to Classify Provider Specialties
When investigating providers, I had access to their self-reported specialty, but the given values are not always reliable. For instance, a common problem was that primary care is reported when in fact the physician's practices are more specialized. Specialty is a key piece of information because it helps to set our expectation of how they should practice. Therefore, we might want to determine specialty based on how the provider practices (what procedures they bill, benes they see) rather than the reported value. One angle could be that providers who do not practice like their reported specialty is suspicious; at the very least, they should update their Medicare application. I investigated here the use of a topic model to generate meaningful specialties (topics).
View Project on GitHub »