An innovative healthcare software product company with a solution for payer reimbursements wanted to improve the way its offering helped healthcare providers determine patient eligibility for procedures. Typically, healthcare providers look at patient procedure criteria (for example, an MRI) on a case-by-case basis. That criteria is then fed into a system to compare against the patient's records. This determines the patient's eligibility for a given procedure and eligibility for reimbursement.
Payer reimbursements for patients varies wildly from case to case requiring a practically unlimited number of rules to be coded and maintained for automation to succeed. The time consuming part for healthcare providers involved having someone read the criteria doc and then convert the criteria into a format that our client’s system could read. The product development team and CTO saw a unique opportunity to transform this process.
The product team believed a “learning approach” would provide the extensibility and scalability necessary to productize the process of determining patient procedure eligibility. However, building this unique capability into their software required access to techniques beyond the skill set of their core software engineering team.
We worked alongside the product team to automate key aspects of the process of determining patient/procedure eligibility using a variety of complementary approaches. We employed some of the following techniques:
Machine Learning (ML) for identifying text fragments of high relevance,
Natural Language Processing (NLP) for parsing, and
Domain-specific Information Extraction (IE) heuristics.
Our machine learning approach leveraged heuristics to extract relevant, easily understandable keywords to input into our client’s solution. The NLP parsed the input documents for criteria, essentially replacing the role of the human reading eligibility requirements. This automated the process of determining patient procedure eligibility by extracting relevant data and presenting a ready-made made output that could then be entered into our client’s tool. No more reading criteria documents!
This project focused on building an “Alpha” version of a product that could automate some of the steps required to take patient information as input and determine whether a procedure was eligible for insurance coverage. We offered guidance on how to productize this solution to extend it to our client’s healthcare provider customers.
Our client has noted that in several cases, our product is able to identify eligibility criteria that manual processing had overlooked. In some other cases, the output of our product was more consistent and comprehensive than human-generated output. The product has successfully processed data originating from a variety of payors and pertaining to a variety of medical conditions. While we originally set out to build an “Alpha” product, our client started showing our solution to its customers towards the end of the project with rave reviews.