Improving revenue cycle management with deep learning: A healthcare case study

Description

Deep learning is attracting attention in healthcare in a variety of areas. A lot of work is underway in training deep learning models to support clinical applications, such as the prediction and diagnosis of disease. While promising in these areas, deep learning also has inherent risks in interpretability of results and quality of data that need to be addressed before it’s more widely used in the clinical environment. On the other hand, there is near-term opportunity to use deep learning to streamline operational processes without running the risk of applications in diagnostic and decision support tools.

OptumLabs recently trained and deployed a deep learning model that’s making an impact in prioritizing work and cases for manual review to achieve operational efficiencies. A significant number of these cases are challenging for provider organizations to evaluate against medical necessity criteria for reimbursement because they are not clear-cut. They often result in denials, rework, or inappropriate reimbursement of insurance claims. Often a physician advisor is engaged to do a second-level review when cases don’t meet initial medical necessity criteria or if the impact of regulatory or payer guidelines needs to be considered. There’s a lot of labeled data available on the inputs and outputs of decisions made in this process, creating fertile ground for a deep learning model to help drive a more efficient and compliant reimbursement process. With this deep learning model, all admissions can now be screened automatically to identify the ones that need a second-level review to support the medical necessity review process. The model helps drive a more efficient and compliant reimbursement process by engaging physician advisors to review only the challenging cases where expertise is most needed.

Sanji Fernando explores his experience from the end-to-end process of building, deploying, and operating a deep learning model that improves hospital revenue cycle management, including use case understanding, performance metric selection, data preparation, tool selection and trade-offs, model development and optimization, model selection, deployment, and model operations. Sanji also details key knowledge gained along the way on what worked well and opportunities for improvement in training future deep learning models in healthcare.