Prediction of Unplanned Hospitalization of Medicare Patients

CMS AI Challenge

This project developed machine learning models to predict patient health outcomes (unplanned hospitalizations and hospital readmissions) for Medicare beneficiaries. We predicted eight types of common diseases and four type of adverse events by analyzing over 15M+ Medicare claims and various public datasets.

We developing a data-driven and knowledge-informed algorithm that divides the population into similar risk groups based on patient demographics, medical history, care quality, and environmental factors we can build more personalized, effective prediction models for each group. We also included a mixed-integer programming-based scoring algorithm to generate monthly hospitalization risk for individual patients.

alt text

This project was led by Dr. Sara Shashaani and Dr. Julie Swann fron North Carolina State University. The team was selected top 25 out of 300+ teams into Stage I of the 2020 national $1M CMS AI Health Outcomes Challenge.

Publications:

  • Mao L, Vahdat K, Shashaani S, Swann J. Personalized predictions for unplanned urinary rract infection hospitalizations with hierarchical clustering. AI and Analytics for Public Health: Proceedings of the 2020 INFORMS International Conference on Service Science (pp. 453-465) (pp. 453-465). https://doi.org/10.1007/978-3-030-75166-1 34 (conference paper) (🏆 Best Student Paper Finalist, INFORMS ICSS 2020) (🏆 CMS AI Challenge)