Dear Colleagues,
Our team at NYU is developing a methodology for evaluation of the effectiveness of AI-based algorithms to predict patient outcomes both across clinical sites and longitudinally. The outcomes of this study will allow us to provide evidence for or against the ability of the tested algorithms to generalize beyond the site where they were trained. If a method does not generalize robustly, we will determine if retraining within sites reinstates its original training performance. We will also study the possible temporal degradation of the algorithms performance and explore if the performance degrades at different rates at different sites. Our three initial models include prediction of sepsis, in-hospital mortality, and medication adherence.
To implement and test these performance evaluation methods, we seek partners that have data standardized to PCORnet Common Data Model (CDM) and OMOP CDM to demonstrate generalizability and scalability. This will be a funded effort.
Please reach out to rimma.belenkaya@nyulangone.org if interested.
Thank you.
