Dr. Narges Razavian is an assistant professor in the Departments of Population Health and Radiology conducting research in the Center for Healthcare Innovation and Delivery Science (CHIDS), and a member of its Predictive Analytics Unit. In this episode we dig deeper into her team’s recent work with using graph neural networks to represent EHR data. Dr. Razavian’s team was able to validate their model predictions by predicting onset of dementia 2 years in advance. We talked about data gaps within an EHR and challenges specially for ICU patients due to lack of data capture. Dr. Razavian spoke about her experience fighting on the frontlines of the COVID pandemic with data science, especially as New York was one of the first cities to be affected by it.
She spoke about how the questions they tackled changed with the course of the pandemic from: how to determine who might be having covid during the early days to how to free up capacity as the pandemic intensity increased. Dr. Razavian also shared her practical thoughts on how data scientists and clinicians can become better collaborators. She is now very excited about scaling machine learning to be deployable at the point-of-care, a skill her team mastered during COVID.