CMS AI Challenge Submission:
The Intelligent Risk Project
We extensively explored and tested machine learning models in our development of a preoperative risk prediction model that provides an optimal combination of robust prediction performance and clinically-meaningful interpretability.
Details on risk context and contributors appear by clicking on specific model aspects (blue boxes). The user may expand any of the four quadrants to a single screen.
The clinician will find patient by name or Medicare ID, view/modify/add preexisting comorbidities, and enter the planned procedure (or identify as medical admission).
The clinician will be able to view a timeline of all previous healthcare interactions, along with procedures performed and/or comorbidities treated during that interaction by clicking on each indicator.
Risk is presented as both an absolute risk (e.g. 30.9%) and relative to all patients undergoing that procedure (e.g. 97th percentile or 5.6 times the average mortality risk).
The factors contributing to risk (e.g. prior comorbidities/procedures) are represented by colored boxes, sized and scaled to reflect the relative impact of each factor on the overall risk.