Abstract for:Supporting Locally Calibrated Modeling at Scale with Front-line Clinical Teams Using a Transparent Data User Interface

Simulation and modeling are powerful tools to improve decision-making, but the models and input data must exist at the level of the decisions. At the Veterans Health Administration, high level data reports guide policymakers, and patient-level data guide individual treatment decisions, but there is no data at the level of the clinical care team, which is where most day-to-day system decisions are made. In fact, the care teams themselves are undefined in the data system! This presentation will cover how we built an innovative information system to address the needs of a modeling project at the care team level. The data system uses Excel as an accessible and familiar platform to build an intuitive user interface. Behind the interface, dynamic SQL queries are generated based on user input. The hyper-local but far-reaching results of those queries are leveraged for validation and parameterization of the models in a user-transparent way in order to inspire confidence in the models. We will cover how we worked with pilot teams to build a tool that would combat user mistrust, how we built the novel data pipeline, and how we solved some of the tricky data issues in this complex service system.