AutoDX: Optimizing an Automated Documentation Tool to Fit Provider Workflow

All providers are required to document their time with patients. But this documentation is labor-intensive and complicated. It also impacts the time providers spend with patients and significantly contributes to physician burnout. Physician documentation is important because it affects how hospitals are reimbursed by Medicare and insurance providers.

As part of a quality improvement effort, a team at UChicago Medicine (UCM) developed a real-time automated documentation assistance tool, called AutoDx, that operates inside their Epic electronic health record system. AutoDX collects patient conditions and risk factors from their health records and summarizes them in the physician’s note. This auto-population ensures that providers include all relevant details in their notes — freeing providers from needing to remember everything that should be charted. Yet despite early signs that this tool improves note quality and saves providers time, adoption levels were lower than expected: 42% of users reported deleting AutoDx suggestions from their notes.

This project sought to explore provider adoption issues and help UCM reach their 80% usage goal. Our output was to generate new UX/UI guidelines that fit provider needs, promote uptake, and could guide the expansion of AutoDX to include more discharge codes and note templates.

We implemented a 14-week, multi-phase design process. We conducted contextual inquiry and usability testing with 15 providers of varying expertise levels in UCM’s Internal Medicine department. This identified critical gaps in the current AutoDx workflow that created barriers to efficient clinical documentation. Throughout this process, we worked with an information technologist to understand the system constraints of Epic and its integration requirements, allowing us to prototype feasible and testable interfaces. We then synthesized our findings into actionable design requirements addressing both technical limitations and user needs.

As part of the development phase, we conducted three cycles of rapid prototyping and testing, using a custom-developed assessment tool to evaluate prototypes. Each prototype was tested with 3-5 clinicians per cycle, ensuring the tool was intuitive and aligned with clinical workflows. Following prototype refinement, we assembled an implementation toolkit with high-fidelity prototypes alongside detailed design specifications for UCM’s development team.