Problem

Pre-Anesthesia Testing teams were spending hours per patient trying to understand risk before surgery.

Risk stratification sat at the intersection of procedure risk and patient risk, but the information needed to make that judgment was scattered across EHR charts, paper records, faxed requests, other clinics, and direct patient calls. Some teams created short clinic questionnaires to help triage whether a patient needed to be seen in person or over the phone, but those questions were often missed in practice.

Understanding

After speaking with several hospital systems, we learned that every team was trying to stratify patient risk in its own way. They did not want to be forced into a single validated instrument, and some instruments were considered too time consuming to score manually.

The strongest pattern was not a single scoring model. It was the underlying taxonomy. Teams were already reasoning through risk in ways that mapped closely to the structure used by CPT codes. That gave us a path toward a flexible product model: pull the same categories of clinical information repeatedly, but allow each system to surface and score that information in the way its clinicians expected.

Early Ideas

The earliest designs mirrored the manual workflows we were hearing about: questionnaires, form-based documentation, and clinician-entered risk details. Those screens helped make the workflow tangible, but they also exposed the scalability problem. If every customer needed a custom questionnaire, the product would become a series of one-off builds.

As discovery continued, the direction shifted from recording answers manually to using Health Information Exchange data to surface risk factors already present in patient records. The design challenge became less about asking every question and more about helping clinicians review, trust, and override what the system found.

Ideation

The Care Pathway Menu became the scalability solve. Instead of hard-coding each system's risk questionnaire, the product could let teams choose from a menu of comorbidities, configure how those findings should be scored, and preserve clinician discretion when a score needed to be overridden.

I iterated from manual questionnaire capture, to ICED-informed risk review using HIE records, to a configurable pathway model that could support different scales while repeatedly pulling the same categories of clinical evidence.

Final Design

The final experience uses the existing Case List to help teams triage patients first. From there, a clinician can open an assessment, review what the Health Information Exchange flagged as potential risk, and drill into each comorbidity with supporting documentation.

The assessment presents risk in the way the hospital system expects to score it, while still showing the evidence behind the generated recommendation. Clinicians can parse a summary, investigate details, and use their discretion when the automated score does not reflect the full clinical picture.

Learnings & Next Steps

The pilot reframed how we measured success. The goal was not to generate as many assessments as possible; it was to generate assessments clinicians intended to review and could trust enough to act on.

That principle continues to guide refinements to permissions, error handling, validation, and the balance between AI-generated findings and clinician override.