Skip to content

Projects

Pick one to read what it is, what was hard about it, and the decisions it produced. Filter by the kind of problem rather than the language.

Filter

CareQueue

Prototype

A triage assistant built around the patients who are missed rather than the ones who are obvious: it scores risk from outcomes instead of from assigned levels, and keeps watching everyone in the waiting room for deterioration.

The problem

Triage is usually built as classification — a nurse assigns a severity level and the model learns to agree. But in the 2022 NHAMCS data, 45.8% of critical-care outcomes came from patients triaged level 3, 4 or 5. They were correctly processed, and then they waited. A model that agrees with the nurse learns to miss exactly the same people.

Built with

  • TypeScript
  • Next.js
  • React
  • NHAMCS 2022