CarePathIQ: Empowering Emergency Medicine Trainees to Embed Clinical Reasoning Into Explainable Artificial Intelligence Systems

AI is rapidly entering emergency medicine, yet trainees seldom learn to inspect, adapt, or operationalize AI-supported tools. Opaque “black box” systems erode trust and slow CDS adoption. Trainees need transparent, editable AI experiences that teach evidence appraisal (e.g., GRADE), decision logic with explicit evidence links, usability and workflow fit, and operationalization into deployable assets. Hands-on building of explainable AI systems can build trust, AI literacy, and management reasoning in emergency medicine.

AI is rapidly entering emergency medicine, yet trainees seldom learn to inspect, adapt, or operationalize AI-supported tools. Opaque “black box” systems erode trust and slow CDS adoption. Trainees need transparent, editable AI experiences that teach evidence appraisal (e.g., GRADE), decision logic with explicit evidence links, usability and workflow fit, and operationalization into deployable assets. Hands-on building of explainable AI systems can build trust, AI literacy, and management reasoning in emergency medicine.

CarePathIQ is a Streamlit web application powered by Google Gemini. The framework implements a structured 5-phase methodology: (1) Scoping & Charter, (2) Rapid Evidence Appraisal with automated PubMed searches and GRADE evaluation, (3) Decision Science generating evidence-linked decision trees visualized via Mermaid flowcharts, (4) User Interface Design with Nielsen's heuristics analysis, and (5) Operationalization producing beta testing guides, feedback forms, PowerPoint presentations, and interactive education modules. The platform necessitates clinician oversight and transparency throughout pathway development.

CarePathIQ produces transparent, evidence-linked pathways with every decision tied to PubMed citations and GRADE ratings. It outputs operational assets (beta testing guide, feedback form, slides, interactive education module) and usability findings (Nielsen heuristics) to address workflow fit. Trainees learn AI literacy by building and editing the AI-augmented logic, improving trust and management reasoning. Planned measures: usability/feasibility, task completion/time-to-build, perceived transparency/trust, and self-reported AI literacy gains.

CarePathIQ is a free, open-source, Streamlit-based AI/LLM platform that teaches emergency medicine trainees to build explainable, evidence-linked clinical pathways. In this hands-on tabletop session, participants will use the live web app to experience end-to-end pathway development with full transparency and clinician control, directly addressing SAEM Theme #5 on AI/LLMs in medical education. The innovation moves beyond passive AI use by asking trainees to construct and edit AI-augmented pathways, emphasizing explainability, PubMed/GRADE evidence linkage, and human oversight to overcome trust and adoption barriers. The platform uses a single Gemini-powered AI agent within a five-phase workflow. Phase 1 captures the condition and setting, auto-generates inclusion and exclusion criteria, a problem statement, and SMART objectives, and produces a project charter with an editable Gantt. Phase 2 runs automated PubMed searches from the last five years, retrieves citations, and applies AI-assisted GRADE ratings with editable rationales. Phase 3 generates evidence-linked decision trees with multi-branch logic, role assignments, and explicit evidence IDs, then visualizes them via Mermaid syntax flowcharts with color-coded nodes and role swimlanes. Phase 4 applies Nielsen heuristic analysis to identify usability issues, offers actionable fixes, and supports undo and natural-language refinement. Phase 5 creates deployable assets, including a beta testing guide, feedback form, branded slides, an interactive education module, and an executive summary. During the demonstration, attendees will enter a mock EM case, run a PubMed search, review AI-generated GRADE ratings and rationales, edit decision nodes, and watch evidence-linked flowcharts update in real time. They will see usability critiques applied and reversed, and then download the asset bundle for implementation and teaching. Planned evaluation focuses on feasibility and usability, task completion and time-to-build a pathway, perceived transparency and trust, and self-reported gains in AI literacy and management reasoning, with qualitative feedback on workflow fit. Educational and operational impact centers on teaching AI literacy, evidence appraisal, decision science, and usability within one integrated workflow, while reducing cognitive load by producing ready-to-use operational and educational assets. By tying every decision to explicit evidence with editable rationales, the platform addresses the black box problem and supports equity and safety by surfacing evidence gaps and usability risks before implementation. A tabletop format allows attendees to interact directly with the live app at https://carepathiq.streamlit.app/, iterate on prompts and edits, and immediately observe changes in evidence tables, decision logic, heuristic findings, and downloadable outputs. The session uses de-identified mock scenarios and demonstrates how EM trainees can build explainable AI systems for clinical pathways, strengthening AI literacy and operational readiness while maintaining clinician oversight.