Clinical Communication Engineering Research Roadmap

Program outcomes

Primary outcomes are correct situation understanding, critical-information detection, decision/action accuracy, communication closure, and preventable use error. Secondary outcomes include time, workload, recall, accessibility, trust calibration, and satisfaction.

Phase 0 — Governance and hazards

• Form a council spanning physicians, nurses, patients/caregivers, clinical informatics, human factors, accessibility, safety/risk, privacy/security, and regulatory/legal expertise.
• Define intended uses and prohibited claims.
• Establish data governance, adverse-event escalation, source/version control, and independent review.
• Produce use-related risk analyses for each setting.

Gate: no patient-data prototype or clinical pilot without governance and hazard controls.

Phase 1 — Cognitive field research

• Contextual inquiry and cognitive task analysis in primary care, emergency, hospital medicine, specialty referral, nursing handoff, telemedicine, and patient result review.
• Sample experts and novices, accessibility needs, language/numeracy variation, interruptions, mobile and print.
• Map decisions, cues, uncertainty, workarounds, ownership transitions, and failure recoveries.

Outputs: role-task models, common and divergent needs, critical cue inventory, baseline workflow/error measures.

Phase 2 — Information-architecture experiments

Compare source order, problem order, task-first summary, timeline, and layered hybrids using de-identified/synthetic cases. Test stable first-layer concepts while allowing role-specific ordering.

Measures: time to coherent summary, diagnostic/plan accuracy, critical omissions, contradiction detection, confidence calibration, NASA-TLX or validated workload measures where appropriate.

Phase 3 — Visual-variable experiments

Factor typography, density, spacing, table versus card/list, trend representation, redundant urgency encoding, and progressive disclosure. Test grayscale, color-vision variance, zoom/reflow, screen readers, glare, print, and interruption recovery.

Rule: do not change multiple visual factors in a way that prevents attribution unless evaluating the complete system as a bundle.

Phase 4 — Communication artifacts

Develop separate validated patterns for:

• emergency/inpatient handoff,
• referral and consultation,
• result notification/follow-up,
• longitudinal summary,
• patient explanation,
• mobile rounding view,
• printed downtime/transfer packet.

Test closed-loop state and ownership, not document completeness alone.

Phase 5 — AI collaboration trials

Use shadow mode first. Benchmark extraction, timeline assembly, summarization, discrepancy detection, and plain-language translation against dual human review. Stratify by specialty, complexity, demographic subgroup, and missing/conflicting data.

Required metrics: unsupported claims, omissions, temporal errors, contradiction preservation, source-link accuracy, correction time, automation bias, over/under-trust, and downstream action errors.

Phase 6 — Prospective controlled deployment

Begin with low-risk, reversible workflows. Use stepped or controlled designs where feasible, predefine stopping rules, monitor balancing measures, and maintain rollback. Do not infer safety from adoption or satisfaction.

Phase 7 — Discipline infrastructure

• Versioned pattern and evidence registries.
• Standard scenario/case bank and benchmark tasks.
• Common outcome taxonomy and incident reporting.
• Certification claims scoped by artifact, role, setting, version, and tested population.
• Public research-debt and invalidation log.

Initial experiment queue

Stopping and invalidation rules

• Stop a study arm on predefined critical-error or inequity signals.
• Reject patterns that improve speed while worsening critical omission beyond approved bounds.
• Reopen the standard when setting transfer fails, source guidance changes, or monitoring identifies new hazards.