VitaSignal: AI-Powered Clinical Decision Support

Gemini 3-powered clinical decision support that detects patient deterioration hours earlier using the Intensive Documentation Index, reducing ICU transfers and improving nurse-sensitive outcomes.

Comment

VitaSignal: AI-Powered Clinical Decision Support

Tagline: Clinical Intelligence Without Equipment — predicting ICU patient deterioration from documentation patterns, not devices.

Inspiration

As an informaticist and clinical researcher, I witnessed firsthand how patient deterioration often goes undetected — not because clinicians aren't attentive, but because the signals are buried in documentation patterns that humans can't process in real time.

The Intensive Documentation Index (IDI) was born from a simple insight: how clinicians document reveals as much about patient status as what they document.

What It Does

VitaSignal™ LLC is a Gemini 3-powered clinical decision support platform that predicts ICU patient deterioration using 9 temporal patterns in existing EHR documentation — without requiring physiological sensors, lab values, or additional medical devices.

Two core systems:

  • IDI (Intensive Documentation Index) — predicts ICU mortality risk from documentation timing patterns, validated across independent international cohorts in the USA and Switzerland
  • DBS (Documentation Burden Score) — predicts nursing workload and burnout risk, validated across U.S. hospitals

Gemini 3-powered features:

  • Clinical Notes Analyzer — extracts early warning signs from free-text nurse observations in real time
  • Explainable Risk Narratives — converts SHAP outputs into plain-language bedside explanations
  • Intervention Recommender — surfaces evidence-based nursing interventions by clinical urgency
  • Health Equity Analyzer — detects demographic disparities in AI risk predictions (NIH AIM-AHEAD aligned)

⚠️ Decision-support only. All findings require clinical verification. Not a medical device.

How We Built It

  • Frontend: React + TypeScript + Tailwind CSS, deployed on cloud-hosted production infrastructure
  • AI Engine: Dual Gemini 3 model integration — Flash for speed-critical analysis, Pro for complex reasoning
  • Backend: Supabase (PostgreSQL, Edge Functions, Row-Level Security)
  • Security: No API keys in client code — all AI routed through secure server-side gateway; HIPAA-safe synthetic data only
  • Research Foundation: Validated across independent international ICU cohorts, with manuscripts under review
  • Patent Portfolio: 11 U.S. Provisional Patent Applications Filed — February 28, 2026

Challenges We Faced

  • Balancing clinical accuracy with real-time performance — solved using Gemini 3 Flash for triage and Pro for deep analysis
  • Making complex ML predictions interpretable for bedside clinicians without statistical backgrounds
  • Ensuring health equity by detecting and surfacing potential demographic bias in risk predictions
  • Demonstrating that documentation timing alone — with no physiological data — could outperform established clinical severity scores validated over decades

What We Learned

Gemini 3's multimodal reasoning capabilities are transformative for healthcare AI. The ability to analyze clinical narratives, generate human-readable explanations, and reason about intervention timing in a single model family dramatically simplifies the clinical AI pipeline.

The strongest signal in clinical AI may not come from what is documented — but from the rhythm and timing of documentation itself, a finding now validated across independent international databases spanning two countries.

What's Next

  • Accepted: ANIA 2026, Boston, MA — March 26–28, 2026
  • IRB-approved prospective clinical validation — pilot site recruitment underway (Q2 2026)
  • Real-time EHR integration via FHIR APIs
  • FDA pre-submission meeting (2026)
  • Expansion to sepsis early warning, shift handoff quality scoring, and global health settings

Built With

react typescript tailwind-css supabase edge-functions postgresql gemini-3-flash gemini-3-pro recharts framer-motion vite

Try It Out

© 2026 VitaSignal LLC. All rights reserved. Patent Pending — USPTO. Submitted to Gemini 3 Hackathon 2026 | Powered by Google Gemini 3 API

Built With

Share this project:

Updates