AI Clinical Reasoning for Chronic Care
Beyond Alerts. Beyond Documentation.
Ardia Health uses Google Cloud's Healthcare API, Vertex AI, and BigQuery to deliver longitudinal clinical reasoning that predicts asthma/COPD exacerbations 7 days in advance — reducing ER visits by 40% while generating $180/patient/month in RPM revenue.
A. How It Works — Clinical Reasoning Engine
End-to-End Flow: From Patient Data to Clinical Action
❌ Why Existing AI Fails
- Transcription ≠ Intelligence: Ambient scribes only document — they don't reason.
- Alerts ≠ Decisions: Rule-based alerts create noise, not insight.
- Burnout ≠ Solved: More tools = more clicks = more frustration.
✓ Ardia's Approach: Longitudinal Clinical Reasoning
We don't just alert providers — we reason across time, predict deterioration, and recommend the next-best-action based on the full patient context.
Clinical Reasoning Engine — 5-Step Flow
Data Ingestion
Sources: FHIR (via Google Healthcare API), EHR integrations (Epic, Cerner), wearables (Apple Health, Fitbit), patient-reported symptoms (mobile app)
Google Cloud: Cloud Healthcare API FHIR store, Pub/Sub for real-time streaming
Symptom + History Normalization
Process: Structured + unstructured data cleaned, mapped to clinical ontologies (SNOMED CT, LOINC)
Google Cloud: Dataflow for ETL, BigQuery for longitudinal patient timelines
Clinical Reasoning Layer (Not Alert-Based)
AI Engine: Temporal models analyze trends, patterns, deviations. No static thresholds — context-aware reasoning.
Example: "Patient's FEV1 declining 3% weekly + rescue inhaler use up 40% + nighttime symptoms increasing → 7-day exacerbation risk = 78%"
Google Cloud: Vertex AI custom-trained models, AutoML for feature engineering
Risk Stratification & Next-Best-Action
Output: Risk scores (0-100), recommended interventions ranked by clinical impact
Example Actions: "Schedule urgent pulmonology visit," "Increase controller dose," "Patient education: inhaler technique"
Google Cloud: Cloud Functions for real-time decisioning, Firestore for action tracking
Provider + Patient Outputs
Provider Dashboard: Prioritized patient list, clinical reasoning explained, one-click interventions
Patient App: Simple, actionable guidance ("Your asthma risk is elevated — use your controller inhaler twice daily")
Google Cloud: Cloud Run for web dashboard, Firebase for mobile app backend
🔒 HIPAA Compliance Built-In
All PHI encrypted at rest (Cloud KMS) and in transit (TLS 1.3). BAA signed with Google Cloud. Role-based access (Cloud IAM). Audit logs (Cloud Logging) for every data access.
B. Google Cloud Architecture
Our Infrastructure Stack (Current + Planned)
Ardia Health on Google Cloud
📊 Data Layer
🤖 AI/ML Layer
⚙️ Application Layer
🔐 Security & Compliance
📈 Monitoring & Observability
Why This Needs Cloud Scale
Longitudinal Data: Each patient generates 10,000+ data points over 12 months (vitals, medications, lab results, symptoms). BigQuery handles petabyte-scale analytics.
Real-Time Inference: When a patient reports worsening symptoms, our Vertex AI models must predict exacerbation risk in <200ms.
HIPAA at Scale: Google Cloud's Healthcare API is the only FHIR-native solution with built-in BAA, de-identification, and consent management.
C. Real Use Case — Asthma/COPD Care (Phase 1)
How Ardia Transforms Respiratory Care
The Problem
- 22 million Americans have asthma. 16 million have COPD.
- ❌ 2 million ER visits/year ($3,000-$15,000 per visit)
- ❌ 60% of patients misuse inhalers → poor control
- ❌ Clinicians see 300+ patients → can't monitor all proactively
- ❌ Care is reactive: patients deteriorate silently until crisis
Ardia's AI Reasoning in Action
Day 1
Patient reports increased breathlessness in mobile app. Rescue inhaler use: 3x/day (normal: 1x/week).
Day 2
Ardia AI analyzes: FEV1 trend ↓ 5%, nighttime symptoms ↑, pollen levels elevated. Exacerbation risk: 68%
Day 3
Provider dashboard flags patient as High Priority. Recommends: increase controller dose, schedule telehealth visit.
Day 7
Patient stabilizes. ER visit avoided ($8,000 saved). Clinic earns $60 RPM reimbursement for remote monitoring.
Sample Provider Dashboard — Detailed Clinical Insights
🔴 HIGH PRIORITY PATIENTS (3)
⚠️ Risk Score: 78% (7-day exacerbation risk)
Last visit: 3 days ago | Next scheduled: 12/24/2026
⚠️ Risk Score: 65% (7-day exacerbation risk)
Last visit: 5 days ago | Next scheduled: 12/26/2026
⚠️ Risk Score: 58% (7-day exacerbation risk)
Last visit: 1 day ago | Next scheduled: 12/28/2026
✅ STABLE PATIENTS (4)
✓ Interactive dashboard — buttons open simple modals for scheduling, messaging, prescriptions