Tampa General Leads With Voice-Driven AI That Actually Helps Clinicians
Voice is the UI. Tampa General Hospital (TGH) is putting that idea to work with two practical tracks: voice biomarkers to predict heart failure decline, and ambient listening to auto-document nursing assessments straight into Epic.
If you build health tech, this is a useful blueprint: small, high-leverage pilots, tight vendor feedback loops, and workflows that make clinicians' lives easier without adding clicks.
Voice biomarkers: early signal for heart failure
TGH participated in an international study of HearO, a voice-analysis app from Cordio. Patients speak five short sentences each morning; the system looks for changes linked to fluid buildup that often appear 2-4 weeks before symptoms.
When detected, the app alerts the care team so meds can be adjusted before the patient deteriorates. Compared to implanted pulmonary artery monitors, this path is cheaper, easier to scale, and doesn't require a device inside the body.
Why this matters for IT and dev teams: voice becomes a passive daily data stream you can turn into actionable signals. It's low friction for the patient and integrates cleanly into existing care pathways.
- Operational notes: baseline enrollment, daily adherence tracking, alert thresholds, and routing to care managers.
- Data: secure storage of audio features (not full audio if policy requires), drift monitoring, false positive review, and clinician override.
- UX: simple morning prompt, offline capture with queued upload, and clear consent language.
Ambient listening for nursing: documentation without the keyboard
TGH is one of eight beta sites co-developing Microsoft's Dragon Copilot for nursing. On a neuroscience med-surg floor, nurses and techs speak head-to-toe assessments while engaging with patients; notes flow into Epic in real time via a secure iPhone app.
Consent is requested at each shift. Sensitive details in semi-private rooms are dictated in a private setting. Adoption friction exists-speaking assessments out loud takes practice-but patients like the increased attention and eye contact.
TGH already rolled out DAX Copilot for physicians in 2024. The nursing version is more complex, and the team meets weekly with Microsoft to fine-tune the app. Expansion is planned for respiratory therapy, PT, case management, and more. Next step: context-aware prompts, like reminding a nurse to delay meds before a procedure or flagging potential side effects.
- Platform: company-issued iPhones under MDM, SSO, least-privilege access, encrypted streaming, and PHI-safe processing.
- EHR integration: Epic APIs, structured fields + narrative notes, audit trails, and rollback options.
- Safety: consent logging, wake-word/record controls, private-mode workflows, redaction of bystanders, and diarization accuracy.
- Reliability: latency targets, on-device buffering, service SLAs, and a fallback path for manual charting.
- Governance: human-in-the-loop verification, error-rate tracking, data retention limits, and incident response playbooks.
What to measure
- Documentation time per assessment and total EHR time per shift.
- Correction rate on generated notes and time-to-finalization.
- Patient consent rate and satisfaction scores.
- Clinician satisfaction, burnout indicators, and training hours needed.
- For HearO: alert precision/recall, intervention lead time, and 30-day readmissions.
- Privacy incidents, uptime, and audio processing latency.
Playbook for health IT teams
Start small: one unit, one workflow. Map the data path, secure it end-to-end, and make it easy for clinicians to correct the output. Meet weekly with your vendor, log every friction point, and ship improvements fast.
- Run a short pilot with explicit patient consent and clear signage.
- Threat model the audio pipeline; verify encryption at rest/in transit; restrict access by role.
- Define note structures and acceptance criteria with nurse leaders and compliance.
- Instrument everything: time saved, corrections, error types, and reasons for opt-outs.
- Train staff on voice workflows and privacy etiquette; keep a crisp cheat sheet in the app.
For more on Tampa General's work bringing a more human touch to AI-driven patient care, visit TGH.org. To see how Microsoft frames ambient documentation for clinicians, explore DAX Copilot.
Your membership also unlocks:
