AI in Cardiac Care: Practical Lessons from Health Systems and Heart+
Credit: Getty Images
Artificial intelligence has moved from idea to everyday tool in cardiology. Algorithms are now embedded across diagnostics and workflows, and health systems are learning where these tools help, where they struggle, and how to use them responsibly.
To ground this in practice, we spoke with Faraz Ahmad, MD, associate director of the Bluhm Cardiovascular Institute Center for Artificial Intelligence at Northwestern Medicine and associate editor for AI in JAMA Cardiology, and Marie-Noelle Langan, MD, former director of operations in the electrophysiology division at Mount Sinai and chief medical advisor at 91Life. Their insights focus on real patient impact, clinician workflows, and what it takes to scale safely.
Where AI is showing value today
Within health systems, ambient listening for clinical note generation has seen swift adoption. Done well, it gives clinicians time back, reduces clerical load, and keeps the patient conversation front and center.
In cardiovascular care, AI is increasingly built into core diagnostics. Algorithms assist with:
- ECG interpretation and signal detection beyond human pattern recognition (eg, flags suggestive of conditions like cardiac amyloidosis)
- Automated measurements and reporting in echocardiography, cardiac CT, and MRI
- Structured data generation that feeds decision support at the point of care
Large language model-based tools are also showing up in clinician education and decision support. The key is using systems that cite trusted medical sources and fit within clinical governance rather than operating in isolation. For broader context, see the American Heart Association scientific statement on AI in heart disease here.
Benefits, tradeoffs, and governance that actually works
The promise is clear: better care delivery, stronger adherence to evidence, and improved patient outcomes. The reality: tools carry costs, onboarding time, and a need for proof they do what they claim.
Risks include workflow friction, alert fatigue, and the potential to widen disparities if models are misapplied. Leading systems are responding with internal AI governance, standardized evaluation frameworks, and clear buy/build/partner pathways. The goal is to measure impact, monitor for unintended effects, and adjust or retire tools that don't deliver.
Staying current without wasting hours
Treat AI literacy like any other clinical competency. Read the studies, attend society sessions, and subscribe to trusted updates so you can separate signal from noise.
If you want a curated, time-efficient way to upskill on applied AI by role, this directory can help: AI courses by job.
What's still missing
Across use cases, the core questions are consistent: Do these tools measurably reduce admin burden? Do they increase patient activation? Do they improve outcomes? And do the benefits outweigh the costs?
Health systems also need reliable processes to detect unintended consequences early and to educate clinicians on when-and when not-to use AI. More pragmatic research and shared best practices will move the field forward.
Heart+: A focused case study in remote cardiac monitoring
Heart+ is a cloud-native platform from 91Life that supports in-clinic and remote monitoring of cardiac-implantable electronic devices, ambulatory ECGs, and wearables. It consolidates data, triages alerts using machine learning, generates reports, pushes structured data into the EHR, and auto-bills. The platform has been implemented at Mount Sinai and Yale.
Adoption and day-to-day impact
According to Dr Langan, electrophysiologists respond well to the platform's practical value once they see it in action. Initial deployment brings understandable anxiety-clinicians are rightly protective of established routines-but staged rollouts and hands-on support help teams settle in quickly.
Two patterns stand out. First, feedback loops matter: clinicians flag issues, the vendor resolves and prevents recurrence, and trust grows. Second, incremental improvements to speed, capacity, and clarity win support when they match how clinicians think and work.
Workflow visibility and patient connectivity
Early on, clinic-level supervision of transmission workflows was hard. Heart+ answered with dashboards and graphics that make it easy to spot delays or gaps and verify task completion.
Keeping patients connected to devices was another challenge. Automated patient connectivity and KPI tracking now support outreach at the volume needed for strong monitoring. Patients don't work directly with Heart+, but they notice quicker, clearer answers from their care teams and benefit from physician-authored instructions delivered through the platform.
Security, privacy, and bias
Per Dr Langan, the platform layers HIPAA requirements on top of an architecture shaped by high-security finance use cases. De-identification is built in, enabling case discussion without unnecessary exposure of patient data.
On bias, Heart+ primarily collates data from commercially deployed sources and presents it consistently across patients. The team pays close attention to language accessibility, personalized outreach, and simple teaching tools to support diverse patient groups.
Environmental footprint
91Life's AI and Data team focuses on lean, purpose-built models instead of large, general models. In practice, that means smaller, classical approaches-such as tree-based models-that are computationally light while maintaining predictive performance for tasks like filtering clinically relevant transmissions.
A practical playbook for health systems
- Start with a narrow, high-impact use case (eg, ambient note generation in cardiology clinic or ECG triage) and define success metrics upfront.
- Integrate with existing workflows and EHRs; avoid creating yet another inbox.
- Set governance: data quality reviews, model monitoring, bias checks, and clear escalation paths.
- Pilot, measure, iterate; retire tools that don't show value within a defined timeframe.
- Invest in clinician training and feedback loops; the best insights come from daily use.
- Build patient-friendly communication: plain language, visual aids, and timely outreach.
- Protect privacy by default and make de-identification seamless for case reviews.
- Track compute needs and model size to meet institutional sustainability goals.
The bottom line
AI is becoming part of the standard toolkit in cardiology. The winners will be the tools that measurably improve care, respect clinician time, and fit cleanly into existing systems.
Heart+ shows how focused deployments-data consolidation, alert triage, EHR integration, smart dashboards, and patient connectivity-can streamline remote monitoring. With strong governance and continuous feedback, health systems can scale these gains without losing sight of equity, privacy, or sustainability.
Your membership also unlocks:
