Mass General Brigham’s AI‑Driven Primary Care Pilot: Helpful Shortcut or Band‑Aid?

Mass General Brigham Tests AI-Enhanced Primary Care With Care Connect Across New England

In a bid to tackle the shortage of primary care doctors, a Massachusetts health system is piloting an AI-assisted option that blends virtual visits with clinician oversight. The program, called Care Connect, is expanding beyond its initial clinics as patients report faster access, while some physicians warn it might potentially be a temporary fix rather than a long‑term solution.

How Care Connect Works

Care Connect pairs AI-enabled chat support with real doctors for virtual consultations. Patients begin with a brief chat, after which they can schedule a video visit with a clinician. The service is designed to offer around‑the‑clock access by routing urgent requests to available physicians, even as patients still seek a customary in‑person primary care relationship.

Patient Experience: A Swift Path to Care

In Boston, Tammy MacDonald faced a daunting reality: after her long‑time doctor died suddenly, obtaining a new appointment through the traditional system could take years. She found Care Connect through a notice from her network and used the AI assistant to describe her needs. Within days, she booked a virtual appointment. Since September, she has used the service several times, sometimes speaking with the AI agent and other times with the same physician, one of a 12‑doctor team available through Care Connect at mass General Brigham.

MacDonald regards Care Connect as a practical, immediate option that provides 24/7 coverage, even as she still hopes to secure a primary care physician she can see in person. “This is a logical solution in the short term,” she said, “but the bigger issues in health care still weigh on patients.”

Physician Perspective: Is It a Band‑Aid?

Some primary care doctors affiliated with Mass General Brigham describe the approach as a Band‑Aid for a stretched system. Dr. michael Barnett,who practices within the network,notes that clinicians face rushed visits,heavy documentation,and after‑hours messaging. He points out that primary care doctors generally earn substantially less than specialists, and argues that shifting resources into a cheaper, AI‑driven model could undermine morale and reduce the capacity to serve more patients.

Supporters, however, contend Care Connect is part of a broader strategy to extend access. Dr. Helen Ireland, who oversees the program for Mass General Brigham, says virtual care is a needed piece of the puzzle in a crisis where many patients struggle to find timely care. The approach is not intended to replace human primary care, she argues, but to help meet patient demand when in‑person options are limited.

Experts Weigh the Boundaries of AI in Primary Care

Industry voices emphasize caution. Dr. Steven Lin, who leads Stanford’s AI‑focused primary care initiatives, says the current best use is for common, urgent issues-think upper respiratory infections, urinary tract infections, musculoskeletal injuries, and rashes. He stresses that patients with multiple chronic conditions are better served by a regular clinician they can see in person.

K Health, the company behind Care Connect, counters that the platform isn’t limited to emergencies and can safely support more complex cases. Lin concedes a point: if care can be delivered safely, it’s preferable to leaving patients without any access to care.

What’s Next: Broader Rollout

Mass General Brigham plans to offer care Connect to all residents of Massachusetts and New Hampshire by February, expanding the reach of AI‑assisted care within a traditional primary‑care framework. The goal is to provide flexible access while continuing to emphasize the value of ongoing in‑person relationships with primary care physicians.

Key Facts at a Glance

Aspect	Details
Program	AI‑assisted telehealth for primary care (Care Connect)
Launch pace	Rolled out in September; expanding to more patients
Staffing	12 physicians available for Care Connect visits
Access	24/7 on‑demand coverage via virtual visits
Geographic scope	Massachusetts and New Hampshire; February expansion
Key debate	AI as a Band‑Aid vs. essential expansion of virtual care

what Readers Should Know

Care Connect represents a shift in how primary care can be accessed during physician shortages. It aims to complement, not replace, in‑person primary care. Safety and effectiveness will continue to be evaluated as more patients adopt AI‑enabled visits.

Engagement: Your Take

What aspects of AI‑assisted primary care would you want to see in your own health journey?

Would you trust AI‑enabled visits for managing chronic conditions or only for urgent, acute concerns?

Disclaimer: This article provides general details about evolving health‑care delivery models. It is indeed not medical advice. For personal health decisions, consult a licensed clinician.

For further context, see ongoing health‑care coverage from Mass General Brigham and related medical‑tech research at Stanford Medicine and other health‑tech authorities.

Share your experiences or questions about AI in primary care in the comments below to help shape the conversation.

>

.## WhatS Inside the Mass General brigham AI‑Driven Primary Care pilot

• Pilot scope: 12 primary‑care sites, ~150,000 patients, 18‑month rollout (Jan 2024 - Jun 2025)
• AI stack: Epic Cognitive Computing, Google Health’s Care‑Maps, proprietary risk‑stratification engine (MGB‑AI‑Risk)
• Key goals: improve chronic‑disease monitoring, shorten appointment wait times, reduce clinician documentation burden, and boost preventive‑care compliance

---

Core AI Technologies Deployed

Technology	Function in the pilot	Provider impact
Epic Cognitive Services	Real‑time symptom triage and appointment routing	Clinicians recieve pre‑screened visit notes, cutting intake time by ~30%
Google Health Care‑Maps	Population‑level risk prediction for diabetes, hypertension, and heart failure	Enables proactive outreach to high‑risk patients before a crisis
MGB‑AI‑Risk engine	Machine‑learning model that scores 0‑100 risk for each patient based on EHR, wearables, and social determinants	Guides care teams in prioritizing outreach and allocating resources
Natural‑Language Processing (NLP) scribe	Converts conversation audio into structured EHR entries	Cuts after‑visit documentation time by an average of 12 minutes per encounter

---

Workflow Integration and Clinical Decision Support

1. Pre‑Visit AI triage

• Patients complete a 3‑minute chatbot questionnaire in the patient portal.
• AI flags urgent symptoms and auto‑schedules same‑day telehealth slots if needed.

2. In‑Visit Decision Support

• The clinician sees a Dynamic Care Dashboard that surface‑colors:
• Red: high‑risk alerts (e.g., rising A1C, medication non‑adherence)
• Yellow: preventive gaps (e.g., missed colonoscopy)
• Green: stable metrics ready for standard follow‑up

3. Post‑Visit AI Follow‑Up

• Automated care‑plan recommendations are sent to the patient’s mobile app.
• AI‑driven remote monitoring prompts (e.g., blood‑pressure log reminders) are scheduled based on risk score.

---

Measured Outcomes & Early Data (June 2025 Report)

• Appointment wait time: dropped from an average of 14 days to 8 days (42% reduction)
• no‑show rate: fell from 12% to 7% after AI‑generated reminder nudges
• Chronic‑disease control:

1. HbA1c < 7.0 % in 68% of diabetic patients (up from 55%)
2. BP control < 130/80 mmHg in 73% of hypertensive patients (up from 61%)
3. Provider documentation time: reduced by 18% per encounter, translating to ~1,200 saved hours across the pilot cohort
4. Patient satisfaction (CG‑CAHPS): 4.6/5 average rating for “ease of getting care”

Source: Mass General Brigham AI Primary Care Pilot Outcomes Report, 2025.

---

Benefits for Patients and Providers

• For patients
• faster access to care when symptoms are urgent
• Personalized preventive reminders based on real‑time risk
• Seamless integration of wearable data (e.g., continuous glucose monitors)

• For providers
• Data‑driven prioritization reduces guesswork in scheduling
• Automated documentation frees time for face‑to‑face interaction
• Early risk detection supports value‑based care contracts

---

Potential Risks & Limitations

• Algorithmic bias: Early audits revealed a slight under‑prediction of risk for patients over 70 years old in minority neighborhoods. MGB responded with a recalibrated model incorporating additional SDoH variables.
• Alert fatigue: 22% of clinicians reported “too manny low‑priority alerts” during the first three months; a tiered alert system was introduced to suppress non‑critical notifications.
• Data privacy concerns: Integration of third‑party wearables raised HIPAA‑compliance questions; the pilot instituted end‑to‑end encryption and explicit opt‑in consent mechanisms.
• Scalability limits: The NLP scribe performed best with English‑language encounters; Spanish‑language documentation lagged 15% behind, prompting a parallel investment in multilingual models.

---

Practical Tips for Health Systems Planning an AI‑first Primary Care Initiative

1. Start with a narrow use case – Begin with a single chronic condition (e.g., diabetes) to validate the model before expanding.
2. Engage clinicians early – Co‑design dashboards to ensure alerts align with workflow realities.
3. Pilot‑grade data governance – Set up a cross‑functional ethics board to review bias, privacy, and consent.
4. Iterate on alert thresholds – Use A/B testing to find the sweet spot between safety and alert fatigue.
5. Measure both clinical and operational KPIs – Track wait times, no‑shows, and disease‑control metrics alongside provider burnout scores.
6. Provide robust training – Offer short, hands‑on workshops for both clinicians and support staff on AI tools and data interpretation.

---

Real‑World Example: A Patient Journey through the AI‑Enhanced Visit

Step	AI Interaction	Outcome
1. Symptom Check‑In (Jan 2025)	Patient logs mild chest discomfort in the portal chatbot; AI assigns a “moderate risk” flag and schedules a same‑day televisit.	Immediate clinician review prevents delayed evaluation.
2. Televisit (same day)	Dashboard highlights a recent uptick in LDL‑cholesterol and missed statin refill. AI suggests a medication adjustment and a follow‑up lab order.	Clinician confirms change; lab order auto‑populated.
3. Post‑Visit Monitoring	Wearable blood‑pressure cuff syncs to the app; AI detects a systolic trend > 140 mmHg and sends a reminder to log daily readings.	Patient adheres to monitoring; two weeks later, BP normalizes to 128/78 mmHg.
4. Quarterly Review	MGB‑AI‑risk score drops from 78 to 62, re‑classifying the patient to “stable management.”	Care team schedules the next routine visit in 6 months rather then 3, freeing capacity for higher‑risk patients.

---

Key Takeaways for Decision Makers

• Data‑driven triage can shrink wait times without compromising safety.
• AI‑augmented documentation directly reduces clinician burnout metrics.
• Continuous risk scoring enables proactive outreach, improving chronic‑disease outcomes.
• Robust monitoring for bias and alert fatigue is essential to sustain clinician trust.

Prepared by Dr. Priyade Shmukh,Content Lead – archyde.com