STAT+: Disagreement Between Patients and Hospitals on AI
Decision brief
Answer first · skim in under a minute
This article delves into the contrasting views of patients and hospitals regarding AI in healthcare, highlighting key insights and implications for the pharmaceutical industry.
Patients and hospitals are not aligned on healthcare AI: most surveyed U.S. adults distrust health systems to use AI safely, while more than half of hospitals already use or plan near-term generative AI inside electronic health records.
Contents9 sections
Key Takeaways
- 65.8% of surveyed adults reported low trust that their health system would use AI responsibly; 57.7% doubted harm protection.
- Among 2,174 hospitals, 31.5% were early generative-AI adopters by 2024 and 24.7% were fast followers planning adoption within a year.
- Patient choice and trust rose most with high AI performance, then with FDA approval, Mayo Clinic-style national certification, local hospital certification, and clinician oversight.
- Pharma and digital-health teams should treat FDA clearance, disclosure, and human-in-the-loop design as market-access requirements, not optional ethics extras.
Where do patients and hospitals disagree on AI?
Hospitals are buying generative AI to draft notes, triage messages, and support documentation workflows that cut labor cost and cycle time.
Patients are asking a different question: will the system use AI responsibly, and will someone stop the tool from harming me?
That gap is now measurable in peer-reviewed surveys rather than anecdote. Trade-press coverage of the divide is secondary; the evidence base sits in JAMA Network Open and related Health Affairs hospital AI studies.
What do patient trust surveys show?
Nong and colleagues surveyed U.S. adults about trust in health systems to use AI responsibly and to protect patients from AI harms.
Most respondents reported low trust on both outcomes: 65.8% for responsible use and 57.7% for harm prevention, according to the JAMA Network Open patient trust study (doi:10.1001/jamanetworkopen.2024.60628).
Trust varied with AI knowledge, health literacy, baseline system trust, and prior discrimination experiences in care. Teams that ignore those covariates will misread average scores.
How quickly are hospitals deploying generative AI?
Everson and colleagues analyzed a national sample of 2,174 nonfederal acute-care hospitals on generative AI integrated with electronic health records.
By 2024, 31.5% were early adopters and 24.7% planned adoption within a year, leaving 43.7% delayed or uncertain, per JAMA Network Open hospital generative AI adoption research (and the related 2025 hospital AI evaluation literature in Health Affairs).
Major teaching and system-affiliated hospitals moved faster. Critical-access and rural hospitals lagged. Hospitals already using predictive AI were far more likely to adopt generative tools early.
Notably, hospitals that ran more comprehensive local evaluations for accuracy, bias, and post-deployment performance were sometimes slower to implement than peers with thinner diligence.
What governance signals raise patient acceptance?
A separate 2026 JAMA Network Open survey of 3,000 U.S. adults tested which attributes move trust and choice in medical AI encounters.
Respondents preferred FDA-approved tools over unapproved ones, nationally certified tools over uncertified ones, and locally certified tools over uncertified ones.
Clinician presence and representative training data also raised preference. The largest effect, however, was AI performance at generalist or specialist levels.
Full methods and average marginal component effects are in Factors for Patient Trust and Acceptance of Medical Artificial Intelligence.
What should pharma and digital-health teams do?
Product teams building AI diagnostics, messaging, or decision support should budget for FDA pathway clarity early. Patients treat clearance as a trust proxy even when they cannot audit model cards.
Commercial teams selling into health systems should assume hospital CIOs will move faster than patient trust recovers. Launch plans need patient-facing disclosure language, clinician review workflows, and equity monitoring.
Medical affairs should prepare evidence packages that speak to both audiences: operational ROI for hospitals, and performance plus governance for patients and advocacy groups.
Regulatory context for software as a medical device and AI/ML lifecycle expectations remains on FDA's Artificial Intelligence and Machine Learning in Software as a Medical Device page.
Related NovaPharma coverage
Frequently Asked Questions
Do patients trust hospitals to use AI responsibly?
In a national JAMA Network Open survey, 65.8% of U.S. adult respondents reported low trust that their health system would use AI responsibly, and 57.7% reported low trust that the system would ensure an AI tool would not harm them.
How fast are U.S. hospitals adopting generative AI?
A JAMA Network Open analysis of 2,174 nonfederal acute-care hospitals found 31.5% already using generative AI with electronic health records by 2024, while another 24.7% planned adoption within a year.
What increases patient acceptance of medical AI?
A 2026 JAMA Network Open conjoint study of 3,000 U.S. adults found trust and choice rose with better AI performance, FDA approval, national or local certification, clinician presence, and representative training data.
Primary Sources
Sources & references 1 primary sources
Sources verified at publication. See our editorial policy and data sources.
This article follows our editorial standards. Report a correction via editorial contact.
