AI is rewriting the future of healthcare, but patients and clinicians aren’t moving at the same speed.
From breast cancer screening to virtual clinicians to health information — the transition is a bit uneven. Providers are using new tools at a quicker clip, but patients are growing uneasier about how their data is gathered, analyzed and protected.
AI Moves From Promise To Proven Cancer Screening Success
A long-running debate over AI’s value in diagnostics now has one of its strongest real-world data sets. A new study in Nature Health analyzed more than 579,000 mammography exams across 109 community imaging centers in California, Delaware, Maryland and New York. It is the largest U.S. study to date measuring how AI affects breast cancer screening in routine practice. It’s important to note that the study was conducted by DeepHealth, an AI-focused subsidiary of diagnostic provider RadNet.
Dr. Greg Sorensen, RadNet’s chief medical officer and a co-author of the study, said during a call that the evidence from the research speaks for itself. “RadNet does about as many mammograms as the entire country of England every year — we have real evidence. AI really is better than the humans, but it also makes the humans better.”
The results stand out. RadNet’s multistage AI workflow, which combines DeepHealth’s computer-aided detection with an AI-guided safeguard review, produced a 21.6% jump in cancer detection compared to standard 3D mammography, while positive predictive rates — an accuracy metric presented as a probability that a patient who tests positive actually has the disease — rose 15%. Recall rates, the percentage of individuals who are called back for further testing after an initial screening, increased only slightly and remained within clinical guidelines.
Given that tens of millions of mammograms occur each year in the U.S., performance gains of this size translate into thousands of earlier diagnoses. The improvements reached every racial and breast-density subgroup, including a 22.7% boost in detection among women with dense breasts, which are especially difficult to accurately screen.
Howard Berger, RadNet’s founder and CEO, said rising imaging complexity has made AI essential. “When a mammogram was done, there were somewhere between two and four images you had to look at. Now there’s well over 1,000 and that’s growing,” he said. “We see the movement of healthcare led by radiology to be more in screening tools to diagnose almost any disease earlier than ever was possible.”
Telemedicine Expands As AI Clinicians Begin Scaling Human Care
Diagnostics aren’t the only areas seeing increases in AI. Telemedicine is entering a new phase driven by virtual clinicians designed to ease pressure on stretched care teams. The American Nurses Foundation conducted a survey with McKinsey that showed 64% of nurses want more AI resources and training as job helps as they manage burnout, declining staff numbers and needier patients.
An example of such an AI tool would be Eleanor, a conversational AI clinician developed by MyndYou. Eleanor engages callers with proactive, empathetic phone conversations without requiring apps or logins, then summarizes what “she” learns during the calls directly into the various patient-care streams.
MyndYou shared in an email exchange that Eleanor has completed more than 2 million calls, with 89% of assessments handled without human intervention. Healthcare systems that have rolled out Eleanor have seen an average 33% drop in hospital admissions, 22% fewer 30-day readmissions and $5.9 million in annual savings from avoided ER visits and readmissions.
The company also recently announced its Epic integration, which now embeds Eleanor directly inside clinical workflows. MyndYou CEO Ruth Poliakine Baruchi stated, “Healthcare professionals can now activate next-generation care models to amplify outcomes and significantly reduce clinician burden.”
According to MyndYou’s website, Eleanor can place calls, flag risks, route concerns and deliver structured insights into the Epic solution, creating an automated loop. Clinicians are freer to work on challenging cases, while AI manages the upfront volume.
AI in Healthcare Is Rising, But Patient Trust In AI Is Not
Even though AI is becoming more common in everyday care delivery and diagnostics, patient confidence is eroding when it comes to the security of their personal health information and AI. A nationwide study from medical health records platform Tebra points to growing skepticism over privacy, governance, data safety and AI.
The numbers tell a striking story:
- Only 1 in 4 Americans would allow their provider to share health data with an AI-powered tech company
- 1 in 3 Americans have experienced a health-data breach in the past three years
- 41% of respondents say HIPAA rules are not keeping pace with current cybersecurity risks and AI advances
Kevin Marasco, chief growth officer at Tebra, said patients respond to clear outcomes, not tech labels. “When we positioned AI specifically for fraud prevention, we reached a 25% acceptance threshold among patients. Merely announcing the use of badges and AI-compliance assurances didn’t mean anything to patients. The key insight we learned is to lead with specific patient protection outcomes, not technology capabilities.”
Past Data Hacks Slows Patient AI Acceptance
Interestingly, the research showed that the strongest predictor of trust in current data systems was whether or not a patient went through an attack on their own data. Patients whose health information was never breached tended to believe healthcare hacks can be avoided, while actual victims were 1.5 times more likely to see breaches as inevitable.
Communication speed bumps are also slowing patient AI uptake. The study showed that one-third of healthcare data leak victims said that after the initial notice, they received no follow-up details from the organization that was supposed to protect their patient information.
Marasco said providers should move quickly on three priorities: faster staff training, updated security risk assessments and improved communications clarity after incidents happen. “Practices demonstrating competence and speedy resolutions will capture the 54% who think breaches are preventable,” he concluded in an email exchange.
When considered as elements of a larger healthcare whole, these trends suggest patients are slow to completely embrace AI and healthcare. AI’s spread into healthcare is speeding up, but improved training and communication, focusing on patient trust and thoughtful AI governance will determine how far that momentum carries.
