Hospital AI Stalls at Integration, Not Accuracy, as Epic Opens Its EHR to Outside Apps

Hospitals' main obstacle to scaling artificial intelligence is shifting from proving that models work to fitting them into daily clinical workflow. A survey organised by Carta Healthcare and reported on 7 August by Healthcare IT News found that EHR integration difficulty was the most-cited barrier to AI adoption, named by 44% of respondents. On 21 October, Epic Systems announced a tool aimed at that problem.

Adoption is already widespread

Federal data show AI is well established in US hospitals. According to a data brief from the Assistant Secretary for Technology Policy (ASTP), based on the American Hospital Association's survey, 71% of non-federal acute care hospitals used predictive AI integrated into their electronic health record (EHR) in 2024, up from 66% in 2023.

Most of it came from EHR makers. In 2024, 80% of hospitals used predictive AI from their EHR developer, 52% used third-party tools and 50% built their own. Use also varied by vendor: 90% of hospitals on the market-leading EHR used predictive AI, against 50% of those on other systems.

A gap separates large systems from small hospitals. The AHA reported that 86% of hospitals in multi-hospital systems used predictive AI in 2024, against 37% of independent facilities.

Generative AI is less common. A study in JAMA Network Open, published in December 2025 and summarised by IntuitionLabs, found 31.5% of hospitals were current users and another 24.7% planned adoption within a year.

Accuracy is being checked

Hospitals do test their models. In 2024, 82% evaluated predictive AI for accuracy, 74% for bias and 79% ran post-implementation monitoring, the ASTP brief found. The brief noted that fewer hospitals did so for all or most of their models, and that 15% to 21% answered "don't know" on these questions.

Where pilots stall

The Carta survey found that 71% of organisations reporting measurable value from AI were still not expanding it at pace. EHR integration ranked well ahead of clinician trust and regulatory concerns, which were each cited by 26%. Carta chief executive Brent Dover told Healthcare IT News that a tool can perform well in a demonstration yet fail once clinicians must leave their workflow to use it.

The survey also found clinical leaders were the most common owners of AI strategy, while 26% of respondents had no clearly defined owner. Carta did not publish the sample size in the article, and the company sells clinical data-management services.

A separate report from Qventus, based on a survey and interviews with more than 60 senior health IT leaders at medium and large systems, found 74% named dependence on EHR vendors as a top execution barrier, and 45% struggled to move past pilots, Healthcare IT News reported in April. Respondents also said managing several third-party AI implementations created bottlenecks.

Epic's response

On 21 October, at the start of its interoperability conference in Verona, Wisconsin, Epic announced Connectors. The capability lets developers and healthcare organisations link outside applications to the EHR, importing data into the record and letting the EHR complete tasks in other apps.

Connectors will work with Epic's AI agents, including Emmie for patients, Art for clinician charting and Penny for revenue cycle and operations, and with custom agents built in its Agent Factory tool, Healthcare IT News reported. Derek De Young, who leads Agent Factory development, said the feature should reduce "swivel-chair" work, meaning switching between apps.

Wayfinding vendor Gozio Health, logistics firm VPL and revenue-cycle company Availity said they are developing Connectors. Health systems will be able to source developer-built Connectors from Epic's website beginning in March 2027, and will be able to review what each one can read and change, according to Epic.

What is not established

Most of the survey data comes from vendors with products to sell, and the samples are small or undisclosed. The surveys also define "use" differently, so figures cannot be compared directly. The federal data cover 2024 and rely on hospitals' self-reports.

None of the sources reviewed shows that better integration improves patient outcomes or that Connectors will ease the problems respondents describe. Epic's announcement is a company statement, and the product is not yet available. Whether opening an EHR to third-party tools reduces the vendor dependence that Qventus respondents reported is an open question.

What to watch next

The first test is March 2027, when Epic says developer-built Connectors become available to health systems. Newer federal data on hospital AI use would show whether the gap between independent hospitals and large systems is narrowing. HIMSS27, scheduled for 5 to 8 April 2027 in Chicago, is likely to feature further announcements. Independent studies linking integration quality to clinical or financial results would give hospital leaders firmer evidence than vendor surveys do.