Artificial intelligence is becoming increasingly common in US hospitals. Nearly half of hospitals surveyed in 2023–2024 reported using AI-based predictive models — but adoption is not evenly distributed across the country. Some regions and health systems are moving quickly, while others — particularly those in healthcare shortage areas — are adopting more slowly.

These findings come from “The Landscape of AI Implementation in US Hospitals,” led by Yeon-Mi Hwang and colleagues and published in Nature Health in 2026.1 The study analyzes data from more than 3,500 hospitals nationwide and maps where predictive AI tools are being implemented — and where they are not.

At first glance, this may seem like a technology adoption story. In reality, it is also a data story.

As healthcare increasingly relies on real-world data (RWD) for research, regulatory decisions, safety monitoring, and value-based payment models, the way hospitals adopt AI could directly influence the quality and coverage of the data being produced across the United States.

AI adoption signals digital maturity

Hwang and colleagues found that interoperability — the ability of hospital systems to exchange and integrate data — was the strongest predictor of AI adoption. Hospitals with better health information exchange capabilities and fewer data-sharing barriers were much more likely to implement predictive AI tools.

This matters because AI systems require structured, standardized, and well-integrated data to function effectively. When hospitals invest in AI, they often strengthen their documentation practices, data governance, and system integration in the process. Those same improvements elevate the overall quality of clinical data.

In other words, hospitals that are ready for AI are often also ready to produce higher-quality RWD.

Why high-adoption regions may produce richer RWD

Predictive AI systems frequently generate structured outputs such as risk scores, alerts, and time-stamped predictions. These outputs are recorded in electronic health records and become part of the clinical data landscape.

As a result, regions with higher AI adoption may generate data that is more complete, more standardized, and better linked across care settings. Their records may contain clearer severity markers, earlier detection signals, and more consistent documentation of clinical decision points.

This is why high-adoption regions may produce richer RWD. The data is not only documented — it is more granular and more measurable.

Because the study shows that AI adoption clusters geographically, these differences in data richness may also cluster by region.

The geography gap

One of the more striking findings in the study is that hospitals in healthcare shortage areas and medically underserved regions were less likely to adopt predictive AI. These areas often include rural and resource-constrained institutions.

If these hospitals have less advanced digital infrastructure, the data they generate may be more fragmented and less standardized. Over time, this could create meaningful differences in data coverage across the country. Regions with strong AI adoption may produce deeper, more analyzable datasets, while underserved areas may remain underrepresented in national RWD pipelines.

That imbalance could influence which populations are most visible in research and regulatory evidence.

AI changes the shape of the data

AI adoption does not simply improve data capture — it can also shape how care is delivered and recorded. Predictive systems may trigger alerts, influence documentation patterns, and alter clinical workflows. These changes become embedded in patient records.

As a result, RWD from high-adoption environments may reflect AI-influenced care pathways, while RWD from lower-adoption settings reflects more traditional workflows. Differences in adoption may therefore create differences not only in data volume, but also in data structure and interpretation.

Why this matters for real-world evidence

Real-world data increasingly underpins post-market surveillance, comparative effectiveness research, regulatory decision-making, and value-based care arrangements. If richer, more granular data clusters in digitally advanced regions, then the evidence generated from national datasets may disproportionately reflect those environments.

This is not necessarily intentional. It is a structural consequence of uneven infrastructure development. But without attention to digital equity, disparities in AI adoption could gradually translate into disparities in evidence generation.

The bottom line

The nationwide analysis by Yeon-Mi Hwang and colleagues offers one of the clearest early views of how AI is spreading across US hospitals. Because AI adoption is closely tied to interoperability, digital maturity, and institutional capacity, it likely influences how real-world data is captured, structured, and represented.

High-adoption regions may produce richer RWD — data that is more complete, more granular, and better connected across care settings. At the same time, uneven adoption raises important questions about representativeness and equity in national datasets.

Understanding how AI adoption is expanding — and where it remains limited — may become a key factor in strengthening the US data ecosystem. If increasing AI adoption leads to more complete and structured RWD, it could significantly enhance the power and reliability of real-world evidence. But ensuring that this digital maturity is broadly distributed will be essential. Otherwise, the strength of future RWE may reflect infrastructure patterns as much as clinical reality.

As AI becomes more embedded in healthcare, how and where it is implemented may quietly shape not only care delivery — but the evidence base that guides it.

Share Via