---
title: "The future of AI and interoperability in healthcare — Datagist"
description: "Hospital exchange and AI adoption figures for 2026, the CMS payer API and prior authorization deadlines, and what health systems should build next."
url: https://datagist.ai/insights/future-of-ai-and-interoperability-in-healthcare
---

[Insights](https://datagist.ai/insights) Point of view

# The future of AI and interoperability in healthcare

Hospitals now exchange records at national scale and most run predictive AI inside the EHR. What the latest numbers show, which deadlines land next, and how to build AI that can use data from outside your own walls.

Datagist November 3, 2026 8 min read

[Healthcare](https://datagist.ai/industries/healthcare) [GenAI & Agents](https://datagist.ai/services/genai-agents) [Modern Data](https://datagist.ai/services/modern-data) [Governance & Operations](https://datagist.ai/services/governance-operations)

For most of the last decade, healthcare interoperability meant getting records from one system to another. A discharge summary reached the primary care office. A lab result showed up in the right chart. AI was a separate conversation, mostly about pilots and risk scores.

Those two conversations have merged. Hospitals now exchange records at national scale, and most of them run predictive models inside the EHR. The models can only work with the data they can reach, and the exchange networks are starting to carry more data than people have time to read. Each now depends on the other.

This report looks at where both stand at the end of 2026, using public figures from ONC, ASTP, HHS, CMS and the AMA. It then covers the deadlines that land next and what we think health systems and payers should build for.

## The numbers at a glance

| Measure                                                                                 | Figure                                                             | Source |
| --------------------------------------------------------------------------------------- | ------------------------------------------------------------------ | ------ |
| Hospitals engaged in all four interoperability domains (send, receive, find, integrate) | 70% in 2023, up from 46% in 2018                                   | ONC    |
| Hospitals that often or routinely exchange data, large vs small                         | 53% vs 38%                                                         | ONC    |
| Records exchanged through TEFCA                                                         | About 10 million in January 2025, more than 1 billion by June 2026 | HHS    |
| Hospitals using predictive AI integrated with the EHR                                   | 71% in 2024, up from 66% in 2023                                   | ASTP   |
| Predictive AI use, system-affiliated vs independent hospitals                           | 86% vs 37%                                                         | ASTP   |
| Hospitals that checked their models for bias                                            | 74%                                                                | ASTP   |
| Physicians using AI in their work                                                       | 38% in 2023, 66% in 2024, 81% in 2026                              | AMA    |

Two patterns stand out. Both exchange and AI use are growing fast. And in both, large and system-affiliated organizations are far ahead of small, rural and independent ones.

## Interoperability: the pipes are mostly built

ONC tracks four parts of interoperability: sending records, receiving them, finding them in outside systems and integrating them into the local chart. In 2023, 70% of non-federal acute care hospitals did all four at least sometimes, up from 46% in 2018\. The gap by size is wide. 53% of large hospitals exchange data often or routinely, against 38% of small hospitals.

National exchange grew even faster. TEFCA, the federal framework that connects health information networks to each other, carried about 10 million records in January 2025\. HHS reported nearly 500 million by February 2026 and more than one billion by June 2026\. The Social Security Administration joined in early 2026 so it can pull medical evidence for disability claims, a sign that exchange now reaches well past hospitals and clinics.

Integration is the weak step. Receiving a record and using it are different things. Much of what arrives is still a document, such as a C-CDA or a PDF, that a clinician has to open and read. Medications, problems and allergies from outside sources often sit beside the local chart instead of inside it, and the same patient can appear under slightly different names and identifiers in every feed.

## AI: already inside the EHR

ASTP’s 2025 data brief found that 71% of hospitals used predictive AI integrated with their EHR in 2024, up from 66% the year before. 80% of hospitals got at least some of those models from their EHR developer. 52% used third-party models and 50% built their own.

The growth is in operations. Hospitals using AI to automate billing rose from 36% to 61% in one year, and scheduling went from 51% to 67%. Clinical uses such as predicting inpatient risk were already common and stayed that way.

Clinicians have moved just as quickly. The AMA found that 38% of physicians used some form of AI in 2023 and 66% in 2024\. Its 2026 survey put the figure at 81%. Ambient documentation tools, which draft the visit note from the conversation, are a large part of that jump.

Oversight has not kept pace everywhere. In 2024, 82% of hospitals using predictive AI checked models for accuracy and 74% checked for bias. Among independent hospitals the figures were 41% and 31%. Many smaller hospitals run models they did not build and have not tested on their own patients.

[Timeline of US healthcare exchange and AI from 2023 to 2027\. Exchange and policy: in 2023, 70% of hospitals send, receive, find and integrate records. In January 2024 a CMS final rule sets payer API and prior authorization deadlines. In January 2025 about 10 million records are exchanged through TEFCA. In early 2026 payers must decide prior authorizations in 72 hours or 7 days with a reason for every denial, and TEFCA reaches 500 million records. By mid 2026 TEFCA passes one billion records and the Social Security Administration pulls disability evidence through it. On January 1, 2027 payer FHIR APIs are due. AI adoption: hospitals running predictive AI in the EHR rise from 66% in 2023 to 71% in 2024, 74% of hospitals using AI test models for bias, and physicians using AI rise from 38% in 2023 to 66% in 2024 and 81% in 2026\. Next, AI agents call the same FHIR APIs at the point of care.](https://datagist.ai/media/img/9ba131a36940487afed791f4f43064c1fc8a5c4b-2160x1022-96b27e96f7.webp)

## Why AI needs interoperability

A model is only as good as the record it reads. A sepsis score that can’t see the antibiotics a patient got at an urgent care clinic yesterday will misjudge risk. A readmission model that only knows about visits inside one health system will miss the patients who go elsewhere. An ambient scribe can write a clean note about today’s visit, but it can’t reconcile medications it never received.

This is where exchange and AI meet. The outside data now arrives through TEFCA, HIEs and payer APIs. If it lands as documents that only a person can read, the models never see it. If it lands as structured, coded and matched data, every model in the EHR gets a fuller picture of the patient without being retrained.

The reverse also holds. Clinicians already struggle to read the outside records they receive. AI that summarizes a 200-page transfer packet, flags the medication that changed, or pulls the last echo result out of a scanned letter turns exchanged data into something a clinician can act on in a short visit. Without that help, a billion records exchanged is a billion records unread.

## What changes next

The biggest near-term change comes from payers. CMS’s Interoperability and Prior Authorization final rule applies to Medicare Advantage plans, Medicaid and CHIP programs and their managed care plans, and qualified health plans on the federal exchanges. Since January 2026, most of these payers have had to decide urgent prior authorization requests within 72 hours and standard ones within seven calendar days, and give a specific reason for each denial. They had to post their prior authorization metrics publicly by March 31, 2026.

By January 1, 2027, the same payers must run FHIR APIs that let providers pull a patient’s claims and clinical data, let a new plan pull data from the old one, and handle prior authorization from inside the EHR. The rule names the standards: FHIR R4, US Core, SMART App Launch, Bulk Data and the Da Vinci guides for coverage requirements, documentation and prior authorization.

For AI, that matters in two ways. Prior authorization is one of the most manual, document-heavy processes in healthcare, and it is about to become an API call with coded questions and coded answers. That makes it a strong candidate for automation on both sides. It also means payer data about a patient, including claims from other providers, becomes reachable through a standard interface at the point of care.

Other shifts are already under way.

**FHIR inside the exchange networks.** TEFCA started with document exchange. FHIR-based exchange is being added, which means networks can return specific resources such as a medication list instead of a whole document. That is the form AI tools can use directly.

**Agents as API clients.** AI agents are starting to call FHIR APIs themselves, often through tool protocols such as the Model Context Protocol. An agent that drafts a prior authorization request can look up the patient’s diagnoses, recent imaging and the payer’s coverage rules in one task. Each of those calls needs the same authorization, scopes and audit trail a human user would get.

**Transparency rules for clinical AI.** ONC’s HTI-1 rule introduced requirements for certified EHRs to show users where a decision support model came from, what data it was trained on and how it was tested. Federal policy on AI keeps shifting, but buyers now expect that information whether or not a rule requires it.

## Risks to plan for

Patient matching is still the quiet failure point. If two records for the same person don’t link, the model sees half the history. If two different people link by mistake, the model sees someone else’s history, which is worse. More exchange makes the problem bigger.

Consent and sensitive data need care. Records about substance use treatment, behavioral health and reproductive care carry their own rules. An AI tool that summarizes everything it can reach may surface information the clinician or the requesting purpose isn’t allowed to see. Data segmentation and purpose-of-use checks have to happen before the model reads the data.

Data quality travels. A wrong code in one system becomes a wrong code in every system that receives it, and then a wrong input to every model that reads it. Provenance helps here. Each fact should carry where it came from and when, so a model and a clinician can weigh a hospital lab result differently from a patient-reported value.

The gap between large and small organizations can widen. System-affiliated hospitals are more than twice as likely as independent ones to use predictive AI and far more likely to test it. If the next wave of AI depends on clean, structured, outside data, the hospitals that already have the integration work done will move faster again.

## What to build now

For health systems, the first job is turning incoming data into structured data. Parse C-CDA documents into FHIR resources, map local codes to standard terminologies such as LOINC, RxNorm and SNOMED CT, and reconcile outside medications, problems and allergies into the chart. Every model you run benefits from this, including the ones you buy.

Invest in identity. A master patient index with tuned match rules, regular review of possible duplicates and clear ownership of the review queue pays off in every AI project that follows. Measure your duplicate rate before you start, so you can show the change.

Keep provenance and consent with the data. Store the source, date and sensitivity label with each record as it lands. Then build AI tools that read through the same access controls clinicians use, so a model never sees data its user couldn’t.

Set up local model review. Test vendor and in-house models on your own patients before go-live, check them for bias across the groups you serve, and watch them after launch. Smaller hospitals can share this work through their HIE, a regional network or a peer group.

For payers, the January 2027 APIs are the base. Build them on clean member and provider data, because the Provider Access and Payer-to-Payer APIs will expose every data quality issue you have to the providers and plans on the other end. Then look at prior authorization rules: the coverage logic you encode for the API is the same logic an AI assistant can use to check a request before it is sent.

For both, start with one workflow where outside data changes a decision, such as medication reconciliation at admission or prior authorization for imaging. Measure how often outside data was available, how often it was used, and what changed. Use that result to decide where to go next.

## Sources

ONC, [70% of hospitals participate in healthcare interoperability](https://healthit.gov/news/70-hospitals-participate-healthcare-interoperability/) (2023 data). ASTP Data Brief No. 80, [Hospital trends in the use, evaluation, and governance of predictive AI, 2023-2024](https://www.healthit.gov/sites/default/files/2025-09/hospital-trends-use-evaluation-and-governance-predictive-ai-2023-2024.pdf) (September 2025). HHS, [ONC strengthens TEFCA as one billion health records exchanged](https://www.hhs.gov/press-room/onc-strengthens-tefca-one-billion-health-records-exchanged) (June 2026). CMS, [Interoperability and Prior Authorization final rule fact sheet](https://www.cms.gov/newsroom/fact-sheets/cms-interoperability-prior-authorization-final-rule-cms-0057-f) (January 2024). AMA, [2 in 3 physicians are using health AI](https://www.ama-assn.org/practice-management/digital-health/2-3-physicians-are-using-health-ai-78-2023) (February 2025) and [More than 80% of physicians use AI professionally](https://www.ama-assn.org/practice-management/digital-health/more-80-physicians-use-ai-professionally-ama-survey) (March 2026).

---

We help health systems and payers turn exchanged data into data their teams and their AI tools can use, from FHIR pipelines and patient matching to model governance. See our [Healthcare](https://datagist.ai/industries/healthcare) work, or [talk to us](https://datagist.ai/contact) about the workflow where outside data would change the most decisions for you.
