The global health data ecosystem is undergoing its most significant transformation since the adoption of electronic health records (EHRs) two decades ago. Driven by new federal rules, explosive growth in wearable sensors, and the maturation of large language models, stakeholders from payers to providers to patients are renegotiating who owns, accesses, and monetizes clinical information. This week’s developments signal that 2025 will be remembered not as a year of incremental updates, but as a structural pivot toward real-time, patient-mediated data exchange—with generative AI acting as both catalyst and complication.

Federal Interoperability Rules Reach Final Implementation Phase

On March 3, the U.S. Department of Health and Human Services confirmed that the final enforcement deadlines for the 21st Century Cures Act’s information-blocking provisions now apply to all health IT developers, including those offering predictive decision-support tools. The rule, which took full effect on March 31, requires that API-based access to patient data—including clinical notes, lab results, and medication lists—be available without “unreasonable” fees or delays. Crucially, the Office of the National Coordinator for Health IT (ONC) has clarified that AI-generated summaries derived from EHR data fall under the same access requirements.

“We are seeing the first enforcement actions against legacy vendors that still charge per-API-call fees,” said Dr. Elena Vasquez, a health policy researcher at the University of Michigan. “The market is shifting from data silos to data utilities. The winners will be platforms that treat interoperability as a baseline, not a premium feature.”

Vasquez’s observation aligns with new market data: a February survey from the Healthcare Information and Management Systems Society (HIMSS) found that 68% of hospitals now use FHIR-based APIs for patient access, up from 41% in 2023. However, the same survey flagged a persistent gap—only 22% of those hospitals allow patients to write or correct their own data, a feature that patient advocacy groups argue is essential for accuracy.

Generative AI: From Pilot to Production—But With New Governance Demands

The most volatile segment of health data news this quarter involves generative AI’s role in clinical documentation and decision support. In January, Epic Systems announced that its AI Scribe tool, which drafts clinic notes from ambient conversations, now processes over 15 million patient encounters per month. Meanwhile, Google Health’s Med-PaLM 3, released in February, demonstrated a 92% accuracy rate on a benchmark of 1,200 consumer health questions—but also produced a 3.4% rate of “hallucinated” medication interactions that required clinician correction.

This dual reality—high utility, measurable risk—has prompted regulators to act. The FDA’s Digital Health Advisory Committee, meeting in late February, recommended that generative AI tools used for diagnostic support be classified as Class II medical devices, requiring premarket notification. More notably, the committee proposed a “human-in-the-loop” mandate: any AI-generated summary that influences a treatment decision must display a clear audit trail, including the source patient data and the model version used.

“The era of silent AI is over,” said Dr. Marcus Chen, chief medical information officer at Stanford Health Care. “We are now asking vendors to expose confidence scores and source citations for every output. That is a massive engineering lift, but it is the only way to maintain clinical trust.”

Chen’s point is echoed by a recent JAMA Internal Medicine study, which found that physicians who used AI-generated chart summaries saved an average of 11 minutes per patient encounter—but also made 8% more documentation errors when they did not verify the AI’s output. The study’s authors call for mandatory “verification prompts” in EHR interfaces, a recommendation already adopted by several large health systems.

Wearables and Remote Monitoring: The Data Volume Explosion

On the consumer side, the health data deluge continues to accelerate. Global shipments of smartwatches and fitness bands with health sensors reached 210 million units in 2024, according to IDC, and these devices now capture continuous heart rate, sleep stage, blood oxygen, and—in newer models—blood pressure and electrodermal activity. Apple’s Health Records API, which allows third-party apps to pull clinical data from participating hospitals, saw a 300% increase in active users last year.

The challenge is no longer data collection but data integration. A report published this week by the Digital Health Collaborative found that the average patient-generated health data (PGHD) stream has a 31% missing-data rate due to device non-adherence, sensor battery failures, or user error. This unreliability makes PGHD difficult to use for formal clinical decision-making, yet insurers are increasingly requiring it for chronic disease management programs.

“We are at a fork in the road,” said Priya Raman, senior analyst at Forrester Research. “Either we build standards for PGHD quality—like the IEEE 2700 series for wearable sensor accuracy—or we risk a tidal wave of noisy, legally risky data entering the medical record. The FDA’s recent guidance on software as a medical device (SaMD) is a start, but it does not cover consumer fitness devices.”

Raman points to the European Union’s new European Health Data Space (EHDS) regulation, which took preliminary effect in January, as a potential model. EHDS mandates that all health data—including PGHD—be stored in a standardized, machine-readable format, and it grants patients the right to port their data between providers and apps without fees. While the U.S. has no equivalent federal law, several states, including California and Colorado, are drafting similar patient data rights bills.

The Patient-Controlled Data Economy Emerges

Perhaps the most consequential trend is the rise of “data wallets” and personal health data intermediaries. Startups such as HumanFirst and Vault Health now offer platforms where patients aggregate their EHRs, genomic tests, pharmacy records, and wearable data into a single consent-based profile. These platforms allow patients to grant temporary, revocable access to researchers or pharmaceutical companies—in exchange for direct compensation or free services.

In February, a consortium of 14 academic medical centers announced a pilot program using such wallets to recruit patients for a large-scale longitudinal study on post-COVID conditions. The program reported a 74% enrollment retention rate over 60 days, compared to a 51% average for traditional recruitment. The key differentiator: patients could see exactly which data fields were being shared and could withdraw specific data points (e.g., mental health notes) without leaving the study.

“This is a fundamental shift from ‘data donation’ to ‘data licensing,’” said Dr. Sarah Lindqvist, a bioethicist at Johns Hopkins. “But it raises serious equity questions. Wealthier, more tech-savvy patients will monetize their data, while vulnerable populations may be excluded or coerced by payers into sharing. We need guardrails against data-based discrimination.”

Lindqvist’s concern is not hypothetical. A 2025 report from the National Consumer Law Center documented five cases where life insurers denied coverage based on algorithmically scored health data purchased from third-party brokers—data that included wearable step counts and sleep patterns. The report urges the Federal Trade Commission to apply the Fair Credit Reporting Act to health data brokers, a move that industry groups have resisted.

Standardization Efforts and the Race for a Global Data Language

On the technical front, the most significant development is the convergence of HL7 FHIR R6, now in final trial use, with the International Patient Summary (IPS) standard. This convergence allows a patient’s core health record—allergies, medications, problems, procedures—to be encoded in a single, portable file that any FHIR-compliant system can render. In January, the Global Digital Health Partnership, representing 38 countries, endorsed IPS as the baseline for cross-border emergency care.

Meanwhile, the open-source community has made strides with the “OpenHealthData” project, which provides a free, de-identified synthetic dataset of 10 million patient records for AI training. The project, launched by a coalition of universities and non-profits, aims to reduce the reliance on proprietary, biased datasets that have plagued medical AI development. Early results show that models trained on OpenHealthData perform comparably to those trained on proprietary datasets, while exhibiting 40% less racial bias in cardiovascular risk prediction.

Expert Outlook: Interoperability Is Not Enough

As the industry absorbs these changes, consensus is forming around a sobering truth: technical interoperability, while necessary, is insufficient to deliver the promised benefits of health data. The missing pieces are governance, trust, and workflow integration.

“We can move files between systems, but we still can’t move meaning,” said Dr. Vasquez. “A lab value from a hospital in rural Montana does not carry the same context as the same value from a quaternary academic center. We need semantic interoperability—shared ontologies for social determinants, for patient-reported outcomes, for clinical reasoning.”

Vasquez and other experts point to a promising pilot from the Office of the National Coordinator: the “Trusted Exchange Framework and Common Agreement” (TEFCA), which now has 12 participating networks covering 90% of U.S. hospitals. Under TEFCA, health information exchanges are required to support “reasonable” patient access and to document every data sharing event for security audits. The next phase, expected in late 202