#90. Standards, Context and Trust

This week’s newsletter explores three stories that are closely connected. HIQA is consulting on new national standards for patient summaries and e-prescribing in Ireland, OpenAI is bringing generative AI directly into clinical workflows through EHR integration, and NHS watchdogs are highlighting the risks of errors in AI-generated clinical documentation.

Together, they reinforce an important lesson: the future of healthcare AI will depend on trusted data, interoperability, strong governance and clear clinical accountability. As AI moves closer to the point of care, standards, context and trust are becoming strategic priorities, not technical considerations.

HIQA consults on the foundations of Ireland’s digital health future

This week the Health Information and Quality Authority (HIQA) launched a public consultation on two new Irish national health information standards: a National Standard for a Patient Summary and a National Standard for Electronic Prescriptions and Electronic Dispensations.

The proposed standards aim to ensure that critical health information is captured and exchanged consistently across healthcare settings. The patient summary standard defines the core information that should be available at the point of care, while the e-prescribing standard sets out how medication information should be recorded and shared between prescribers and dispensers.

Ireland’s digital health ambitions depend on information being able to follow the patient across organisational and system boundaries. A common patient summary will give clinicians a consistent view of essential information at the point of care, while standardised prescribing and dispensing information will strengthen the continuity and safety of medicines management.

Publishing national standards is an important step, but the real test will be adoption. The organisations that will ultimately implement these standards, particularly clinicians, pharmacists, GPs and technology suppliers, should engage now to ensure they can be embedded into supplier roadmaps, system design and frontline clinical workflows.

OpenAI takes AI closer to the clinical frontline

OpenAI recently announced new capabilities that allow healthcare organisations to connect EHRs and selected healthcare data sources directly to ChatGPT. The release includes integration with Epic, enabling clinicians to review authorised patient information within ChatGPT, alongside a new healthcare public data plugin that provides access to sources such as PubMed, ClinicalTrials.gov and medication reference databases.

Instead of clinicians manually searching across notes, laboratory results, medication lists etc AI can help bring relevant information together, summarise what has changed and highlight areas requiring attention. OpenAI positions this as a way to reduce the administrative and cognitive burden of navigating increasingly complex patient records.

Until now, many generative AI tools have operated largely outside core clinical systems. By moving directly into the EHR workflow, AI shifts from being a general productivity tool to becoming part of how clinicians access and interpret patient information

The potential benefits are clear, however, the challenges are equally significant.

Summarisation is not the same as understanding. Healthcare organisations will need confidence that important information is not omitted, misinterpreted or presented without appropriate context. Even if AI achieves impressive accuracy rates, clinicians remain accountable for decisions and will need to be able to verify how conclusions were reached.
There is the question of trust. Healthcare has spent decades building processes around clinical governance, auditability and evidence-based decision making. AI systems need to fit within those frameworks rather than bypass them.
If AI becomes the primary interface through which clinicians access information, interoperability becomes even more important, not less. AI is most valuable when it can draw on comprehensive, high-quality data from across the care pathway. Fragmented records, inconsistent standards and poor data quality remain barriers that no AI model can fully solve on its own.

AI scribes promise to give clinicians time back, but errors are testing trust

Healthwatch England has warned that AI scribes can introduce potentially harmful errors into patient records. Reported cases include changing “null demyelination” to “demyelination”, confusing two similarly named medicines, and omitting an instruction about a repeat prescription. In each case, the patient spotted the error rather than the clinician.

The potential benefit remains significant. AI scribes could reduce documentation time and allow clinicians to focus more fully on patients. But these examples reinforce the challenge raised by OpenAI’s move to connect AI directly to EHR data: the closer AI gets to the clinical record, the greater both its potential value and the consequences of a subtle, plausible error.

“Human in the loop” is not enough if clinicians lack the time to check every word against the source. Healthcare organisations will need traceability, realistic verification workflows, clear accountability and rapid correction processes. Patients can provide an additional safety net, but they should not become the final quality-assurance function for AI-generated records.

Originally published on LinkedIn.