Overview of the HHS ONC HTI-1 Final Rule on AI Transparency in Clinical Software
The Health and Human Services (HHS) Office of the National Coordinator for Health Information Technology (ONC) issued the Health Data Technology and Interoperability Certification Program Updates final rule, known as HTI-1, on January 9, 2024. This rule marks the first federal requirement in the United States mandating transparency for artificial intelligence (AI) systems embedded in certified electronic health record (EHR) software. The rule applies specifically to predictive decision support tools, which are AI-driven functionalities designed to assist healthcare providers in clinical decision-making. Since January 2025, certified EHR vendors must comply with these transparency requirements.
The rule establishes a framework for disclosing critical information about AI algorithms used in clinical decision support. This includes details about the training data, validation processes, and fairness assessments of these predictive tools.
Key Requirements for AI Transparency in Predictive Decision Support Tools
- Disclosure of Training Data: Vendors must provide detailed information about the datasets used to train AI models.
- Validation and Testing Information: Vendors are required to disclose the methodologies used to validate and test their AI models. This includes information about the types of validation studies conducted, the performance metrics evaluated, and the real-world testing environments.
- Fairness and Bias Assessments: The rule mandates that vendors assess and disclose potential biases in their AI models. This includes evaluating fairness across different demographic groups, such as age, race, ethnicity, and gender, to ensure that the tool does not disproportionately disadvantage any subgroup. Vendors must also provide mitigation strategies if biases are identified.
These requirements are designed to foster trust in AI-driven clinical decision support by ensuring that the underlying algorithms are scrutinizable and accountable. The rule applies to all certified EHR systems that incorporate predictive decision support tools, which are increasingly used for tasks such as diagnosing conditions, recommending treatments, and predicting patient outcomes.
Implementation Timeline and Compliance
This timeline allows certified EHR vendors sufficient time to update their systems, documentation, and internal processes to meet the new transparency standards. Vendors must ensure that their predictive decision support tools comply with the disclosure requirements before the effective date to maintain their certification status under the ONC’s Health IT Certification Program.
The rule does not impose additional certification criteria but instead amends existing certification requirements to include algorithm transparency. Vendors must integrate these disclosures into their certification applications and ongoing compliance reporting.
Scope and Applicability of the Rule
The HTI-1 final rule applies to all certified EHR technology developed by ONC-approved certification bodies. This includes software used in hospitals, clinics, and other healthcare settings where clinical decisions are influenced by AI-driven recommendations. The rule does not extend to non-certified EHR systems or AI tools used outside of certified EHR environments, though it may set a precedent for broader regulatory expectations in the future.
Predictive decision support tools covered under this rule include, but are not limited to, AI models used for:
By requiring transparency in these tools, the rule aims to mitigate risks associated with AI-driven clinical decisions, such as misdiagnosis, inappropriate treatment recommendations, or disparities in care due to algorithmic bias.
Potential Impact on Healthcare Providers and Patients
- Enhanced Trust in AI Tools: Healthcare providers will have greater confidence in AI-driven recommendations if they can independently verify the training data, validation processes, and fairness assessments. This transparency reduces the risk of blind reliance on AI systems without understanding their limitations.
- Improved Patient Outcomes: By ensuring that AI tools are thoroughly evaluated for bias and accuracy, the rule aims to reduce disparities in healthcare delivery.
- It establishes a baseline for algorithmic transparency that could influence future regulations in the sector.
- Vendor Accountability: EHR vendors will need to invest in documentation and compliance processes to meet the new requirements. This may lead to higher development costs but could also incentivize the creation of more transparent and reliable AI tools in the long term.
The rule does not mandate specific technical standards for AI models but instead focuses on disclosure. This approach allows flexibility for vendors to innovate while ensuring that critical information about their AI systems is accessible to stakeholders.
Challenges and Considerations for Compliance
While the HTI-1 final rule introduces important transparency requirements, several challenges may arise for EHR vendors and healthcare providers:
- Resource Intensiveness: Complying with the rule may require significant resources, including time and expertise, to document training data, validation processes, and fairness assessments.
- Interoperability and Standardization: The rule does not specify standardized formats for disclosing algorithmic information. Vendors may need to develop or adopt common frameworks to ensure that transparency reports are consistent, comparable, and usable by healthcare providers and regulators.
Despite these challenges, the HTI-1 final rule represents a critical step toward responsible AI deployment in healthcare. By prioritizing transparency, the rule helps to address ethical, legal, and clinical concerns associated with AI-driven clinical decision support.
Required actions
- Developers of certified health IT must make "source attributes" of predictive decision support interventions available to users: training data, fairness, validation and performance information - Maintain intervention risk management practices covering validity, reliability, safety, security and fairness, and publish summaries