ONC HTI-1 Rule: Unlocking AI Transparency for Health IT Investors

The era of opaque artificial intelligence making critical decisions in healthcare is rapidly drawing to a close. For Health IT Product Managers and Compliance Officers, the Office of the National Coordinator for Health Information Technology (ONC) has ushered in a new mandate for transparency, particularly for predictive algorithms embedded in clinical decision support (CDS) tools, through its HTI-1 Final Rule. This critical regulation demands a fundamental shift in how AI-powered health tools are developed, documented, and deployed, moving beyond mere functionality to rigorous explainability.

The End of Black-Box AI in Clinical Systems Under New Federal Rules

The ONC HTI-1 Final Rule, officially known as the Health Data, Technology, and Interoperability: Certification Program Updates, Algorithm Transparency, and Information Blocking Revisions, marks a key moment for AI in healthcare. Effective December 31, 2024, for specific provisions, this rule significantly expands the scope of the ONC Health IT Certification Program. It directly addresses the growing concern that AI algorithms, while powerful, often operate as “black boxes,” making decisions without clear, interpretable rationale. For any Health IT developer, particularly those building or integrating AI-driven CDS, understanding and implementing these transparency requirements is not merely good practice but a regulatory imperative. The rule aims to help clinicians with the necessary context to critically evaluate and appropriately use algorithmic insights, thereby enhancing patient safety and reducing health disparities that can arise from biased or poorly understood AI.

Deconstructing DSI Source Attribute Requirements and Information Blocking Provisions

At the heart of the HTI-1 rule’s algorithmic transparency initiative are the stringent requirements for Decision Support Interventions (DSIs), especially those that use predictive algorithms. The ONC defines a DSI as “interventions designed to be presented to a person, or to their agent, to help them make decisions about health care.” The rule mandates that certified health IT modules providing DSIs, particularly those that are “predictive,” must enable users to access specific “source attributes” at the point of care. These attributes are designed to shed light on the algorithm’s methodology, data provenance, and potential limitations. Specifically, for predictive DSIs, the HTI-1 rule requires the following source attributes to be made available, encompassing 31 attributes across 9 categories:

• Details and output of the intervention: Including the name and contact information for the developer, funding sources, and a description of the output produced (e.g., prediction, classification, recommendation).
• Purpose of the intervention: The intended use, target patient population, and intended user of the DSI.
• Cautioned out-of-scope use: A description of tasks, situations, or populations where the user is cautioned against applying the intervention, including known limitations and biases.
• Intervention development details and input features: A clear explanation of the data elements and types of information the algorithm uses, including demographic representativeness of training data and the underlying methodology.
• Process used to ensure fairness in development: A description of the approach taken to reduce or eliminate bias.
• External validation process: References to scientific literature or clinical studies that support the algorithm’s validity and effectiveness, and a description of any external validation.
• Quantitative measures of performance: Relevant metrics demonstrating the algorithm’s accuracy, reliability, and fairness, ideally across different demographic groups.
• Ongoing maintenance of intervention implementation and use: Details on how the validity and fairness of the DSI are monitored over time.
• Update and continued validation or fairness assessment schedule: The schedule for updates and ongoing validation or fairness assessments.

These requirements are not merely theoretical. They are directly tied to the information blocking provisions of the 21st Century Cures Act. Failure to provide these source attributes, especially when requested by a healthcare provider or patient, could be deemed information blocking. The HHS Office for Civil Rights (OCR) coordinates enforcement of these provisions, underscoring the severity of non-compliance. This means that health IT developers must not only generate this information but also ensure it is readily accessible and understandable within the clinical workflow, avoiding any practices that could impede its availability.

A Step-by-Step Checklist for Updating Software Documentation and User Interfaces

For Health IT Product Managers and Compliance Officers, translating these regulatory mandates into actionable steps requires a systematic approach. The compliance deadline for HTI-1 transparency requirements for DSIs is December 31, 2024, emphasizing the urgency of these updates. Here is a practical checklist to guide your efforts:

1. Inventory and Categorize All AI-Powered DSIs:

• Identify every instance of a predictive algorithm within your certified health IT modules that qualifies as a DSI.
• Distinguish between “predictive” and “non-predictive” DSIs, as the full source attribute requirements apply primarily to the former.

2. Develop or Enhance Internal Documentation for Each Predictive DSI:

• For each identified predictive DSI, compile all the required source attributes (identification, developer, version, update date, inputs, outputs, methodology, evidence, performance, limitations).
• Ensure this documentation is clear, complete, and kept up-to-date with every algorithm iteration.
• Collaborate with your data science and engineering teams to accurately capture the technical details in an understandable format for clinicians.

3. Integrate Source Attribute Display into User Interfaces:

• Design user interface (UI) elements within your health IT module that allow clinicians to easily access the source attributes at the point of care. This could be via an “info” icon, a dedicated panel, or a hover-over feature.
• Prioritize conciseness and clarity in the UI presentation, offering more detailed information via drill-down options or links to complete documentation. The American Medical Association (AMA) has consistently advocated for transparent and understandable AI in clinical practice, aligning with this need for intuitive presentation. AMA guidance on AI in medicine
• Test the usability of these UI elements with target users to ensure they are intuitive and do not disrupt clinical workflows.

4. Review and Update Software Development Life Cycle (SDLC) Processes:

• Embed the generation and documentation of source attributes into your standard SDLC for all new and updated predictive DSIs.
• Implement version control for algorithms and their associated documentation to track changes over time.
• Establish clear roles and responsibilities for maintaining algorithm transparency information.

5. Address Information Blocking Considerations:

• Ensure that your health IT module’s design and your organizational policies do not create barriers to accessing these source attributes.
• Train your support staff and product teams on the information blocking regulations as they pertain to algorithmic transparency.

6. Prepare for Certification and Audits:

• Confirm that your certified health IT module meets all the technical and functional requirements for algorithmic transparency under HTI-1.
• Maintain strong records of your compliance efforts, as these will be critical during ONC certification and potential audits.

The ONC HTI-1 rule represents a significant leap forward in ensuring the responsible and ethical deployment of AI in healthcare. By proactively addressing these transparency requirements, Health IT Product Managers and Compliance Officers can not only ensure regulatory adherence but also build greater trust and confidence in the AI tools that are increasingly integral to modern clinical practice.

Methodology and Source Note

This framework and analysis are directly based on the official text of the ONC HTI-1 Final Rule (45 CFR Parts 170 and 171), as published in the Federal Register, and corroborated by official ONC fact sheets pertaining to Decision Support Interventions and algorithmic transparency. All data points regarding compliance deadlines and specific source attribute elements have been verified against these authoritative sources. Federal Register publication of ONC HTI-1 Final Rule