ONC’s Health Data, Technology, and Interoperability (HTI-1) final rule was enacted on March 11, 2024, and has since been the center of all news in healthcare IT. A particular section of the HTI-1 rule, Decision Support Interventions (DSI), has been the primary focus. In this blog, we will unpack what DSI is and what it means for various stakeholders in healthcare. Let's start with some basic history and terminology.

Health IT developers and EHRs have been required to incorporate Clinical Decision Support (CDS) functionalities within their EHR systems. CDS has been part of a certification criterion called 170.315(a)(9) or, in short, a9 since 2012. With HTI-1, the terminology Clinical Decision Support (CDS) is being transitioned to Decision Support Interventions (DSI) and will now be covered under a new certification criterion called 170.315(b)(11) or, in short, b11.

There are two reasons why DSI has been the focus of HTI-1.  

1.      The buzz around AI: This rule is the first set of regulations around using AI in healthcare through EHRs.  

2.      December 31, 2024 deadline: EHRs currently certified to the CDS a9 criterion must meet the DSI b11 criterion requirements and provide the health IT with the updated certification to their customers by the end of 2024.  

DSI criterion represents the first substantial revision to the certification requirements for clinical decision support-related capabilities since 2012. It will be required to meet the “Base EHR” definition starting January 1, 2025. DSI requirements have been categorized into two sections.  

• Evidence-based DSIs 

• Predictive DSIs 

Evidence-based DSIs are essentially the original Clinical Decision Support (CDS) systems. There are some additional requirements that EHRs need to fulfill to maintain them. There is a lot of guidance on that provided by ONC. On the other hand, Predictive DSIs are the new kids on the block and are the ones where there is a lot of confusion. Here is how the Office of the National Coordinator for Health Information Technology (ONC) has defined Predictive DSI -  

Predictive Decision Support Intervention means technology that:  

• Supports decision-making based on algorithms or models that;  

• Derive relationships from training data and then;  

• Produces an output that results in prediction, classification, recommendation, evaluation, or analysis. 
  

There are a couple of crucial points to note here. Predictive DSI needs to be based on some training data. This is different from rule-based or evidence-based DSIs. Secondly, the outcome doesn’t necessarily need to be “predictive”. As outlined in the last item above, prediction is only one of the intended outcomes. The figure below depicts some use cases around Predictive DSI, commonly called AI in healthcare. It comes from a report on using Artificial Intelligence in Healthcare jointly created by the United States Government Accountability Office (GAO) and the National Academy of Medicine.

How can AI be used in healthcare?

EHRs, AI App Developers and DSI Criterion (b)(11)

So the key question is - What exactly do the EHRs need to do to meet this criterion?  

EHRs are required to enable their users to support, configure, and manage DSI in the EHR. The users should be able to:  

·         Select both evidence-based and Predictive DSIs  

·         Provide electronic feedback data for evidence-based DSIs and export such feedback data 

·         Access complete and up-to-date source attribute information for evidence-based and Predictive DSIs 

·         Record, change, and access source attributes for evidence-based and Predictive DSIs 

It is also essential to explain what EHRs are NOT required to do. EHRs are NOT required to author, develop, or otherwise “supply their own” Predictive DSI. The criterion includes a variety of contemporary functionalities and configuration requirements as well as an expanded set of information, or “source attributes,” related to DSI. There are also requirements for risk management practices to be applied to Predictive DSIs. These requirements build transparency and, ultimately, accountability for these DSIs. The information will help users to determine on their own whether Predictive DSIs within the EHRs are fair, appropriate, valid, effective, and safe (FAVES). These requirements are conditional and only required if the EHR develops or manages DSIs as part of the EHR. This regulation provides an excellent framework for building responsible AI in healthcare by separating the development of Predictive DSIs from usage.  

The HTI-1 rule does not specify any details on how to implement and support these interventions in the EHR. With the recent updates in FHIR®, we believe it will be the most common implementation platform.  The requirement to use USCDI version 3 for EHRs was delayed till 2025. However, the rule requires DSIs to leverage the USCDI data elements, especially during data imports. Based on discussions with multiple stakeholders, we believe, most implementations will use FHIR and leverage USCDI 3 from the start. The initial Patient Access API criterion did not require FHIR, but many EHRs implemented FHIR anyway. We expect a similar path here, where most implementations will use FHIR and USCDI v3 until the rule ultimately requires FHIR.  

Although this makes the certification trajectory very clear for the EHRs, there is some ambiguity regarding the requirements for DSI (AI app) developers. It is unclear if they must certify their technology to the b11 criterion. It will be interesting to see if a related criterion is defined in the future to enforce that all DSIs must get certified before being allowed to be configured in an EHR.   

We are developing a plug-and-play solution to assist our EHR partners in meeting these requirements. The new solution will leverage our g10 Certified BlueButtonPRO solution, which already supports USCDI v3 and v4. We welcome feedback and thoughts from all facets of the healthcare IT community on how they are implementing these requirements. We are leveraging CDS hooks as the core technology for the implementation, which we believe is the best way to enable DSIs at the point of care and seamlessly integrate into the EHRs. We are also working with various potential DSI algorithm suppliers. If you are one, we would like to hear from you and explore the potential of tackling some of those challenges together.

Interested in learning more about HTI-1 and DSI?

Latest Posts on HTI-1