In Tier II, EHR clients shape the performance based on the desired “solution” strategy, style guides, and standardizations via approval and scheduling of recommended changes. Solutions are defined as a discrete set of functionality, including templated notes, auto text shortcuts (aka “dot phrases”), orders, order sets, and alerts. This tier optimizes connected EHR solution builds. For example, the Oracle Health EHR (formerly known as Cerner Millennium) contains up to 850 content and configuration tools, each with dozens to hundreds of options and subtools. The output of each of these tools may be connected to one or multiple other solutions. To build a simple form, we may require up to 12 distinct tools and a week of skilled architect time. The proper front-end flow depends on the coordinated build of solutions, but the tools to configure these solutions are siloed, and the solution architect or informaticist may be blind to the full system impacts. Changing the content of this form may negatively impact multiple other solutions, including note templates, orders, interfaces, and discrete data capture. Tier II ensures this hypothetical form is built both to institution-set standards and aligns with these other solutions.

A solution architect needs to update a form, a context-dependent collection of discrete data entry fields. However, multiple other solutions may populate or use these data fields, including results or data review solutions, “smart” documentation or ordering templates, outbound interfaces, logical rules or alerts, or patient registries. Without querying the database or system configurations for each potential impact, the architect is largely unaware of possible downstream ramifications. Tier II addresses this by providing AI-assisted guidance on how these solutions interact, automatically suggesting any necessary edits to keep interrelated components synchronized. In this process, a middleware layer becomes invaluable; it orchestrates data exchange among siloed EHR modules, allowing the AI engine to integrate seamlessly with the relevant system components. By maintaining consistent data structures and communication channels, middleware ensures that the architect’s updates are executed safely and comprehensively. Technically, this would include AI-generated queries of relevant database tables and system files, as well as a graphical user interface overlay that helps the architect visualize potential impacts and either approve or deny suggested changes.

Tier II is designed to streamline how institutions sustain their EHRs, according to enterprise standards, while reducing configuration silos and ensuring that changes to one component do not inadvertently disrupt others. Evaluation metrics may include configuration turnaround time, error rates, compliance with institutional standards, and architect or administrator feedback.

Tier III proactively suggests configuration changes to optimize EHR choice architecture. Optimal choice architecture in this context entails a configurable design that incentivizes the minimum number of clicks, keystrokes, and mouse miles to achieve an intended, quality outcome. Optimal architecture makes the right, efficient choice path the intuitive option, enhancing EHR usability [2]. With the current complexity of EHR design to account for the high variability within patient care, following the most efficient choice paths is not easy or intuitive. By globally monitoring user behavior and determining pockets of efficient users achieving defined process or outcome metrics, Tier III finds the ideal choice paths and suggests configuration changes to democratize them across all relevant user populations. It makes the easy path, the right path. Examples include identifying missing orders or default selections within order sets; optimizing note template content to reduce the manual insertion of discretely captured information or the unnecessary use of free text; updating default naming conventions and selections to reduce the misrouting of orders, notes, or messages; and consolidating unnecessary user positions, preferences, or roles.

An example scenario involves monitoring the ordering patterns of outpatient primary care physicians treating acute nasopharyngitis (common cold). Across hundreds of outpatient clinics, ordering times for these encounters vary widely, despite similar order volumes and medication classes. Tier III AI identifies a subset of providers who achieve faster, more efficient workflows by using personal order sets. The system consolidates these personal sets into a recommended, enterprise-level order set and queues its integration into the primary care physician’s workflow position. Once approved, the AI executes the change. To achieve this level of real-time monitoring and seamless deployment, a robust middleware solution can mediate data traffic, collecting operational metrics from disparate EHR modules, and pushing approved configuration changes into production.

Tier III aims to optimize user workflows by identifying and disseminating best practices across relevant roles and settings. Key metrics may include user efficiency, clinical process and outcome metrics, adoption and utilization rates, and user satisfaction and burnout scores.

Tier IV proactively maintains the currency of EHR clinical content and decision support through two mechanisms. First, content may be directly linked to its derived source. For instance, registries and their integrated actions (orders, forms, laboratory or radiographic studies) could be linked to the United States Preventive Services Task Force (USPSTF) guidelines [3]. When the guidelines change, Tier IV proactively offers the configurations required to incorporate these updates. These linkages could also extend to nonclinical sources including governmental regulations or issuances, institutional policies, or payer requirements. When the Centers for Medicare and Medicaid Services update the essential elements for clinical note content, Tier IV offers the configurations to add or remove the applicable sections for efficient documentation. Second, Tier IV crawls sources of evidence and peer-reviewed literature and cross-checks these findings with existing EHR configured content. As evidence becomes available, Tier IV suggests its EHR incorporation, either as de novo content or updating existing solutions. If a trusted source of truth publishes a new clinical practice guideline, then Tier IV offers a set of EHR solutions to incorporate this clinical practice guideline across the relevant EHR workflows and positions.

The USPSTF updates their breast cancer screening recommendation to begin at age 40 years versus age 50 years [3]. Because the USPSTF had been identified as a source of truth as part of Tier IV, a web-crawling, agentic AI identifies the change and suggests the requisite configuration changes to incorporate this update into the corresponding EHR patient registry. With the underlying Tier II AI system in place, changes to other associated solutions can also be performed concurrently, such as any related forms, rules, or clinical documents.

Tier IV proactively updates clinical content and decision support based on changing guidelines, regulations, and published evidence. Key metrics include update lag time, completeness of updates, accuracy of incorporated guidelines, and regulatory compliance.

Finally, Tier V involves context-dependent functions that serve to enhance care efficiency, quality, and user experience for both patients and providers. These are the copilots. This tier is the current, almost exclusive focus of integrating AI within EHRs. Examples are robust, but popular ones include Microsoft’s GenAI copilot integration within Epic Systems [4] and the Nuance Dragon Ambient eXperience (DAX) AI copilot [5].

Integrating AI into EHR maintenance and configuration carries inherent risks that require careful mitigation strategies. These safeguards must address proper human oversight, data security, safety testing, legal and regulatory compliance, and data standards.

We need to maintain comprehensive records of all AI-generated recommendations and subsequent human-validated changes. This includes versioning, timestamps, rationale for acceptance or rejection, and who approved the changes.

All AI-recommended changes should come with manual build and backout steps in case manual, direct human architect involvement is required to modify or rollback the change.

Authority for approving final changes should be restricted. Only designated administrators, informaticists, or clinicians with appropriate privileges should “sign off” on AI-driven recommendations [6].

Clearly define who bears responsibility if AI-suggested changes negatively impact patient care—whether it is the vendor, the health care institution, or a combination. Incorporate these details into institutional policies and vendor contracts.

Some aspects of Tiers II–IV that meaningfully affect patient care (eg, advanced clinical decision support) may require regulatory oversight or approval from agencies. Understand and follow applicable guidelines when introducing these features [11].

Align AI-driven changes with industry standards (HL7 [Health Level Seven], FHIR [Fast Healthcare Interoperability Resources], SNOMED CT [Systematized Nomenclature of Medicine Clinical Terms], LOINC [Logical Observation Identifiers Names and Codes], etc.) to maintain interoperability.

Consider periodic third-party evaluations of AI systems and processes, focusing on data handling, software quality, and patient safety standards.