Implementation Guide

    Building a Section 1557 Inventory of Clinical Algorithms That Use Race, Color, National Origin, Sex, Age, or Disability Inputs

    A Section 1557 clinical algorithm inventory is a maintained record of every patient care decision support tool a covered entity uses, automated or not, with each tool's owner, location, input variables, clinical use and mitigation decision. Scope it from the definition in 45 CFR Part 92, flag inputs tied to race, color, national origin, sex, age or disability, and record the reasonable efforts made under 45 CFR 92.210(c). The inventory documents identification and mitigation. It does not prove nondiscriminatory outcomes.

    1. 1

      Identity and type

      Stable ID, name, version, and whether the tool is automated, non-automated or combined.

    2. 2

      Location and origin

      EHR-embedded, vendor software, third-party model or local build, with the vendor or internal team named.

    3. 3

      Ownership

      An accountable clinical or operational owner and a technical contact.

    4. 4

      Inputs

      Each input variable, with flags for race, color, national origin, sex, age or disability and the way each is used.

    5. 5

      Clinical use

      Use case, decision-making role (informs, augments or replaces), intended population and intended users.

    6. 6

      Mitigation and review

      Mitigation decision, rationale, any disability modification considered, approver, review date and validation evidence on file.

    7. 7

      Intake trigger

      Add an inventory check to procurement, vendor onboarding, EHR change control and local tool development.

    8. 8

      Classification

      Apply the scope statement and decide whether the tool supports clinical decision-making.

    9. 9

      Input review

      Record inputs and flag protected-characteristic variables, including output adjustments.

    10. 10

      Mitigation decision

      Document the decision, owner and rationale for each flagged tool.

    11. 11

      Scheduled review

      Set a review date per tool and re-review on version change, population change or retirement.

    Step 1: Set the scope from the regulatory definition

    Start with the text, not with a list of known software. In the final rule published in May 2024, HHS OCR replaced the proposed term "clinical algorithms" in 45 CFR 92.210 with "patient care decision support tools" and added a definition. Under Part 92, the term means any automated or non-automated tool, mechanism, method, technology, or combination of these, used by a covered entity to support clinical decision-making in its health programs or activities. OCR described the definition as more precise and as reaffirming that the section applies to tools used in clinical decision-making that affect the care patients receive.

    Two consequences follow for scope. First, the definition is not limited to software. Flowcharts, guideline-based scoring and manual calculators can qualify alongside predictive models. Second, paragraph (c) requires a covered entity to make reasonable efforts to mitigate the risk of discrimination resulting from the use of each tool identified under paragraph (b). The inventory is the working record that connects identification to that mitigation duty. As of November 2026, § 92.210 appears in the current regulation, and no amendment, stay or repeal is reported.

    Write the scope statement in your own policy using the regulatory wording, and cite the current rule text rather than the 2022 proposed rule, which used the superseded term. Treat disability as a distinct review track. OCR guidance states that covered entities may not discriminate on the basis of disability through these tools and ties that principle to the reasonable modification requirement in § 92.205, unless a fundamental alteration would result.

    Step 2: Identify where tools live and how to find them

    Clinical decision support is rarely held in one system. Plan for four deployment locations: calculators and rules embedded in the EHR, vendor software, third-party models, and tools built locally by clinical or analytics teams. Add a fifth category for non-automated methods such as paper protocols and reference tables, because the definition reaches them.

    The regulation does not prescribe discovery methods, so the following are practitioner suggestions. Review EHR configuration for order sets, best-practice alerts, embedded calculators and result-interpretation logic. Read vendor contracts and procurement records for clinical modules and analytics add-ons. Pull entries from any clinical informatics or content-governance catalog. Then interview service-line leaders, because local spreadsheets and departmental tools seldom appear in enterprise systems.

    Use use-case prompts to widen the search. Commenters on the proposed rule asked OCR to confirm coverage of tools that assess health status, recommend care, determine eligibility, allocate resources, conduct utilization review, and provide disease management guidance. That was a commenter request, not regulatory text, but it works as a checklist of places to look. The 2022 preamble also described uses such as screening, risk prediction, diagnosis, prognosis and treatment planning.

    Known examples help reviewers recognize what to flag. The preamble noted that race and ethnicity are often explicit input variables, described race correction as adjusting an algorithm's output based on race or ethnicity, and cited tools that evaluate kidney function. Pulmonary function calculators and many risk scores raise the same question. Age and sex are also common inputs to risk scores, so the review must cover every listed characteristic, not race alone.

    What the inventory can and cannot establish

    The standard in § 92.210(c) is reasonable efforts. An inventory shows that the organization identified its tools, understood which ones use protected-characteristic inputs, and made and recorded a mitigation decision with an accountable owner and a date. That is the record a regulator or auditor can examine.

    It does not establish that a tool produces nondiscriminatory outcomes. Outcome analysis is a separate activity, and the inventory can reference its results through the validation evidence field without claiming more than the evidence shows. The preamble to the proposed rule also stated that the section is not intended to prohibit or hinder the use of clinical algorithms. It is meant to make clear that discrimination occurring through their use is prohibited. A mitigation decision may therefore be to retain a tool with a documented rationale, modify it, or replace it, and the record should say which and why.

    Keep the legal basis visible in the record: 45 CFR 92.210 and the definition in Part 92. Where a tool affects people with disabilities, note whether a reasonable modification was considered under § 92.205.

    Where runtime enforcement supports the compliance record

    An inventory identifies tools and documents decisions. For AI-driven tools that operate in the execution path, such as models, agents and tool calls, a second question remains: what was the system actually allowed to do at the time, and what did it do?

    Trussed addresses that narrower question. It enforces policy in the execution path for individual AI interactions and tool calls, with actions to allow, redact, block, rate limit or escalate for human approval. Enforcement is applied at runtime, not reviewed after the fact. That enforcement generates continuous audit-ready evidence mapped to HIPAA and NIST, which a compliance team can attach to the relevant inventory record. Trussed runs in production with healthcare organizations.

    This is a supporting layer and not a substitute for the inventory. Trussed does not discover tools, decide whether a tool is in scope, or demonstrate that outcomes are nondiscriminatory, and it makes no claim about model provenance, lineage or dataset provenance. The inventory remains the organization's own record of identification and mitigation.

    Inventory lifecycle

    Scope

    Apply the regulatory definition of a patient care decision support tool.

    Discover

    Find tools across EHR, vendor, third-party and local sources.

    Record

    Capture inputs, use, owner and protected-characteristic flags.

    Decide

    Document the mitigation decision and review date.

    Maintain

    Update the record when tools are added, changed or retired.

    Connect your inventory to enforceable, evidenced controls

    Discuss how policy enforcement in the execution path and audit-ready evidence can support the compliance record for AI-driven tools in your environment.

    Talk to an Expert