ONC HTI-1 Decision Support Transparency Testing Guide

Statutory requirements

1. 01

   Scope: certified health IT developers

   45 CFR Part 170

   HTI-1 applies to developers of certified health IT under the ONC Health IT Certification Program, not directly to hospitals or AI startups. AI vendors felt it through their EHR partners, who needed source attribute information to publish for predictive models they supplied or enabled through the EHR.

2. 02

   Predictive DSI definition

   45 CFR 170.102

   HTI-1 defined a predictive decision support intervention as technology that supports decision-making based on algorithms or models that derive relationships from training data and then produce an output that results in prediction, classification, recommendation, evaluation or analysis. LLM-based features embedded in an EHR generally fit that definition.

3. 03

   Source attributes (removed by HTI-5)

   45 CFR 170.315(b)(11)(iv) and (v), as originally finalized

   HTI-1 required certified Health IT Modules to let users review source attributes for predictive DSIs (31 for predictive, compared with 13 for evidence-based DSIs), covering intended use, cautioned out-of-scope use, development details, quantitative validity and fairness measures, and ongoing maintenance. HTI-5 removed these requirements in 2026, citing limited clinician time to review them during care.

4. 04

   Intervention risk management (removed by HTI-5)

   45 CFR 170.315(b)(11), as originally finalized

   HTI-1 required developers to apply intervention risk management practices for predictive DSIs, covering risk analysis, risk mitigation and governance, organized around FAVES characteristics: fair, appropriate, valid, effective and safe. HTI-5 removed this requirement from certification, but health systems' own governance committees and contracts often still ask for the same evidence.

5. 05

   Related nondiscrimination duty

   45 CFR 92.210

   Separately, the HHS Section 1557 rule requires covered health programs to make reasonable efforts to identify uses of patient care decision support tools that employ input variables measuring race, color, national origin, sex, age or disability, and to mitigate the risk of discrimination. Check the current status of that provision, which takes a different route from HTI-1 and is not affected by HTI-5.

Evaluation methodologies

Testing artifacts to retain

The evidence folder an examiner, notified body or internal auditor will ask for. GONATURELLE compliance reports export most of it from signed runs.

GONATURELLE assertion rules

Each assertion maps to a requirement above. Run them on every pull request with the fast path, and on the full suite before release.

from gonaturelle import EvalSuite, assertions as A

suite = EvalSuite.load('healthcare/soap-note-summarization', site='local_deid_2026q3')
results = suite.evaluate(agent=ehr_summary_agent, max_concurrency=10)

# Valid: accuracy against local clinician labels
results.assert_all([
    A.judge(rubric='soap-fidelity-v3', threshold=0.90, require_kappa=0.80),
    A.no_fabricated_findings(source='encounter_record'),
])

# Fair: subgroup performance gap within tolerance
results.fairness(group_by=['age_band', 'sex', 'preferred_language']).assert_parity('omission_rate', max_gap=0.03, ci=0.95)

# Appropriate: out-of-scope populations trigger abstention
oos = EvalSuite.load('healthcare/out-of-scope-pediatric').evaluate(agent=ehr_summary_agent)
oos.assert_all([A.abstains_or_warns()])

# Safe: medication and allergy omissions are zero-tolerance
results.assert_zero_contraindications()

# Model card claims reproduced by this run
results.reconcile_claims('model_card/summary_agent.yaml').assert_all_within(tolerance=0.01)
results.export_manifest('governance/dsi_faves_run.json', sign=True)

Audit checklist