Comment on the CDC Data Intermediaries RFI: Data Quality Certification as a Universal Intermediary Baseline

The health data exchange ecosystem has achieved a robust, independently certified infrastructure for cybersecurity, most notably through the HITRUST certification required by TEFCA Common Agreement Section 12.1.2 for all Qualified Health Information Networks. No analogous certification exists for data quality, accuracy, or provenance.

HITRUST certifies the pipe. Nobody certifies the water.

As data intermediaries increasingly feed AI pipelines for clinical decision support, prior authorization, and public health surveillance, this asymmetry exposes the ecosystem to compounding risks. Box Commons recommends that the CDC incorporate independent, third-party data quality and provenance certification as a core component of the proposed Universal Intermediary Baseline.

USCDI standardizes the format of exchanged data elements. It does not certify the accuracy of their content. A USCDI-compliant exchange can transmit perfectly formatted records that are clinically inaccurate, duplicated, or stripped of the provenance metadata needed for downstream AI use.

The NCQA Data Aggregator Validation (DAV) program is the closest existing analogue to "HITRUST for data quality": a 12-to-18-week cycle of process review, system conformance testing, and primary source verification. But DAV certifies fidelity to the source record, not suitability for AI training. The Sequoia Project's Data Usability Workgroup guide addresses provenance thoughtfully, but adoption is voluntary and unaudited.

Data quality certification for intermediaries should encompass four elements:

1. Semantic accuracy: clinical data faithfully represents the source encounter, validated through methods analogous to NCQA DAV.
2. Provenance integrity: the chain of custody through each intermediary transformation is documented and verifiable.
3. Consent documentation: verifiable authorization for data entering secondary use pathways, including AI training.
4. Demographic representativeness metadata: enough granularity to support the fairness testing required by ONC HTI-1 and state frameworks such as Colorado Regulation 10-1-1.

We recommend a three-tier maturity model: Tier 1 baseline compliance (USCDI, HITRUST, FHIR Provenance Resource support); Tier 2 DAV-equivalent data quality certification; Tier 3 AI-readiness certification covering consent chains, representativeness, and metadata survivability through de-identification.

Data intermediaries are often both the entities being certified and the entities writing the certification criteria. Any certifying body should operate under multi-stakeholder governance with equal-weight chambers for industry, civil society, and academic research, plus four structural safeguards: a 25 percent cap on revenue from any single entity, cooling-off periods for individuals moving from regulated entities into governance roles, funded participation for consumer advocates and independent researchers, and an annual public independence audit.

Certification standards should default to royalty-free IP, specify outcomes rather than technologies, and issue machine-readable credentials queryable through open registries.

1. Include data quality and provenance certification in the Universal Intermediary Baseline.
2. Adopt and extend the NCQA DAV methodology with AI-readiness criteria.
3. Establish tiered certification maturity levels.
4. Require FHIR Provenance Resource support as a baseline capability.
5. Condition certification frameworks on multi-stakeholder governance.
6. Address AI-specific risks, including the de-identification provenance paradox and consent gaps in ambient AI documentation, in intermediary assessment.