Section 1: The UDS Reporting Burden on Health Center Program Awardees
The Uniform Data System represents the annual HRSA reporting framework binding every Health Center Program awardee and look-alike, combining patient demographics, service utilization, clinical quality measures, staffing, and financial data into a standardized national submission. Clinical documentation inside a Federally Qualified Health Center must therefore serve two masters simultaneously: the immediate care of the patient in front of the clinician, and the retrospective population-level reporting that HRSA uses to evaluate performance, ensure compliance with legislative mandates, and identify disparities across underserved communities.
Because the workflow is not merely 'capture a diagnosis,' each reportable fact must satisfy a chain of conditions before it belongs in an aggregate submission. The patient must be eligible for the measure. A qualifying encounter must have occurred. The required clinical finding or intervention must be documented. The finding must fall within the correct measurement period. Any exclusion or exception must be recorded. And the data must trace cleanly back to source. Failing any single link converts a well-intended note into an unreportable or, worse, a falsely reported data point.
For FQHCs specifically, this coordination spans clinicians, rooming and nursing staff, health information management, quality-improvement personnel, informatics teams, and billing and finance. Merry AI positions itself inside this chain as an assistive abstraction and mapping layer—not as the authoritative source of UDS truth. The EHR, the signed encounter record, and the device result remain the record of truth; the platform accelerates abstraction and prepares auditable candidate data for human validation.
Section 2: Blood-Pressure Reporting Under CMS165v13
Under the 2025 HRSA materials, the Controlling High Blood Pressure measure aligns with CMS165v13 and concerns patients aged eighteen to eighty-five with a hypertension diagnosis whose most recent blood pressure during the measurement period fell below 140/90 mmHg. The documentation requirements here are exacting, and they are the single most common source of abstraction error in FQHC quality reporting. Both systolic and diastolic values must be distinct numeric results; ranges or threshold-only statements are categorically insufficient.
Acceptable readings may originate from a clinician measurement, an automated office device, or an acceptable remote monitoring device transmitted to the clinician. A patient-conveyed reading from a home automated monitor qualifies differently from unsupported self-reporting, and the clinician must judge whether the device and reading are reliable. When multiple readings occur on the last day the patient was seen, the measure uses the lowest systolic and the lowest diastolic. Emergency-department and acute-inpatient readings are excluded outright, and the patient must have a countable visit reported on UDS Table 5 during the measurement period to be included at all.
The Structured Extraction Contract
Given these constraints, Merry AI extracts substantially more than the phrase 'BP controlled.' The structured output records the numeric systolic value, the numeric diastolic value, the measurement date and time, the encounter type and location, the documented measurement source or device, whether the reading is the most recent qualifying reading, whether multiple same-day readings exist, the associated hypertension diagnosis such as I10, and the evidence supporting inclusion or exclusion. A note reading 'blood pressure stable' or 'at goal' never becomes a numerator result without numeric evidence—the system routes it to review instead.
Section 3: Separating Tobacco Screening From Cessation Intervention
Tobacco-related reporting depends entirely on distinguishing five status categories—current use, former use, never use, unknown, and not documented—alongside the tobacco product type, screening date, and any documented cessation counseling, medication, referral, or quitline intervention. The critical design failure to avoid is conflating screening with intervention. A note stating 'smoker' supports tobacco status, but it establishes nothing about whether cessation counseling actually occurred, and 'discussed smoking' may be insufficient unless the measure specification accepts that language as a qualifying intervention.
To keep these facts independent, Merry AI produces separate objects: a tobacco_screening field carrying status, product, and documented date, and a cessation_intervention field carrying a present-or-absent flag, the exact supporting evidence text such as 'cessation counseling provided; nicotine replacement prescribed,' and its own documentation date. Both fields preserve the note location so a quality analyst can audit the abstraction against the measure-specific intervention definition rather than trusting a keyword match.
The Anchor-Truth Pipeline
Every fact travels a defined pipeline: clinical note to extracted fact, to validated clinical concept, to UDS field, to quality review, to submission. This staged separation of raw evidence, extracted fact, normalized fact, measure eligibility, and final reporting value is what prevents an AI interpretation from ever being mistaken for an original clinical observation. Prompt templates for each measure are versioned and validated in the Scribing Template Directory before any run touches production data.
Section 4: Architecture Comparison and Governance Standards
When health centers weigh their options, the meaningful comparison is not manual charting versus automation in the abstract, but the specific defensibility each approach offers under a HRSA data audit. The table below contrasts three postures against the requirements that actually determine whether reported data survives reconciliation.
| Capability | Manual Charting | Standard Generic AI Scribe | Merry AI Compliance Architecture |
|---|---|---|---|
| Numeric BP extraction with same-day lowest-value rule | Analyst applies by hand, high abstraction burden | Captures 'controlled' as text, misses numeric logic | Extracts distinct systolic/diastolic, applies CMS165v13 rules |
| Screening vs intervention separation | Depends on reviewer discipline | Frequently conflates 'smoker' with counseling | Independent fields with evidence span and location |
| Source traceability to signed encounter | Paper or manual cross-reference | Often lost after transcript discard | Every fact linked to source document, author, date |
| Measure-version control | Manual policy binder | None; static keyword logic | Versioned mappings with annual HRSA approval |
| PHI retention posture | Physical and EHR records | Variable, often persists transcripts | Zero data retention, RAM-only session shredding |
| Human attestation before reporting | Clinician signs note | Bypassed for structured output | Mandatory review queue with reviewer identity logged |
Governance obligations extend well beyond extraction accuracy. Because the platform processes protected health information, deployment requires a business associate agreement, minimum-necessary role-based access, encryption in transit and at rest, authentication and session controls, audit logging, and defined breach-response and deletion procedures under HIPAA §164.312. Where substance-use-disorder records are involved, 42 CFR Part 2 confidentiality controls determine whether those records are segregated, specially tagged, or excluded from processing absent appropriate authorization. Statutory references and measure logic are validated against authoritative sources including the peer-reviewed literature indexed at the National Library of Medicine PubMed Central archive.
AI Governance and Validation
Before any extracted data feeds a UDS submission, the health center should conduct a retrospective validation study against manually adjudicated records, measuring sensitivity, specificity, positive and negative predictive value, and—most importantly—agreement with the final UDS numerator and denominator classification. A high document-level extraction score is not the endpoint; measure-level denominator and numerator agreement is. Validation runs separately for tobacco status, cessation intervention, blood pressure, diabetes measures, preventive screenings, immunizations, and depression screening, with ongoing bias testing across language, race, ethnicity, age, and site.
Section 5: Physician Attestation, Reconciliation, and Failure Modes
No AI-generated data may silently enter the legal medical record as clinician-authored content. If write-back is enabled, it is clearly labeled, reviewable, and governed by the organization's documentation policy, and the attesting physician retains full override authority. The human review interface lets quality teams view the original documentation, see the extracted phrase, accept or correct the result with a recorded reason, review conflicting sources, filter by measure or provider or confidence, and export an audit file—all with reviewer identity and timestamp preserved. Notice, per the AMA guidance, that UDS does not require a depression screening at every encounter; eligible patients are screened once per measurement period, and the system encodes that eligibility logic rather than over-triggering.
Pre-submission reconciliation compares AI-derived results against EHR-generated quality reports, laboratory and vital-sign tables, billing and encounter records, registries, and prior-year results, because HRSA submission guidance requires data auditing to surface errors and exceptions before submission. The recurring failure modes this catches are specific and well documented: treating 'BP controlled' as a qualifying numeric result, counting a patient-reported reading without device context, including emergency-department readings, using an out-of-period reading, ignoring the most-recent and same-day lowest-value rules, interpreting tobacco use as proof of counseling, treating missing documentation as a negative finding, and mapping a concept to the wrong UDS year.
The strongest compliant positioning is precise: Merry AI identifies and normalizes UDS-relevant evidence in FQHC documentation and maps it to HRSA reporting fields with source traceability, measure-version controls, and human validation—it does not independently complete UDS reporting. Health centers ready to see how this abstraction and attestation chain fits their existing EHR and quality workflow can Book a 15-Minute Workflow Audit and review the measure-specific prompt library in the Scribing Template Directory.
