Sources and review notes
These links support workflow-level research and do not establish the regulatory status, clinical safety, diagnostic performance, or suitability of any product.
CMS defines quality measures as tools for quantifying processes, outcomes, patient perceptions and organizational structures associated with quality goals. The eCQI Resource Center shows that electronic measures have named identifiers and versions, explicit populations, numerators, denominators, exclusions and exceptions, data criteria, value sets and direct-reference codes, and annual updates aligned with changing evidence and code systems. Candidate measures are not eligible for CMS reporting until adopted through the applicable program process. CMS stratification guidance explains that subgroup reporting can expose differences hidden by aggregate rates and is not the same as risk adjustment. ONC's Health IT Playbook describes population-health tools, panel management, registries and targeted outreach, but it does not define a universal care gap or certify a vendor. These sources do not make every missing data element a missed service, make a reported measure applicable to a local improvement workflow, or prove that a message, appointment, order, documentation update or numerator event improved health. Buyers should define each gap with the measure or clinical-program owner, exact version and reporting or observation period, target population and attribution, denominator, numerator, exclusions and exceptions, value sets and code-system releases, timing and lookback, required data and provenance, evidence hierarchy, closure event, reopen logic, clinical action, responsible team, patient communication and escalation. Keep separate states for potentially eligible, evidence incomplete, service not documented, service documented outside the network, not yet due, excluded or excepted, clinically inappropriate, patient declined, unreachable, scheduled, ordered, completed, result reviewed, documented, reported and reconciled. A completed task or claim is not automatically a closed gap, and absence from one EHR, HIE or claims feed is not proof that care did not occur. Every generated gap should retain patient match and attribution, source facts and dates, calculation and terminology versions, reason and uncertainty, missing evidence, reviewer and corrections, outreach attempts and preferences, clinical decision, service and result evidence, measure disposition, report submission and later reconciliation. AI may retrieve evidence, classify possible gaps, prioritize queues and draft outreach, but qualified clinical and quality owners should determine applicability, exceptions, appropriate action and closure; outreach staff should not convert a measure result into diagnosis or treatment advice. Acceptance testing should use independently reviewed cases across new and changing attribution, enrollment and age boundaries, different measurement periods, completed care inside and outside the network, delayed and corrected claims, duplicate and merged identities, code and value-set updates, historical versus current records, exclusions and exceptions, contraindications, patient refusal, multiple acceptable services, partial series, recurring intervals, ordered but uncompleted care, missing results, conflicting sources, recent discharge, pregnancy and pediatric transitions, language and disability access, deceased and moved patients, outreach opt-outs, unavailable services, source outages and measure-version changes during active work. Test negative cases and missed true gaps as well as false positives. Measure source and population coverage, data latency, matching errors, gap precision and recall against reviewed truth, false outreach and missed eligible patients, evidence retrieval and reviewer edits, exclusion and exception accuracy, time from identification to appropriate action, successful contact, patient decline and opt-out, scheduled and completed service, result and documentation closure, reopened gaps, report reconciliation, staff workload, complaints, adverse events and outcomes by relevant subgroup. Report each funnel denominator and distinguish operational, measure and clinical closure. Higher list volume, outreach, scheduling, numerator completion or reported performance does not alone establish appropriate care, equity, improved outcomes, savings, compliance or causation. Governance should include clinical, quality, data, patient-access, privacy, security and equity owners; version and release management; prospective shadow calculation; bidirectional auditing; subgroup review; source and decision logs; role controls; minimal PHI in outreach; accessible human alternatives; incident response; rollback; export and vendor exit. Automation must not invent evidence, silently alter source records or measure logic, treat area-level or predicted attributes as patient facts, suppress patients who are harder to reach, close gaps from unsupported inference, send sensitive or urgent clinical content without approved review, or make autonomous diagnosis, treatment, coverage or reporting attestations.
