The measure is not the improvement loop

The measure is not the improvement loop.

That distinction matters for dental AI in Medicaid-facing organizations. A system can know the measure, calculate the numerator, summarize the benefit, and still fail the work. The useful question is not only whether a child received a preventive dental visit. It is whether the organization can show what it changed after seeing the gap.

CMS’s Medicaid Oral Health Quality Improvement Resources page makes the frame clear. The oral-health problem is not presented as a code-table problem alone. CMS describes technical assistance for states, QI resources, state examples, delivery of preventive services, and ongoing dental care for children enrolled in Medicaid and CHIP. The page points state teams toward small tests of change, learning collaboratives, and project design.

Taken together, the material describes a work system, not a lookup table.

The measure names the gap

Measures are necessary. They give the project a shared target and keep the team from mistaking activity for improvement.

The CMS page on Promoting Children’s Preventive Dental Visits uses the Child Core Set Oral Evaluation, Dental Services measure as a key indicator for an affinity group focused on increasing children’s preventive dental visits. That is exactly the kind of measure a Medicaid dental AI should understand.

But the same CMS page immediately surrounds the measure with machinery. The affinity group is planned as a 21-month effort. The pre-implementation phase includes process flow mapping, an aim statement, QI partners, and a strategic plan. The implementation phase uses small tests of change through Plan-Do-Study-Act cycles.

The measure does not contain those facts. It points at the place where those facts have to be built.

A weak system stops at the score. It reports that preventive dental visits are low, ranks the population, and recommends outreach. A stronger system asks what the team is trying to change, where the baseline came from, which partner owns which part of the workflow, what intervention is being tested, how often the result will be measured again, and what evidence will prove that the process changed.

The measure is a signal. The loop is the work.

A PIP preserves the route of change

The Medicaid Oral Health Performance Improvement Projects manual gives the route a practical shape.

The manual does not begin and end with a dental measure. It walks states through selecting the PIP topic, identifying the population, defining the aim, selecting measures, establishing a data collection plan, planning the intervention, implementing improvement strategies, analyzing results, and sustaining improvement. That sequence is the ontology an operator actually needs.

An aim statement has to be specific, measurable, achievable, relevant, and time-bound. A baseline has to exist before improvement can be claimed. CMS requires at least three measurement points for a PIP: baseline and two remeasurement points. Measures need sources, numerator and denominator specifications, measurement periods, benchmarks, and goals.

Those fields are not administrative decorations. They decide whether the project can be reviewed.

When a plan says it improved sealant use, the next operator should be able to inspect the population, the starting rate, the target, the intervention, the data source, the remeasurement dates, the barriers found during implementation, and the strategy chosen for the next measurement period. A claim without that chain is only a better number.

Hadto’s thesis is that domain experts become owners when their judgment becomes teachable and reviewable. In Medicaid oral-health QI, the teachable unit is not the metric. It is the route of change around the metric.

The data plan changes the intervention

QI work also refuses the fantasy that data is a neutral afterthought.

The PIP manual asks health plans to define data sources, collection frequency, responsible staff, analysis procedures, and data workflows. It also names the practical work that makes data usable: timely extraction, validation, training on key terms, sharing across quality staff and frontline providers, and regular maintenance of current sources.

The choice matters because the data source can change the intervention.

Claims may show utilization. Enrollment files may define the denominator. Provider files may show network capacity. Member surveys, complaint logs, focus groups, and interviews may explain why families do not complete preventive visits. A workflow failure may live in referral timing, appointment availability, transportation, language access, provider participation, or misunderstanding about cost.

An AI that only reads claims can make the problem look cleaner than it is. An AI that preserves the QI record can show which source is answering which question.

The improvement loop needs both kinds of evidence: the measure that shows whether the rate moved, and the field evidence that shows why the workflow did or did not change.

PDSA is a governance surface

Plan-Do-Study-Act cycles can sound like QI vocabulary, but the CMS material treats them as governance for change.

A PDSA cycle records the planned test, the implementation, what happened, and what the team changed afterward. The PIP manual ties those cycles to intervention tracking measures, course correction, pilot testing on a smaller population before scaling, and comparison of multiple interventions to find what works.

One-shot automation moves in the opposite direction.

A Medicaid dental AI should not simply recommend a campaign and disappear. It should help preserve the loop: why this intervention was chosen, which driver it addresses, which staff or partner owns it, when the test runs, what data will be reviewed, what barrier appeared, what adjustment was made, and whether the next measurement period improved.

State, plan, provider, and community coordination becomes part of the record here. CMS’s PIP manual names state levers such as incentives, public reporting, contract requirements, data support, provider training, cross-agency partnerships, stakeholder engagement, EQRO support, and peer learning. These are not side notes. They are the levers by which the measure becomes a changed workflow.

Ownership lives in the loop

The dental expert already knows that a measure alone is not enough.

They know when a low visit rate is really a provider participation problem. They know when a referral process breaks between primary care and dental care. They know when the denominator is right but the outreach list is stale. They know when one health plan needs data support, another needs provider cooperation, and a third needs a different intervention because the first test did not move the work.

If that knowledge stays private, the organization still depends on the expert being nearby. If the system stores it as an improvement loop, another operator can learn from it.

An owner-ready QI record should preserve the aim, baseline, population, partners, measure specifications, data sources, intervention rationale, implementation owners, PDSA cycles, remeasurement results, barriers, next tests, sustainability plan, and evidence that the workflow changed.

Reporting a metric and owning a system are different jobs.

The Medicaid oral-health QI material gives Hadto a practical rule: AI enablement should not compress Medicaid dental work into benefit facts, measure facts, or source snippets. Those inputs are necessary, but the owner surface is the loop that makes the work reviewable after the first answer.

A measure tells the team where to look.

The improvement loop tells the next operator what changed, who changed it, what evidence survived, and what has to be tested next.

---

Source evidence used in this note: public Medicaid.gov page Medicaid Oral Health Quality Improvement Resources; public CMS PDF Medicaid Oral Health Performance Improvement Projects: A How-To Manual for States; public Medicaid.gov page Promoting Children’s Preventive Dental Visits, reviewed 2026-05-17.