Nartis
A first-principles guide

HEDIS, Star Ratings,
and a Quality
Control System

Healthcare quality is measured, scored and paid for in billions of dollars every year. This walkthrough follows the chain from one member's lab result to a plan's Quality Bonus Payment, and shows why quality works better as a control loop than as a report.

Section 01
The Foundation

What is healthcare quality measurement?

Health plans are graded on whether their members actually receive good care. Not whether claims were paid, but whether members with diabetes have their blood sugar under control, whether women 50 to 74 had a mammogram, and whether members keep filling their prescriptions. Each measure is a percentage: of everyone who should have received the care, how many did?

Example: glycemic control for members with diabetes
Denominator
2,400 members
Members 18 to 75 with diabetes
Numerator
1,872 members
Most recent A1c below 8.0% this year
0%
Performance Rate
Denominator
Who qualifies for measurement
The eligible population: members who meet the criteria for a given measure (age, diagnosis, enrollment length). This is the "who should be getting this care" group.
Numerator
Who met the standard
The subset of the denominator who actually received the care or met the clinical threshold. Numerator ÷ Denominator = your performance rate.
Every care gap is a patient who fell out of the numerator.
The entire business of healthcare quality improvement is: find those patients, close the gaps, and move the rate.
Section 02
The Standard

What is HEDIS?

HEDIS (Healthcare Effectiveness Data and Information Set) is the measure set maintained by NCQA. It defines exactly how each measure is calculated: who qualifies, what counts as a pass, and the lookback window. Plans report HEDIS across commercial, Medicaid and Medicare lines of business. Medicare Star Ratings then draw on a subset of HEDIS measures plus other sources, covered in the next section.

Explore a measure
HEDIS turns clinical judgment into auditable math.
When a plan knows exactly what counts as a pass, it can find every member who has not passed yet and act before the measurement year closes.
Section 03
The Score

What are Star Ratings?

CMS publishes Star Ratings for Medicare Advantage and Part D plans only. Medicaid and commercial plans are not rated by this program. A contract's overall rating, from 1 to 5 stars in half-star steps, combines HEDIS measures with the CAHPS member experience survey, the Health Outcomes Survey (HOS), Part D measures such as medication adherence, and CMS administrative data. Each measure is scored against cut points and weighted, so some measures count three times as much as others.

Star Rating scale and the Quality Bonus Payment
1.0
No QBP
2.0
No QBP
3.0
No QBP
3.5
No QBP
4.0
+5%
4.5
+5%
5.0
+5%
Below 4.0
No Quality Bonus Payment. The plan is paid against its unadjusted county benchmark.
4.0 and up
CMS raises the plan's benchmark by 5%. The same 5% applies at 4.0, 4.5 and 5.0 stars.
Double bonus
In qualifying counties, the benchmark increase doubles to 10% for plans at 4 stars or higher.

Star Ratings also set the share of bid savings a plan keeps as rebates (higher ratings keep a larger share), so ratings affect benefits and premiums as well as payment.

The 4.0 line is the cliff.
The bonus is a step, not a slope. A 3.5-star plan earns no QBP and a 4.0-star plan earns the full 5%. That is why plans manage individual measures against their cut points: moving one measure above a cut point can be what lifts the overall rating over 4.0.
Section 04
The Money

How Star Ratings turn into payment

Medicare Advantage plans are paid against county benchmarks. The Quality Bonus Payment raises the benchmark by 5% for plans at 4 stars or higher (10% in double-bonus counties), and that higher benchmark applies to every member in the contract, not only the members in the measures that improved. How much of the increase becomes plan revenue depends on the plan's bid; for plans that bid below the benchmark, part of it is paid out as rebates that fund extra benefits. The figures below are an illustrative upper bound.

Members
50,000
Benchmark PMPM × 12
$14,400
Annual benchmark
$720M
QBP
5%
Up to
+$36M
Estimate the bonus for a plan
50,000
$1,200
Annual benchmark base
$720.00M
50,000 members × $1,200 PMPM × 12 months
QBP at current rating
$0
At 3.5 stars, below the 4.0 threshold
Lift at target rating (upper bound)
+$36.00M
3.5 to 4.0 stars, 5% of the benchmark base
Lift per member per year
+$720
Spread across all enrolled members
A small change in measure performance can be worth tens of millions.
A measure might touch 8,000 members, but the benchmark increase applies to all 50,000. That is why quality improvement is a financial priority as well as a clinical one.
Section 05
The Problem

Why current systems fall short

Legacy quality platforms are built around one constraint: data is processed after care happens. By the time a gap appears on a dashboard, the visit that could have closed it may be over. A person has to notice it, interpret it and act on it, and at the scale of a health plan most of those moments never happen.

The legacy quality improvement loop
01
Clinical event
Member visit
02
Claims batch
Days or weeks later
03
ETL pipeline
Overnight run
04
Dashboard
Updated weekly
05
Human review
If someone looks
06
Manual action
Months later
07
Missed outcome
Gap stays open
Batch processing
Data arrives in nightly or weekly batches. By the time a gap is visible, the clinical event that could have closed it may have passed.
Human dependency
Every gap needs a person to see it, prioritize it and act on it. Across 50,000 members and 30 or more measures, the math does not work.
No feedback loop
When an action is taken, the system does not know whether it worked until the next batch, so it cannot learn which interventions work.
Section 06
The Design

The closed loop

Nartis replaces that sequence with a continuous loop. Each clinical event is evaluated against measure policy when it arrives. Open gaps are scored and ranked. Actions are routed to the right person or channel, and the outcome feeds back into the next decision.

1
Event
A clinical event occurs: a lab result, a prescription fill, a missed appointment. The system ingests it as it happens.
2
Policy
Measure rules run as code. The event is checked against each affected measure's denominator and numerator criteria.
3
Scoring
Each open gap is scored by severity, likelihood of closing, distance to the measure's cut point and value. A language model can explain the gap in plain words; it does not decide whether the gap is open.
4
Action
The right intervention is proposed and routed: provider alert, member outreach, scheduling or pharmacy follow-up. A person approves anything clinical.
5
Outcome
The result is captured. Did the gap close? Did the measure rate move toward its cut point?
6
Feedback
Outcomes show which interventions work for which members, so the next round of prioritization is better informed.
This is not a faster dashboard. It is a different kind of system.
Legacy platforms report what happened. Nartis is designed to change what happens next, across every member and every measure at once.
Section 07
The Difference

Insight versus control

Moving from legacy analytics to a quality control system is not an upgrade of the same thing. One produces reports. The other is built to produce outcomes.

Legacy quality analytics
Batch processing: data available days or weeks after events
Retrospective reporting: shows what already happened
Human-dependent review: gaps are found only if someone looks
Measure logic embedded in ETL pipelines
Annual HEDIS updates take months of engineering
Dashboard alerts handed to care managers for manual follow-up
No learning: the system behaves the same regardless of outcomes
Nartis
Event-driven: evaluated when the clinical event arrives
Proactive: acts before the measurement year closes
Ranked work: every open gap scored and prioritized
Policy-as-code: measure rules versioned and tested
Spec updates ship as reviewed code with a test suite and audit trail
Workflow integration: care teams receive ranked action queues
Feedback: outcomes inform the next intervention cycle
Section 08
The Scale

From one patient to plan-wide payment

This is the chain the platform is built around. One clinical action, such as a lab drawn, a refill picked up or an outreach call that lands, is a better outcome for one member. It also adds one member to a measure's numerator. Enough of those move the measure past its cut point, which can move the overall rating, which changes payment for the whole plan.

01
Patient
One gap closed
A member with diabetes gets an A1c under 8.0%. One numerator event is recorded.
02
Measure rate
+1 / denominator
In a 2,400-member denominator, one closure adds about 0.04 points. About 20 closures move 80.2% to the 81.0% cut point.
03
Star Rating
Threshold crossed
Weighted measure scores lift the overall rating from 3.5 to 4.0 stars, the Quality Bonus Payment threshold.
04
Plan payment
Up to +$36M
The 5% benchmark increase applies across all 50,000 members in the illustration, not only the members in the measure.
A small clinical action can scale into plan-wide financial impact.
One intervention costs tens to hundreds of dollars. The rating change it contributes to can be worth hundreds of dollars per member across the whole plan. That asymmetry is what the platform is designed to exploit.
Section 09
The Prototype

Try the working prototype

The Nartis prototype runs in the browser on a synthetic population of 10,004 members. It evaluates any member against 34 simplified HEDIS and Stars measures, ranks open gaps, shows the rate math behind each closure, and records an audit trail. One feature calls a live model: Nartis explains why a selected gap is open and drafts an outreach note. Everything else is deterministic code, and the prototype says so on the page.

Synthetic data
10,004 synthetic members
Generated deterministically across 100 diagnosis codes, with labs, screenings, fills and dates relative to the current measurement year. No real patient data.
Policy-as-code
34 HEDIS and Stars measures
Simplified denominator and numerator rules with thresholds and reasoning. Stars-only measures apply to Medicare Advantage members only.
Rule-based scoring
Gap ranking
Open gaps ranked by measure weight, distance to an illustrative cut point, visit recency and adherence level. These are transparent heuristics, not a trained model.
Live model
Ask Nartis
For a selected member with an open gap, Nartis explains why the gap is open and drafts a short outreach note for review. Only synthetic, de-identified fields are sent.
Calculator
Stars bonus calculator
Model the Quality Bonus Payment for any plan size, and see how many closures a measure needs to reach its cut point.
Audit
Audit trail
Each evaluation, simulated action and model call is logged with user, agent, purpose, data used and decision.
Open the prototype How it's built Architecture brief (PDF)
Designed for
Medicare Advantage plans, HEDIS reporting teams and quality analytics groups operating at scale.