# Measuring quality and outcomes that CMS actually pays for
A 68-year-old patient with heart failure is discharged on a Tuesday. Eighteen days later, she is back in the emergency department, short of breath. To the clinical team, this is a sad but routine event. To the hospital's finance office, that single readmission is a data point in a formula that can swing millions of dollars in Medicare payments across the year.
This lesson shows you how raw discharge data becomes a quality score, why risk-adjustment is the whole game, and how the Centers for Medicare and Medicaid Services (CMS), the federal agency that runs Medicare, turns those scores into real money.
For most of the 20th century, hospitals were paid for volume: more admissions, more procedures, more revenue. That is "fee-for-service."
Starting in the 2010s, CMS shifted toward "value-based payment," meaning a portion of what a hospital earns depends on quality and outcomes, not just activity. The logic: a readmission often signals that something went wrong (poor discharge planning, medication confusion, no follow-up appointment).
Two programs matter most for hospital outcome data:
Both are budget-neutral in design, meaning CMS withholds a slice of payments and redistributes it based on performance. If you do worse than peers, you fund the bonus for hospitals that do better.
Two outcome families dominate the scorecard.
30-day readmission rate. Of patients discharged with a given condition, what share return to any hospital within 30 days? CMS focuses on conditions like heart failure, pneumonia, chronic obstructive pulmonary disease (COPD), and hip/knee replacement.
30-day mortality rate. Of patients admitted for a condition like heart attack or heart failure, what share die within 30 days?
Note the "any hospital" detail. A patient can be discharged from Hospital A and readmitted to Hospital B, and it still counts against A. This is why hospitals rely on claims data, which follows the patient across facilities, not just their own internal records.
Let's build the raw numerator and denominator. Imagine a discharge table with one row per admission.
-- Simplified 30-day heart failure readmission
WITH index_admissions AS (
SELECT patient_id, admit_date, discharge_date
FROM admissions
WHERE primary_dx = 'heart_failure'
AND discharge_status <> 'expired' -- can't readmit a death
),
readmits AS (
SELECT i.patient_id, i.discharge_date
FROM index_admissions i
JOIN admissions a
ON a.patient_id = i.patient_id
AND a.admit_date > i.discharge_date
AND a.admit_date <= i.discharge_date + INTERVAL '30 days'
)
SELECT
COUNT(DISTINCT r.patient_id) * 1.0 /
COUNT(DISTINCT i.patient_id) AS raw_readmit_rate
FROM index_admissions i
LEFT JOIN readmits r USING (patient_id);This gives you a *raw* rate. And a raw rate is almost useless for comparing hospitals. Here is why.
Suppose Hospital A is a downtown academic center treating the sickest, oldest, poorest patients. Hospital B is a suburban hospital with healthier patients. Hospital A will have more readmissions no matter how good its care is.
Comparing their raw rates would punish A for doing hard work well. Risk-adjustment fixes this. It asks: given this hospital's specific mix of patients (their ages, other diagnoses, prior admissions), how many readmissions would we *expect*? Then it compares expected to observed.
CMS reports a ratio-based measure. The key output is often expressed as a ratio of predicted to expected outcomes:
CMS multiplies that ratio by the national average rate to produce a risk-standardized readmission rate (RSRR) or risk-standardized mortality rate (RSMR). These are the numbers that actually drive payment.
The models use hierarchical logistic regression and adjust for clinical variables (comorbidities from diagnosis codes) but deliberately do *not* adjust away everything. For example, socioeconomic factors have been handled through peer grouping in HRRP rather than baked directly into the clinical model, a design choice meant to avoid setting lower standards for hospitals serving disadvantaged populations. This remains a debated topic.
You do not need to build these models yourself. CMS publishes the methodology, and the underlying results are public on Medicare's Care Compare tool, where anyone can look up a hospital's performance.
🎬 [VIDEO: "How Risk Adjustment Works in Value-Based Care" — youtube.com — a clear explainer of why observed-to-expected ratios matter more than raw rates]
CMS compresses many measures into the Overall Hospital Quality Star Rating, a 1-to-5 star summary shown to the public.
The star rating groups dozens of measures into categories such as:
Each hospital gets a summary score, and hospitals are sorted into star buckets. Patients see stars; executives see the measures underneath. A drop from 4 stars to 3 rarely changes a single patient's decision, but it shapes payer negotiations, reputation, and referral patterns.
Important: the public star rating and the payment programs (HRRP, HVBP) use overlapping but not identical measures. Do not assume a good star rating means zero payment penalty.
Here is the mechanism that makes this a finance topic, not just a clinical one.
Under HRRP, CMS calculates an "excess readmission ratio" for each measured condition. If your hospital readmits more than expected, CMS applies a payment reduction to *all* your Medicare inpatient payments for the year, up to a capped percentage (the maximum penalty has been 3 percent of base Medicare inpatient payments).
Do the arithmetic on scale. Take a hospital with, say, 200 million dollars in annual Medicare inpatient payments (illustrative figure). A 1 percent penalty is 2 million dollars. A move from a 1 percent to a 0.4 percent penalty, driven by a modest improvement in heart failure and pneumonia readmissions, is worth well over a million dollars, recurring every year.
That is why hospitals invest in:
The return on a nurse who calls discharged heart failure patients is measured partly in avoided readmissions, and those avoided readmissions move the observed-to-expected ratio.
Vérification des acquis
1. A hospital performs slightly better than the national average on readmissions, yet still receives no bonus and only a marginal payment adjustment. Which design feature of HRRP and HVBP best explains why performing 'above average' does not guarantee a large reward?
2. Why did CMS shift from fee-for-service toward value-based payment for hospitals?
3. The lesson states that 'risk-adjustment is the whole game.' What is the core reason risk-adjustment is essential when comparing hospital readmission rates?
4. Select ALL correct answers about how HRRP and HVBP function.
Sélectionnez toutes les réponses correctes.
5. Select ALL correct answers about why a single readmission matters beyond the clinical event itself.
Sélectionnez toutes les réponses correctes.
If you are the analyst asked to "pull our readmission rate," these mistakes are easy to make and expensive to miss.
Reporting raw rates to leadership. Always label whether a number is raw or risk-standardized. Presenting a raw rate next to a CMS-published risk-standardized rate is comparing two different things.
Ignoring transfers and planned readmissions. A planned return for chemotherapy or a staged surgery is not a quality failure, and CMS excludes many planned readmissions. Counting them inflates your rate.
Missing out-of-hospital readmissions. If you only use your own EHR (electronic health record), you miss patients who bounced to a competitor. Claims data or a health information exchange closes that gap.
Small denominators. For a low-volume condition, one or two extra readmissions swing the percentage wildly. CMS handles this with statistical shrinkage in its models, pulling small-sample hospitals toward the national average. Your internal dashboard should flag conditions with too few cases to interpret month to month.
Confusing the measurement window with the fiscal impact. CMS uses a multi-year performance period (often three years of data) to calculate a single fiscal year's penalty. The improvement you make today shows up in payments a year or more later. Set expectations accordingly.
A useful internal scorecard has three layers:
1. Raw operational metrics, updated frequently, so clinical teams see trends fast.
2. Approximate risk-standardized estimates, so you can predict where CMS will land.
3. Financial translation, converting projected ratios into estimated payment adjustments so leadership understands the stakes.
The data skill that matters most is not fancy modeling. It is faithfully reproducing CMS's definitions: the right conditions, the right exclusions, the right 30-day window, and honest labeling of raw versus adjusted.