CMS scores GUIDE's cost and nursing home measures as observed-to-expected (O/E) ratios, which compare what actually happened for your aligned patients with what would be expected for similar patients. For both measures, lower is better. For performance year 2025, the benchmark was 0.95 for total per-capita cost and 0.47 for admissions to long-term nursing home stay. Each is the 50th percentile of a cohort that blends GUIDE participants and non-model providers. Meeting a benchmark adds 1.5 percent to your payment adjustment. Missing it subtracts 0.5 percent.
This guide explains how the ratios are calculated, how CMS builds the benchmarking cohort, where the benchmarks sit in the distribution, and how to read your own results. It draws on CMS's GUIDE Performance Measurement Manual, version 3.0, published May 18, 2026. It is written for GUIDE program leads and analysts.
What Is an Observed-to-Expected Ratio?
An O/E ratio divides actual results by expected results. Expected results come from a risk model that accounts for how complex the patients are. A ratio of 1.0 means results matched expectations. On an inverse measure like cost or nursing home admissions, a ratio below 1.0 means the outcome was lower than expected, which is better.
As a hypothetical example, an O/E ratio of 0.80 on cost would mean observed spending was 20 percent below what the risk model predicted for that group of patients. The ratio is what CMS scores, not the raw dollar figure or raw admission count. That is how CMS compares a participant with a sicker panel and one with a healthier panel on the same scale.
How Total Per-Capita Cost Is Calculated
According to the manual, TPCC is a payment-standardized, risk-adjusted measure of the overall cost of care provided to a participant's aligned patients. Here is how the pieces fit:
Observed cost. CMS aggregates standardized Medicare fee-for-service claims payments for each eligible patient-month.
Expected cost. CMS estimated expected spending for each patient-month with an ordinary least squares regression model that accounts for patient risk factors.
The ratio. Observed and expected monthly costs are added up at the TIN level, which is the taxpayer identification number the participant bills under, to produce an O/E ratio.
Minimum size. CMS scores TPCC only if a participant has at least 240 patient-months, which equals 20 patients with a full year of data each. It sets no continuous alignment requirement, because the minimum case count, risk adjustment and outlier truncation already support stable estimates, and partial-year alignment is handled by the exposure-based design.
Participants report nothing for this measure. CMS calculates it from claims.
How Admissions to Long-Term Nursing Home Stay Is Calculated
The LTNH measure compares the observed number of aligned patients who move into a long-term nursing home stay with the number expected. CMS counts a stay as long-term when it lasts 101 days or more. It uses Medicare claims and Minimum Data Set (MDS) 3.0 nursing home data.
The definition affects timing. Because a stay must last 101 days to count, some stays extend more than three months past the end of a performance year. To build the PY 2025 benchmark, CMS used data from July 1, 2024 through April 30, 2025 and looked forward through August 9, 2025. That date is 101 days after April 30, the last day a new admission could begin and still be observed long enough to become a long-term stay.
Other details from the manual:
Continuous alignment. A patient has to be aligned for at least 90 days to count in the measure. CMS explains that preventing or delaying long-term nursing home entry requires ongoing care coordination, and that early institutionalization is often driven by patient severity rather than care quality.
Minimum size. CMS scores LTNH only if there are at least 20 denominator-eligible patients.
Aggregation. The observed value is the total of all observed long-term stays among the TIN's patients, and the expected value is the total of expected transitions, calculated with risk adjustment. The ratio is the two divided.
How CMS Builds the Benchmarking Cohort
For the claims measures, CMS doesn't set benchmarks from GUIDE participants alone. The manual describes a benchmarking cohort that blends GUIDE participants with non-model providers identified in Medicare fee-for-service claims. CMS builds it in four steps:
Find patients with dementia. Patients need at least one qualifying dementia diagnosis code on a Part B professional claim with a date of service in the year before the performance year.
Check GUIDE eligibility month by month. A patient counts in a month only if they are in Parts A and B, Medicare is the primary payer, they aren't in Medicare Advantage or PACE, aren't receiving hospice and aren't in a long-term nursing home stay of 101 days or more, and are alive at the start of the month.
Keep providers with real dementia volume. CMS identifies the TINs serving these patients and keeps those with at least 50 eligible patients living with dementia, where those patients make up at least 25 percent of the TIN's eligible panel. CMS says these thresholds ensure the cohort reflects providers with meaningful dementia experience.
Assemble the risk adjustment data. The final cohort includes demographics, original reason for Medicare entitlement, prior healthcare utilization, dual eligibility, hierarchical condition categories and measures of dementia severity.
What the Benchmarks Are and Where They Sit
For PY 2025, CMS derived benchmarks for both program tracks from claims for July 1, 2024 through June 30, 2025, and it used the 50th percentile of the cohort for each:
TPCC: benchmark of 0.95. Meeting it means an O/E ratio of 0.95 or lower.
LTNH: benchmark of 0.47. Meeting it means an O/E ratio of 0.47 or lower.
The manual also publishes the distribution of ratios in the benchmarking cohort, which shows how wide the spread is. For TPCC, ratios ran from 4.62 at the weakest end to 0.20 at the strongest. The 25th percentile was 1.29 and the 75th percentile was 0.73. For LTNH, the median was 0.47, the 25th percentile was 1.30, and from the 85th percentile up the ratio was 0.00, meaning at least 15 percent of providers in the cohort had no long-term stays observed.
Two cautions follow from those numbers. First, the LTNH distribution is steep, so small changes in a small number of admissions can move a ratio a long way. Second, the benchmark is the middle of the cohort by design, so meeting it means performing better than about half of comparable providers.
CMS chose different reference points for other measures. The quality of life benchmark for established track participants was 0.91, set at the 30th percentile of GUIDE participant performance alone.
How Benchmarks Change Over Time
For PY 2026, CMS says the TPCC and LTNH benchmarks will use PY 2025 data. It also says it will publish the following year's benchmarks in the manual by June 1 at the latest, at least 30 days before the performance year starts, and that for later years it may hold benchmarks constant or update them based on additional data.
That means the number to aim for can move. Check for a version of the manual newer than 3.0 for the PY 2026 benchmarks before you set targets.
How Scoring Works
Because both measures are inverse measures, CMS scores them with a simple rule:
Meet the benchmark: +1.5 percent contribution to the performance-based adjustment.
Miss it: -0.5 percent.
The difference between meeting and missing one benchmark is therefore 2 percentage points of your monthly payment. For PY 2026, each of the two measures carries a potential of -0.5 percent to +1.5 percent within a total range of -3.5 percent to +10 percent.
How to Read Your Own Results
The Data Feedback Tool shows performance results over a 12-month period, updated quarterly. As of May 2026, the manual says TPCC and LTNH results were available to established track participants, who can see their own value, an established-track average and a benchmark comparator. New track participants' performance data will be available in 2027. CMS also provides a performance-based adjustment workbook each December that shows results for the prior year and compares you with other participants.
When you review results:
Check your denominators first. A ratio built on 20 patients moves more than one built on 200.
Compare with the benchmark, and watch the trend. A ratio just above 0.95 for TPCC is a very different position from one above 1.2.
Remember the lag. LTNH events are counted well after the performance year ends, so today's care affects results months from now.
Look for patterns in the care delivery. The manual's appendix links 24/7 access and proactive care coordination to cost, and advance care planning, consistent navigator support and routine respite to nursing home admissions. CMS describes the appendix as a supportive reference and not a formal protocol.
Where Elli Cares Fits
Both measures depend on what happens between visits: whether a problem is noticed early, whether a caregiver gets support and whether a transition is managed. Elli Cares gives care teams a clinician dashboard, AI-generated summaries, daily wellbeing check-ins and family observations, so a care navigator can see who may need earlier outreach. Elli can't change how CMS calculates a ratio, doesn't guarantee lower costs or fewer nursing home admissions and doesn't replace your Data Feedback Tool. You can read more on our GUIDE page.
Frequently Asked Questions
What is an observed-to-expected ratio in GUIDE?
It compares actual results for a participant's aligned patients with the results expected for similar patients, based on a risk model. A ratio of 1.0 means results matched expectations. For cost and nursing home admissions, lower ratios are better.
What is the GUIDE TPCC benchmark?
For performance year 2025, the total per-capita cost benchmark is 0.95, the 50th percentile of the benchmarking cohort. Participants with an O/E ratio at or below 0.95 receive +1.5 percent, and those above it receive -0.5 percent.
What is the GUIDE long-term nursing home stay benchmark?
For performance year 2025, the benchmark is 0.47, also the 50th percentile of the benchmarking cohort. CMS counts a stay as long-term when it lasts 101 days or more.
Who is in the benchmarking cohort?
A blend of GUIDE participants and non-model providers that serve at least 50 eligible patients with dementia, where those patients make up at least 25 percent of the provider's eligible panel.
Will the benchmarks change?
Yes, they can. CMS says PY 2026 benchmarks for these measures will use PY 2025 data, and that it may hold benchmarks constant or update them in later years.
