Since January 2024, Medicare Advantage plans have had to judge inpatient admissions by the same standard traditional Medicare uses. That standard, the two-midnight benchmark in 42 CFR § 412.3, turns on the expectation the admitting physician formed, not on how long the stay turned out to be. CMS told plans in February 2024 that whether an admission actually crossed 2 midnights “is not a factor” in the criteria.
The evidence since then points to partial compliance. In claims data from more than 1,900 hospitals, the share of MA claims classified as observation fell by about 4 percentage points after the rule took effect. It stayed more than 3 times the traditional Medicare rate. No public source shows how individual plans decide these cases, and the prior authorization metrics plans began posting in March 2026 do not separate them out.
This year brings 2 changes that raise the stakes. Decisions a plan makes during a hospital stay are now organization determinations with appeal rights. CMS has also begun removing procedures from the inpatient-only list, which moves more surgical admissions into the expectation test. The change argued for here is in how plans review admissions after the fact, so that the reviewer judges the expectation without knowing the outcome.
Key takeaways
- The test is the expectation, not the stay. MA plans must apply the inpatient admission criteria in § 412.3. CMS has said the actual length of the stay is not a factor in those criteria.
- Compliance looks partial. MA observation rates fell by about 4 percentage points after January 2024 but stayed more than 3 times the traditional Medicare rate, according to an analysis by Kodiak Solutions.
- Public data cannot settle the question. No public source reports inpatient denials or observation conversions plan by plan.
- 2026 adds protections during and after the stay. Decisions made during a stay are now organization determinations with appeal rights. An approved admission can be reopened only for obvious error or fraud.
- The inpatient-only list is shrinking. CMS removed 285 procedures from the list in 2026 and plans to end it by January 2028. In MA, those procedures now face the expectation test.
- Retrospective reviews should be blind to the outcome. A reviewer who knows the patient went home after 1 night is answering a different question. Plans can build the review file as of the admission order and record the judgment before the outcome is shown.
What the rule requires
The requirement comes from the CY2024 Medicare Advantage and Part D final rule, CMS-4201-F, finalized in April 2023. It added § 422.101(b)(2), which requires MA plans to follow the general coverage and benefit conditions in traditional Medicare law. The regulation names “payment criteria for inpatient admissions at 42 CFR 412.3” as an example. It also names the procedures CMS designates as inpatient-only, a detail that matters more in 2026 than it did when the rule was written.
Section 412.3 sets out 3 ways an admission can be appropriate for inpatient payment. Each starts with a physician’s order. Each leaves a plan something different to review.
| Path | What the regulation says | What a plan can review |
|---|---|---|
| Two-midnight benchmark, § 412.3(d)(1) | Inpatient admission is generally appropriate when the admitting physician expects the patient to need hospital care that crosses 2 midnights | Whether that expectation was reasonable, based on the medical record |
| Inpatient-only procedures, § 412.3(d)(2) | Procedures CMS designates as inpatient-only are appropriate for inpatient payment “regardless of the expected duration of care” | Whether the procedure is on the list |
| Case-by-case exception, § 412.3(d)(3) | An admission expected to last less than 2 midnights can still be appropriate, based on the admitting physician’s clinical judgment and support in the record | Whether the record supports that judgment |
From 42 CFR § 412.3, paraphrased except where quoted.
The benchmark asks the physician to weigh factors such as the severity of the patient’s signs and symptoms and the risk of an adverse event. Those factors must be documented in the medical record to count. They are judgments made at a point in time, with the information available then.
The regulation also anticipates stays that end early. Under § 412.3(d)(1)(ii), if an unforeseen circumstance shortens a stay the physician expected to cross 2 midnights, the patient may still be treated as an inpatient. CMS’s two-midnight fact sheet, updated in March 2026, lists “rapid clinical improvement” among those circumstances. The regulation asks what the physician expected. It does not ask whether the patient in fact stayed 2 midnights.
Why status matters
Status decides payment. An inpatient admission is paid as an admission, while observation is billed as an outpatient service, usually at a lower rate. For patients, status can change cost sharing. In traditional Medicare it also decides whether a later skilled nursing stay is covered, because that benefit requires a qualifying inpatient stay. Many MA plans waive that requirement, so in MA the dispute is mostly about hospital payment and the enrollee’s cost sharing.
That is why the standard matters so much to hospitals. A plan that applies a stricter test than § 412.3 shifts part of the cost of the patient’s care from the plan to the hospital, case by case, without changing a single published benefit.
The presumption does not travel
In traditional Medicare, the benchmark has a companion. The two-midnight presumption tells Medicare’s review contractors that stays crossing 2 midnights generally should not be the focus of medical review. CMS’s February 2024 answers to plans’ questions separated the 2 ideas. The presumption “is a medical review instruction given to Medicare post-payment audit and compliance contractors,” and plans do not have to follow it. The benchmark is a coverage criterion, and plans must follow it.
So an MA plan can review any inpatient admission, including one that lasted a week. What it cannot do is apply a different standard. The same answers said a plan may evaluate whether the admitting physician’s expectation was reasonable. That evaluation “should defer to the judgment of the physician as long as that judgment was reasonable,” and it rests on “the complex medical factors documented in the medical record.” Whether the admission actually crossed 2 midnights “is not a factor in the inpatient admission criteria at § 412.3.”
Plans also cannot route around the standard with their own screening criteria. Section 422.101(b)(6) allows internal coverage criteria only “when coverage criteria are not fully established” in Medicare law, and for inpatient admissions § 412.3 establishes them. HFMA’s coverage of the guidance noted that plans may not use commercial screening tools to override the two-midnight rule. The tools can inform a review. The standard that decides the case is still § 412.3.
What the evidence shows
No public data source reports how often MA plans deny inpatient admissions or convert them to observation. The best available evidence comes from hospital claims that vendors aggregate. The most cited analysis, from Kodiak Solutions, covered more than 1,900 hospitals from July 2023 to June 2024, a period that spans the rule’s start.
| Coverage | Monthly observation rate | Period |
|---|---|---|
| Medicare Advantage | 18.1% to 20.2% | July to December 2023 |
| Medicare Advantage | 14.4% to 16.1% | January to June 2024 |
| Traditional Medicare | 3.7% to 5.2%, with a spike to 8.5% in December | July 2023 to June 2024 |
| Commercial insurance | 15% in January, rising to 18.5% in June | January to June 2024 |
Compiled from Becker’s Hospital Review and HFMA coverage of the Kodiak analysis (October 2024). Rates are the share of claims classified as observation stays.
The drop in MA rates came as CMS confirmed that the rule applied to plans. The gap with traditional Medicare remained. Kodiak’s Matt Szaflarski said the gap “suggests that MA plans are not fully complying with the federal rule.” He also said MA observation rates were similar to those for commercial insurance. HFMA’s Shawn Stack put the obligation more bluntly. “It has always been clear that Medicare Advantage plans absolutely, 100% need to be following the two-midnight rule.”
Hospital systems saw the change in their results. HCA reported 11% growth in MA admissions in the third quarter of 2024 and attributed 2 percentage points of it to the rule. Not every system saw relief. Community Health Systems reported that its third-quarter denials had doubled from a year earlier, a $10 million headwind.
What the numbers cannot show
An observation rate is not a compliance rate. MA enrollees differ from traditional Medicare patients, and some observation stays are correct. A plan could follow § 412.3 exactly and still see more observation stays if its enrollees arrive less sick. Hospitals also anticipate plan behavior. A hospital that expects a denial may place a patient in observation from the start, and a claims count records that as the hospital’s choice.
The public metrics that should help do not. The prior authorization reports MA plans began posting in March 2026 aggregate all items and services, as our review of the first public metrics described. Inpatient admission decisions disappear into the totals. Until plans report admission decisions on their own, the public will keep measuring compliance from the hospital side.
The other end of the stay shows a related pattern. A study in JAMA Internal Medicine, published in September 2025, analyzed more than 89 million hospitalizations from 2017 to 2023. Average stays for MA patients rose from 6 to 7.1 days, compared with 5.8 to 6.3 days in traditional Medicare, according to AHA News. The researchers pointed to prior authorization requirements and limited post-acute networks as possible causes. Under per-admission payment, the hospital usually bears the cost of those extra days. Our post-acute analysis covers that side of the problem.
Where disputes concentrate
Before the rule took effect, the American Hospital Association told CMS in October 2023 that some plans had said they would not change their protocols. It also reported denials relabeled as “payment integrity administrative review[s].” This analysis groups the disputes into 4 patterns, each tied to a specific rule.
| Dispute | What the plan does | What the rules say |
|---|---|---|
| Hindsight denial | Denies a short stay after discharge because the patient recovered quickly | Whether the admission crossed 2 midnights is not a factor. CMS counts rapid clinical improvement as an unforeseen circumstance that can shorten an inpatient stay |
| Screening tool override | Applies admission criteria stricter than § 412.3 | Internal criteria are allowed only where Medicare’s criteria are not fully established |
| Reversed approval | Approves the admission, then denies payment in a later audit | § 422.138(c) bars a later denial for lack of medical necessity. Since 2026, approved admissions can be reopened only for obvious error or fraud |
| Downgrade during the stay | Directs observation status while the patient is still in the bed, without a formal decision | Since 2026, a decision made during the stay is an organization determination, with notice and appeal rights |
From CMS’s February 2024 answers and the CY2026 final rule, applied to the regulations cited in the text.
The case-by-case exception sits beside these patterns. An admission the physician expected to last less than 2 midnights can still be inpatient under § 412.3(d)(3), but only with “medical record support for that determination.” Plans can review that support, and the record decides these cases.
The first pattern deserves the most attention, because it is the hardest to see. A hindsight denial does not announce itself. The letter cites the criteria and the record, and the reviewer may sincerely believe the expectation was unreasonable. The trouble is that the reviewer read the record already knowing how it ended.
What 2026 changes
Decisions during the stay become reviewable
The CY2026 MA final rule, issued in April 2025, expanded the definition of an organization determination to include “MA plan decisions made concurrent to the enrollee’s receipt of services.” A plan that directs a downgrade to observation during a stay is now making a decision with notice and appeal rights attached. The rule also requires plans to give providers notice of these decisions.
The rule’s fact sheet adds a limit at the back end. When a plan has approved an inpatient admission, it may reopen that approval only for obvious error or fraud. The February 2024 answers had already said a plan that pre-authorized an admission “may not deny coverage later on the basis of lack of medical necessity.” The CY2026 change narrows what remains.
The fact sheet describes no change to the admission criteria themselves. Proposed guardrails on AI in coverage decisions were not finalized. Our review of UM committee rules covers how plans govern the criteria they use, and our analysis of the CY2027 rule covers the changes that follow in 2027.
The inpatient-only list starts to disappear
In the CY2026 outpatient payment rule, released in November 2025, CMS began eliminating the inpatient-only list, which is set to end by January 2028. The first step removed 285 mostly musculoskeletal procedures, according to a Holland & Knight summary. In traditional Medicare, those procedures remain exempt from site-of-service denials and Recovery Audit Contractor referrals until CMS determines that a majority are furnished in outpatient settings.
For MA, the change runs through § 422.101(b)(2), which names the inpatient-only list alongside § 412.3. A procedure on the list is inpatient by regulation. A procedure that leaves it must qualify under the benchmark or the case-by-case exception. The traditional Medicare exemption is an instruction to Medicare’s own contractors, like the presumption. On CMS’s reasoning in February 2024, it would not bind plans.
The effect in MA is that surgical admissions move from a list into a judgment. A musculoskeletal procedure that once qualified because of its code now qualifies because of what the surgeon expected and documented. That makes the quality of the review, and the question it asks, matter for more admissions than before.
Traditional Medicare patients gain a status appeal
An October 2024 final rule followed the court order in Alexander v. Azar. Under it, certain traditional Medicare patients who are switched from inpatient to observation during a stay can receive a Medicare Change of Status Notice and request a fast appeal. A retrospective process covered admissions on or after January 1, 2009. Its filing window closed in January 2026, according to CMS.
The appeals “do not extend to enrollees in Medicare Advantage (MA) plans,” as LeadingAge noted. For MA enrollees, the closest equivalent is the CY2026 change that makes a plan’s decision during the stay an appealable organization determination.
Review the expectation blind to the outcome
The two-midnight benchmark asks a question about a moment. The question is whether, at the time of the order and given what the record showed, it was reasonable to expect the patient to need hospital care past a second midnight. A retrospective reviewer answers that question knowing something the physician could not have known, which is how the stay ended.
Outcome knowledge changes clinical judgment, and the evidence for that is decades old. In a 1991 study in JAMA, 112 practicing anesthesiologists judged the appropriateness of care in 21 cases. Each case also existed in a version identical to the original except for a plausible outcome of opposite severity. When the outcome changed from temporary to permanent harm, ratings of appropriate care fell by 31 percentage points. When it changed from permanent to temporary, they rose by 28 points.
The outcome in that study was harm, not length of stay, but the mechanism carries over. A reviewer who knows the patient went home after 1 night is likely to find the reasons for the quick recovery in the record and read them back into the admission. The expectation looks less reasonable because the ending is known. CMS’s own fact sheet treats rapid improvement as unforeseen. To a reviewer who already knows it happened, it can look inevitable.
The rule asks what the physician expected. A reviewer who knows how the stay ended is answering a different question.
Plans can design retrospective review to remove the ending. The approach borrows from the way diagnostic tests are validated, where the reader interprets the result without knowing the answer. Our case for treating prepayment review like a diagnostic test made a similar argument about measuring review accuracy.
- Build the file as of the order. The review file holds the record up to the time of the admission order, along with the physician’s documented reasoning. Discharge times and later notes stay out of the first read.
- Record the judgment first. The reviewer records whether the expectation was reasonable, and why, citing facts from the file. Only then does the system show how the stay ended.
- Mix the queue. If a plan samples only short stays, the sample itself reveals the outcome. Mixing in stays that crossed 2 midnights keeps the selection from leaking the answer.
- Decide during the stay when possible. The cleanest way to avoid hindsight is to decide while the expectation is current. A review in the first hours, with a peer-to-peer conversation before any denial, gives the physician a chance to explain. Since 2026, that decision is appealable. An approval holds except for obvious error or fraud.
- Measure the drift. On a sample, a plan can compare blinded and unblinded reads of the same admissions. The gap between them measures hindsight in the program directly, and it belongs in the same reports as inter-rater reliability.
The design has limits. Records are not always time-stamped cleanly, and a lab result drawn before the order may be charted after it. A file built as of the order has to use the time information became available, not the time it was filed. Blinding also costs reviewer time, and it does nothing about a plan that applies the wrong standard on purpose. It does make the right standard easier to apply honestly.
Plans have a fair objection. Some short stays reflect admissions that should not have happened, and a short stay is a reasonable signal for choosing which cases to review. That is true, and it is compatible with the design. Selection can use the outcome. The judgment should not.
Where Redlake stands
Redlake builds tools that health plans use in utilization review. We hold those tools to 3 rules. A clinician makes every denial, so a machine can say yes, but only a person can say no. Every recommendation shows its reasons. Every decision can be reconstructed.
For inpatient status, the rules translate directly. Our tools can approve an admission when the record supports the physician’s expectation. A denial or a downgrade to observation goes to a physician reviewer. CMS points the same way. Its February 2024 answers said “algorithms or artificial intelligence alone cannot be used as the basis to deny admission or downgrade to an observation stay.”
We think the review file should be built as of the admission order, with the outcome withheld until the reviewer has recorded a judgment and its reasons. The record of the review should show what the reviewer saw and when, so that an appeal can test the decision against the information that existed at the time.
We do not claim this ends disagreement about admissions. Physicians and reviewers will still read the same record differently. The design moves the disagreement to the right question, which is what the physician could reasonably have expected.
What to watch in the rest of 2026
- The next inpatient-only removals. CMS usually proposes its outpatient payment rule in July. The CY2027 proposal is the next step in the phase-out, and its list will show which admissions move into the expectation test in 2027.
- Appeals of decisions made during the stay. The CY2026 rule makes these decisions appealable. Appeal volumes and outcomes would show how often plans downgrade admissions in real time, if plans or CMS report them.
- Oversight of § 422.101(b)(2). Watch whether CMS program audits cite plans for applying criteria stricter than § 412.3, and whether those findings become public.
- Data that separates admissions. Plans’ March 2027 postings will cover 2026 decisions. A breakout for inpatient admissions would let the public measure compliance from the plan side for the first time.
About this analysis
Method. The inpatient admission criteria in 42 CFR § 412.3 and the MA coverage rules that apply them are described from CMS’s February 2024 answers on the CY2024 final rule and from the CY2026 final rule fact sheet. Evidence on observation rates comes from press coverage of the Kodiak Solutions analysis.
Data notes. Kodiak’s figures cover more than 1,900 hospitals and may not represent all hospitals. The 2 press accounts report slightly different upper bounds for MA rates in 2024. The table uses the Becker’s figures, and HFMA reported a range of 14.4% to 16.3%. The JAMA Internal Medicine findings are described from AHA News. The 1991 study measured judgments about harm, not length of stay. The blinded review design is this analysis’s proposal, not a method CMS has published.
Sources. CMS, answers on coverage criteria and utilization management in CMS-4201-F (February 2024) and CY2026 final rule fact sheet (April 2025). CMS, two-midnight rule fact sheet (updated March 2026). 42 CFR §§ 412.3 and 422.101. Becker’s Hospital Review and HFMA (October 2024). AHA News (September 2025). AHA letter to CMS (October 2023). Holland & Knight (November 2025). CMS, patient status appeals page (updated March 2026). LeadingAge (October 2024). Caplan and colleagues, JAMA (April 1991).
Frequently asked questions
Do MA plans have to follow the two-midnight rule?
Yes. Since January 2024, § 422.101(b)(2) has required Medicare Advantage plans to follow the inpatient admission criteria in 42 CFR § 412.3, including the two-midnight benchmark. Plans do not have to follow the two-midnight presumption, which is an instruction to traditional Medicare’s review contractors.
Can a plan deny an admission because the patient went home after 1 night?
Not on that basis. CMS has said whether the admission actually crossed 2 midnights is not a factor in the § 412.3 criteria. CMS also lists rapid clinical improvement among the unforeseen circumstances that can shorten an inpatient stay. A plan may review whether the admitting physician’s expectation was reasonable, based on the medical record.
Can a plan reverse an inpatient admission it approved?
A plan that approved an admission may not later deny it for lack of medical necessity under § 422.138(c). Under the CY2026 rule, a plan may reopen an approved inpatient admission only for obvious error or fraud.
What does the end of the inpatient-only list mean for MA plans?
CMS removed 285 procedures from the list in 2026 and plans to eliminate it by January 2028. In Medicare Advantage, procedures that leave the list must qualify for inpatient payment under the two-midnight benchmark or the case-by-case exception.