In 2019, UnitedHealth Group published a one-page brief claiming that two-thirds of hospital ED visits, 18 million a year, are avoidable, at a cost of $32 billion. The number has been repeated in pitch decks and vendor blogs ever since. Read the actual PDF and the second footnote quietly gives the game away: 27 million of 46 million annual visits were classified, and 19 million were left out of the math. Two-thirds of the classified subset is 39% of the whole. And the population behind it is privately insured, which tells you close to nothing about a Medicare book.
That footnote is a good place to start, because almost everything written about non-emergent ER use has the same shape: a real problem, measured loosely, rounded up. The problem deserves better numbers. Non-emergent visits are real, they are expensive, and in a risk-bearing contract they come straight out of your medical spend. But the popular picture of the problem, a waiting room full of sore throats that belonged in urgent care, points at the wrong population, the wrong intervention, and the wrong week.
What counts as a non-emergent ER visit
The taxonomy most analyses trace back to is the NYU ED algorithm, built by John Billings in the late 1990s from clinician review of roughly 6,000 full ED records. It sorts visits into four buckets:
- Non-emergent. Care was not needed within 12 hours. A timing judgment.
- Emergent, primary-care treatable. Care was needed promptly, but a primary care setting could have delivered it. A setting judgment.
- Emergent, ED needed, preventable. The ED was the right place that day, but adequate ambulatory care upstream would have prevented the episode. This is the ambulatory-care-sensitive bucket: diabetes, COPD, asthma, heart failure.
- Emergent, ED needed, not preventable. The visit was always going to happen.
Two things about the algorithm are routinely ignored. First, it assigns each discharge diagnosis a probability distribution across those buckets, not a label; no single visit is ever "non-emergent" under it. Second, its own developers have said it was never meant to judge individual visits, a point a 2024 JAMA Network Open commentary made bluntly after two decades of misuse.
The deeper problem is that discharge diagnosis is the wrong lens entirely. A 2013 JAMA study found that only 6.3% of ED visits ended with a primary-care-treatable diagnosis, but the presenting complaints behind those visits showed up in 88.7% of all ED visits. Chest pain that turns out to be reflux looks identical, at the front door, to chest pain that turns out to be an infarct. That is why the prudent layperson standard exists: since the Balanced Budget Act of 1997, plans must cover emergency visits based on the symptoms the member showed up with, not the diagnosis they left with. Non-emergent ER use is a care-management target. It cannot be a payment target, and the plans that tried are the cautionary tales later in this post.
How many ER visits are actually non-emergent
Pick your instrument and you can land almost anywhere:
- By discharge diagnosis: 6.3% of visits are primary-care treatable (JAMA, 2013).
- By site substitutability: 13.7% to 27.1% of visits could have been handled at an urgent care center or retail clinic, worth roughly $4.4 billion a year (Health Affairs, 2010).
- By triage acuity: in the CDC's 2022 NHAMCS tables, 19.4% of all visits were triaged semiurgent and 1.8% nonurgent, with the caveat that about a quarter of visits carry no recorded triage level.
- By payer classification: UHG's 39%-once-you-read-the-footnote, privately insured only.
The honest summary is a range, roughly 6% to 27% depending on what question you are actually asking, with every method agreeing on one thing: the number is well short of "most."
The Medicare picture is different
The US logged 155 million ED visits in 2022, 47 per 100 people. Medicare beneficiaries went at a rate of 56 per 100; adults 75 and older, 76 per 100. People 65 and older made 32.9 million visits, about a fifth of the national total.
But look at what those older patients came in with. Among visits by patients 65 and older, 8.8% were triaged semiurgent and 1.4% nonurgent, about 10% combined, versus 21% across all ages. Meanwhile 40.3% were urgent and 15.4% emergent, both far above the all-ages shares.
This is the part most ED-diversion content gets backwards. In a Medicare Advantage book, the sore-throat-in-the-ER problem is small. The expensive, recurring, genuinely avoidable pattern is the third NYU bucket: the COPD member who went from a missed refill to a 2 a.m. exacerbation, the heart failure member whose weight gain nobody caught. The visit was emergent by the time it happened. The failure happened in the two weeks before.
There is an uncomfortable corollary for MA specifically. A 2023 JAMA Health Forum analysis of 10.6 million beneficiaries found MA plans had 44% higher ED direct-discharge rates for ambulatory-care-sensitive conditions than traditional Medicare and 2.4 times the observation stays. Admissions were slightly lower, but adding the three settings together, MA carried 30% more potentially avoidable acute care. Some of what MA reports as avoided admissions is the same unmanaged exacerbation, relabeled as an ED discharge or an observation stay. It still costs money, and it still marks a member whose condition nobody is steering.
What a non-emergent ER visit costs
Use numbers whose construction you can explain. AHRQ's HCUP brief puts the average resource cost of a treat-and-release ED visit at $750 in 2021, across 107.4 million visits and $80.3 billion in aggregate cost. For Medicare patients the average is $1,040; for patients 65 and older, $1,110. On the commercial side, the Peterson-KFF tracker found an average total cost of $2,453 per visit, of which 80% was facility fees; even the least complex visits averaged $592.
These are different constructs. HCUP estimates what the visit consumed in resources; KFF measures what was allowed on commercial claims. Both are real, neither is the other, and any analysis that quietly swaps between them, or leans on a payer's own hand-picked comparison, is selling something.
Why members use the ER for non-emergencies
Four forces, each with evidence behind it:
- Appointment lag. The average wait for a physician appointment across large metros hit 31 days in 2025. A member who cannot get seen this week solves the problem tonight.
- After-hours gaps. The ED is the only door that is always open. Extending primary care hours helps, but only on the slice you would expect: a large UK difference-in-differences study found evening and weekend access cut patient-initiated visits for minor problems by 26%, and total ED visits by a statistically insignificant 3%.
- Social needs. In Medicare survey data, an unmet health-related social need was not associated with a single ED visit but was associated with repeat visits (adjusted odds ratio 1.47), and transportation barriers alone raise ED use in older adults by about a third.
- Habit. Frequent users, 4.5% to 8% of ED patients, account for 21% to 28% of all visits, and prior ED use is one of the strongest predictors of the next visit. HEDIS draws its Medicare 65+ outlier line at just 4 visits a year.
The measures that pay: FMC, not EDU
Two HEDIS measures live in this territory, and operators regularly confuse them. EDU (Emergency Department Utilization) is the risk-adjusted observed-to-expected ED visit ratio. It is reportable, but it is not a Star Ratings measure; it appears on neither the 2026 nor the 2027 CMS measure list. FMC (Follow-Up After ED Visit for People With Multiple High-Risk Chronic Conditions) is: a process measure, weight 1, on both lists. It pays when a member with multiple high-risk chronic conditions gets a qualifying follow-up within 7 days of the ED visit, and the qualifying list is generous: an outpatient visit, a telephone visit, an e-visit, transitional care management, or case management all count.
The design of that measure is worth reading twice. CMS does not rate you on preventing the ED visit. It rates you on whether anyone touched the member in the week after. The neighboring measures point the same direction: Plan All-Cause Readmissions is an outcome measure at weight 3, and Transitions of Care scores four notification-and-follow-up steps on every inpatient discharge. The rated surface of this entire problem is outreach speed.
What doesn't work
The record on the tempting shortcuts is unusually clear.
Retrospective denials. Anthem tried denying claims it judged non-emergent in 2017 and denied about 12,200 across three states in six months, per a Senate staff report. Of the denials that were appealed, 62% were overturned, with Missouri's monthly overturn rate climbing from 58% to 73%, and the policy was carved down with exceptions. UnitedHealthcare announced a similar policy in 2021 and shelved it within weeks. A JAMA Network Open analysis explains why these keep failing: 87.9% of commercially insured ED visits share presenting symptoms with visits that ended in non-emergent diagnoses, yet 65.1% of those received emergency-level care on the day.
Steering to urgent care. A Health Affairs study of about 29 million enrollees over 12 years found it took 37 additional urgent care visits to displace one lower-acuity ED visit, and each $1,646 visit averted came with $6,327 of added urgent care spend. Retail clinics showed the same pattern: 58% of low-acuity retail visits were new utilization, not substitution. Convenience supply creates its own demand.
Alerts without outreach capacity. Washington State's mandated statewide EDIE program, the most studied ADT-alert intervention in the country, cut ED visits among frequent Medicaid users by 8.2% in year one. By year two the effect was statistically indistinguishable from zero, and expenditures never moved. The alert tells you the visit happened. It does not call the member. The programs the widely circulated savings claims were built on self-reported their results; the peer-reviewed evaluation is the one worth reading.
What works: the 72-hour window
The index ED visit is a signal with a half-life. The national 72-hour revisit rate runs 4.5% to 5.1%; among older adults discharged from the ED, roughly one in five is back within 30 days, and single-site cohorts consistently find the return risk peaking in the first 24 hours. Older adults are also where the stakes concentrate: they account for 18.3% of all ED revisits but 38.8% of high-risk revisits, the kind that end in the ICU or the OR, with 2.5 times the adjusted odds of a serious adverse outcome on return.
The intervention evidence matches the clock. Structured nurse-led calls after discharge cut ED visits 28% at 7 days but only 12% at 30; the benefit lives in the first week and decays from there. A randomized patient-navigation trial for frequent users cut treatment-group ED visits 13.2% against 4.3% in controls, and dropped ED-plus-hospital costs 26.6% against 17.5%, while raising PCP visits; the margins are real but modest, which is worth saying out loud. The evidence is not unanimous either, one trial of telephone follow-up in older adults found no effect, but every positive result shares a feature: the touch came fast, and it was a conversation rather than a letter.
The operational bind is one an IPA medical director will describe to you unprompted: the alert feed already delivers the signal. An ADT feed flags every ED registration and discharge in near real time. What most care management teams lack is not the signal but the capacity to act on it inside the window: more discharges on Monday morning than a team can call by Friday, and a triage habit of starting with the sickest, which quietly deprioritizes exactly the members whose visit was preventable and will be again. The visits keep landing after the first issue is solved, because keeping a member engaged after the acute episode is harder than reaching them during it.
How Pelica handles non-emergent ER use
Pelica's Care Management Copilot watches the ADT feed and works every ED discharge rather than a triaged subset. It reaches each member by the channel they answer, AI voice call, text, or a routed task to a human care manager, inside the window that both the revisit data and the FMC measure care about, and it keeps the follow-up cadence going after the first contact instead of closing the loop at one call. Care managers stop spending their week deciding whom to call and start spending it on the calls that need a human. Customers pass 70% on Transitions of Care within 30 days of going live, and the same discharge-to-outreach machinery is what moves FMC's 7-day clock.
Sources and methodology
Every figure above comes from a primary source: CDC/NHAMCS 2022 web tables and data briefs for visit volumes and acuity, AHRQ HCUP Statistical Brief #311 for costs, the NCQA EDU and FMC specifications, the CMS 2026 and 2027 Star Ratings measure lists, and the peer-reviewed studies linked in context. Where a familiar industry number failed verification, the UHG two-thirds claim, the Washington State savings figures, we said so rather than repeating it. Payer-published cost comparisons were excluded by policy.