The average emergency room bill in 2023 was $5,675. What actually got paid out was $1,435. I looked into federal survey data to find out what ER care really costs, and how much the answer depends on why you're there and who's paying.
Every year, Americans make tens of millions of trips to the emergency room, and almost nobody can say what one costs. Hospital "chargemaster" prices are fiction, and what insurers negotiate is opaque. So I went to the closest thing the U.S. has to ground truth: the Medical Expenditure Panel Survey (MEPS), a federal survey that records what was actually paid for each medical visit by insurers and patients combined.
All figures are survey-weighted national estimates computed from 4,241 sampled ER visits in MEPS 2023.
I compared four of the most common broad reasons for an ER visit: injuries & trauma, cardiac & circulatory events, neurological events, and digestive problems. Together they account for roughly 28 million visits a year.
The first finding: the average and the typical visit are wildly different numbers. The mean is dragged upward by a minority of catastrophic cases, and how far it's dragged varies by category:
Cardiac visits have the most extreme skew: the typical circulatory visit costs about $505 (many are chest-pain scares that end with tests and a discharge) but the average is 3.4× higher, pulled up by the visits that turn out to be the real thing. Digestive visits show the same pattern at higher cost: a long tail of GI bleeds and obstructions makes them the most expensive of the four on average ($2,398), despite a modest typical cost. Injuries are the most predictable: high volume, and the smallest gap between typical and average.
Broad categories hide the interesting variation, so here's the drill-down — every specific condition inside each group with a large enough sample to report:
One number here looks bizarre, and it's the most instructive in the whole dataset. A heart attack averages just $696 in the ER, less than half of a hypertension visit. That's not because heart attacks are cheap. It's because heart attack patients get admitted: the meter stops on the ER event and the real costs — cath lab, surgery, ICU — land on the inpatient bill, which is a different file entirely. It's a perfect example of why knowing where a dataset's boundaries are matters as much as analyzing it.
For every category, hospitals charged several times what was ultimately paid. The markup isn't uniform, either — neurological visits were billed at 4.7× what payers settled for, the steepest of the four:
If you've ever received a terrifying ER bill, this is the mechanism: the sticker price is an opening offer that virtually no payer accepts. The gap is also why "average ER cost" figures in the media vary so wildly — it depends entirely on whether the source counted charges or payments.
The same category of visit costs dramatically different amounts depending on coverage. Privately insured patients' visits consistently involve higher payments than publicly insured ones (Medicare/Medicaid) — and for digestive visits the gap is a startling 5.1×:
Part of this is negotiating power — public programs set administered prices, private insurers negotiate from the chargemaster. Part is patient mix. Either way, the payment attached to an identical diagnosis varies more by insurance card than by almost any clinical factor in this data.
Three things. First, "the cost of an ER visit" is not one number — the typical visit ($591) and the average ($1,315) tell different stories, and both are honest. Second, billed charges are a fiction — a 4× markup on average — so any cost figure is meaningless until you know whether it's charges or payments. Third, datasets have edges: the "cheap heart attack" isn't a finding, it's a boundary — and finding those boundaries is most of what real data analysis is.
Methods. Data: MEPS 2023 public-use files (HC-248E ER visits, HC-249 conditions, HC-248I condition–event link, HC-251 person file), Agency for Healthcare Research and Quality. Conditions grouped by AHRQ CCSR categories; "subsequent encounter" suffixes trimmed for display. All estimates weighted with PERWT23F; point estimates only (complex-survey standard errors not computed). Visits with missing payment excluded; per-condition figures require ≥30 sampled visits; a visit with multiple linked conditions counts under each. ER-event payments only — costs after inpatient admission are out of scope. Analysis in Python/pandas: notebook & code.