The Autopsy / Case File · Medical Error & Preventable Care
Sent Home With It
The most dangerous thing an emergency department can do with a septic patient is decide they aren't sick enough to keep. Federal-scale data says that decision doesn't land evenly.
Sepsis is a timing disease. Every hour that passes between the infection turning systemic and the first dose of antibiotics measurably raises the odds of death, which makes the emergency department's disposition decision — admit, observe, or discharge — one of the highest-stakes judgment calls in American medicine. A cross-sectional study of 3.68 million acute-care encounters found that the diagnostic testing behind that call was ordered at significantly different rates depending on the patient's race, and that the difference existed almost entirely in the emergency department. Here's what the data actually shows.
Join the discussion →
Sepsis does not present as sepsis. It presents as a bad flu, a urinary tract infection, a skin infection that looks a little angrier than it did yesterday, a person who is tired and warm and breathing a bit fast. The mechanism that kills is invisible at the bedside: the body's own response to an infection turns against its circulatory system, blood pressure falls, organs stop getting the perfusion they need, and a patient who looked survivable at 9 p.m. is in septic shock by morning. The clinical literature is unusually blunt about what determines the outcome, and it is not the sophistication of the intensive care available later. It is whether somebody recognized it early and started antibiotics.
The Test That Doesn't Get Ordered
In 2024, a team led by researchers at Johns Hopkins published a cross-sectional analysis in JAMA Network Open covering 3,683,055 acute-care encounters across Kentucky, Maryland, North Carolina, and New Jersey between 2016 and 2018. They looked at four of the most common and most diagnostically ambiguous ways a person arrives at a hospital: nausea and vomiting, abdominal pain, chest pain, and syncope — the presentations where an infection turning septic hides most easily, and where the decision to test or not test is most discretionary.
The finding was consistent and one-directional. Compared with white patients presenting the same way, Black patients had an adjusted odds ratio of 0.74 for receiving diagnostic testing in the emergency department — roughly a quarter lower odds of getting the workup that would find something. The authors were careful about what that does and doesn't prove, writing that their results "raise the concern that undertesting among Black patients may lead to more missed diagnoses." What it establishes on its own is narrower and still significant: at the point in the encounter where a clinician decides how hard to look, they did not look equally hard.
The Gap Closes at the Admission Door
The most forensically useful part of that study isn't the emergency department number. It's the comparison. The same analysis measured the same testing decision at two other points of care, and the disparity behaved very differently at each. For patients in observation stays, the adjusted odds ratio was 0.92 — still a gap, but a narrower one. For admitted inpatients, it was 1.00. Exactly parity. Once a patient was through the door and in a hospital bed, the difference in how thoroughly they were investigated disappeared entirely.
That pattern points somewhere specific. A disparity that persisted across every setting would suggest something about the patients — different presentations, different comorbidities, different care-seeking behavior. A disparity that exists only in the emergency department and vanishes on the inpatient ward suggests something about the setting: the time pressure, the volume, the incomplete history, the reliance on rapid pattern-matching by a clinician who has never met this person and has four other patients waiting. The emergency department is where medicine does its fastest thinking, and fast thinking is where documented bias has the most room to operate.
In the emergency department, the odds of being tested were 26% lower. On the inpatient ward, the odds were identical. The gap wasn't in the patients. It was in the room.
What the Delay Looks Like in Minutes
A separate research group put a clock on the downstream consequence. Investigators from Harvard Medical School and the Harvard Pilgrim Health Care Institute — Pak, Sánchez, McKenna, Rhee, and Klompas — retrospectively identified more than 48,000 adults admitted to five Massachusetts hospitals between 2015 and 2022 with suspected sepsis or septic shock, defined by an objective clinical trigger rather than a billing code: blood cultures drawn and intravenous antibiotics started within 24 hours of arrival. Then they measured how long each patient waited for that first dose. The work was published in Critical Care Medicine in 2024.
Black patients with sepsis waited a median of 212 minutes for antibiotics. White patients waited 193. In septic shock — the more obviously critical presentation, where urgency is hardest to miss — the gap narrowed but did not close: 157 minutes versus 147. After detailed risk adjustment, Black patients had 1.23 times the adjusted odds of falling into the three-to-six-hour antibiotic window rather than the zero-to-three-hour window for sepsis, and 1.27 times the odds in septic shock. Women waited longer than men on the same measure, and women with septic shock had higher in-hospital mortality that persisted even after accounting for the antibiotic delay.
The Study That Didn't Find It
Honest surveillance means reporting the analyses that cut the other way, and there is one worth putting directly alongside the others. In 2025, a team publishing in the American Journal of Emergency Medicine examined exactly the scenario this case file is built around: patients discharged home from the emergency department with an infection, who then came back sick enough to be admitted. Across 10,179 unique patients at two University of California San Diego emergency departments between 2019 and 2022, 113 — 1.11% — returned within seven days for admission with sepsis, severe sepsis, or septic shock.
What predicted that return was largely physiological and structural rather than demographic. Arriving by ambulance rather than walking in carried 2.55 times the odds. Every 10-beat-per-minute increase in heart rate at the index visit carried 1.29 times the odds. Cancer, hypertension, cardiovascular disease, and chronic kidney disease all independently raised the risk. Race and ethnicity showed no statistically significant association in the final model, and insurance type was not retained in it at all. One hospital system, one metro area, and 113 events is a thin base for detecting a modest disparity even if one exists — but the finding stands as reported, and it belongs in the record next to the ones that found a gap.
What Actually Closed It
The most consequential result in this literature is not a measurement of the problem. It's an intervention that moved it. Researchers publishing in Academic Emergency Medicine in 2025 studied 8,151 sepsis patients across a Midwestern health system — one academic Level I trauma center and two community hospitals — for 18 months before and 18 months after the July 2020 rollout of a standardized sepsis order set. The order set was not a new drug or a new machine. It was a checklist: the standard labs, imaging, and blood cultures for suspected infection, plus fluid resuscitation and antibiotic orders selected against local resistance patterns, bundled so that ordering them was the default rather than a series of individual decisions.
Before the order set, Black race was independently associated with higher sepsis mortality in multivariable analysis, with an odds ratio of 1.41. After it, that association was 1.10 and no longer statistically significant. The share of patients receiving antibiotics within three hours rose across the board, but it rose furthest for the patients who had been waiting longest: from 57.5% to 74.1% among Black patients, compared with 64.7% to 73.3% among white patients. The two groups converged at roughly the same place. Removing discretion from the front of the process removed the gap that discretion had been producing.
What the Data Doesn't Settle
None of these studies followed a single patient from an emergency department waiting room to a death certificate, and none of them claim to. The testing study measured orders, not outcomes. The antibiotic study measured minutes among patients who were all eventually admitted, which means it never saw the people who were sent home. The return-admission study looked directly at the discharged group and found no racial signal in a sample too small to rule out a modest one. The order set study is the closest thing here to a causal argument, and it is a single health system's before-and-after.
What survives across all four is narrower than a proven chain of causation and harder to explain away than a single anomalous result: the diagnostic effort at the front door varies by race in a way it does not vary once a patient is admitted, the treatment clock runs measurably slower for the same patients, and when a health system replaced individual judgment with a default order set, both the timing gap and the mortality association compressed toward parity. That last finding is the one with an operational implication, because it is the only one that describes something a hospital can actually do on a Monday.
And Then You're Dead
Sepsis is one of the few conditions where American medicine knows precisely what saves the patient and precisely how fast it has to happen. The failure documented here isn't a knowledge gap or a technology gap. It's a triage judgment, made in a few minutes by a tired clinician in a crowded room, that the evidence says isn't being made the same way for everybody — and a standardized checklist, in at least one health system, was enough to make the difference mostly go away.
That's it. That's the whole thing.
Sources
- 3,683,055 encounters across Kentucky, Maryland, North Carolina and New Jersey, 2016–2018; adjusted odds of diagnostic testing for Black vs. white patients of 0.74 in the emergency department, 0.92 in observation, and 1.00 inpatient; the authors' "undertesting…may lead to more missed diagnoses" conclusion — Ellenbogen MI, Weygandt PL, Newman-Toker DE, et al., "Race and Ethnicity and Diagnostic Testing for Common Conditions in the Acute Care Setting," JAMA Network Open 2024;7(8):e2430306
- Median time-to-antibiotics of 212 minutes for Black vs. 193 for white sepsis patients (157 vs. 147 in septic shock); adjusted odds of 1.23 and 1.27 for the 3–6 hour window; women's longer times and higher septic shock mortality; more than 48,000 adults, five Massachusetts hospitals, 2015–2022 — Pak TR, Sánchez SM, McKenna CS, Rhee C, Klompas M, "Assessment of Racial, Ethnic, and Sex-Based Disparities in Time-to-Antibiotics and Sepsis Outcomes in a Large Multihospital Cohort," study abstract; published in full as Critical Care Medicine 2024;52(12):1928–1933
- 10,179 emergency department patients discharged with an infection at two UC San Diego Health sites, 2019–2022; 113 (1.11%) returning within 7 days for sepsis admission; ambulance arrival OR 2.55, heart rate OR 1.29 per 10 bpm; race, ethnicity and insurance not significant in the final model — Chen AY, et al., "Risk factors associated with return sepsis admission following emergency department discharge with infection," American Journal of Emergency Medicine 2025;97:207–215
- 8,151 patients across one academic Level I trauma center and two community hospitals, 18 months before and after a July 2020 sepsis order set; Black-race mortality odds ratio falling from 1.41 to a non-significant 1.10; antibiotics within 3 hours rising from 57.5% to 74.1% among Black patients vs. 64.7% to 73.3% among white patients — Fernandez Olivera ML, et al., "Decreased racial disparities in sepsis mortality after an order set–driven initiative: An analysis of 8151 patients," Academic Emergency Medicine 2025;32(6):598–603
Photo: "Trinity Health - St. Joseph Mercy Livingston Hospital - Emergency Room - Howell - Michigan" by Tony Webster, via Wikimedia Commons (CC BY-SA 2.0).
Comments
Loading…