The research, published in the Journal of Eating Disorders, was led by Quang Le and Majd Al-Ahmad of the University of Missouri together with colleagues at several other United States medical centers. The team performed what epidemiologists call a serial cross-sectional analysis, drawing on the National Inpatient Sample for the years 2016 through 2022. That database, maintained as part of the Healthcare Cost and Utilization Project, is the largest publicly available all-payer inpatient database in the United States, capturing a representative sample of hospital discharges that can be statistically weighted to estimate national totals. Because it records every diagnosis coded during each hospital stay, it allows researchers to ask which co-occurring conditions travel with fatal outcomes.

To build their cohort, the investigators identified every adult admission carrying an any-listed diagnosis of anorexia nervosa, using the ICD-10-CM code family F50.0x. Any-listed means the diagnosis appears anywhere in the record, not just as the reason for admission, which matters because many of these patients enter the hospital for medical crises rather than eating-disorder treatment. After survey weighting, the sample represented roughly 61,000 hospital admissions nationwide over the seven-year window. Within that population, 148 patients died before discharge, an in-hospital mortality rate of 1.21 percent. That figure may sound modest, but it is striking for a psychiatric diagnosis in a general hospital population, and it represents only deaths during the indexed admission, not the longer-term mortality that makes anorexia nervosa so feared.

The analytical engine of the study was survey-weighted logistic regression, a technique that accounts for the complex sampling design of the National Inpatient Sample while estimating how strongly each complication is associated with death. The models adjusted for age, sex, and the Charlson comorbidity index, a validated score that summarizes a patient’s burden of chronic illness. Because the researchers were testing many potential risk factors simultaneously, they applied Holm correction, a sequential method that tightens the threshold for statistical significance and guards against false positives when multiple comparisons are made. This detail matters: several associations that looked compelling in a primary model faded once every complication was modeled together, a pattern that honest analysis must surface rather than hide.

The results upend conventional assumptions. Severe malnutrition carried an adjusted odds ratio of 3.32, meaning patients with that diagnosis had more than three times the odds of dying compared with similar patients without it. Hypoglycemia was even more potent, with an adjusted odds ratio of 3.65. Heart failure showed an association in the primary model, with an odds ratio of 2.09, but this attenuated to 2.01 when all complications were modeled together, and the Holm-adjusted p-value of 0.07 fell just short of conventional significance. In other words, the heart’s pumping failure may matter, but the signal is far weaker than the metabolic collapse of starvation itself.

Most provocative of all, the rhythm abnormalities that dominate bedside vigilance showed no association with death at all. Bradyarrhythmia, the pathologically slow heart rate that clinicians watch so closely, had an adjusted odds ratio of 0.82, with a confidence interval spanning well below and above unity. Tachycardia fared no better, at 0.62. Coronary artery disease, another plausible cardiac culprit, was likewise unassociated, at 1.14. The authors are careful about what this does and does not mean. A slow heart rate may look harmless in administrative data precisely because it triggers closer monitoring and intervention, a form of detection bias that observational datasets cannot fully untangle. The finding is not an argument for reducing cardiac monitoring, but it is a strong argument that heart rate alone is a poor proxy for who will die.

One of the starkest findings concerned sex. Women make up the overwhelming majority of anorexia nervosa cases, yet men in the cohort died at roughly twice the rate. After adjustment, female sex carried an odds ratio of 0.42, meaning men had about two and a half times the odds of in-hospital death. The authors note that men did not appear more medically unwell on admission, which makes the disparity harder to explain. Possible contributors include delayed recognition of eating disorders in men, longer time to treatment, and differences in body composition that allow less physiological reserve before starvation becomes lethal. Whatever the mechanism, the signal suggests that clinicians should resist the stereotype that anorexia nervosa is a disease of young women whose presentation is always recognized early.

When the researchers examined what actually killed these patients, the picture sharpened further. Cause of death was approximated by the principal diagnosis and its Major Diagnostic Category, a coarse but informative classification. Among fatal admissions, infection and sepsis led the list, accounting for 27.0 percent of deaths. Most people who died had been admitted for a serious acute medical illness rather than for treatment of the eating disorder itself. This reframes the clinical problem: the hospitalized anorexia nervosa patient in danger is often not the one with a dramatic cardiac rhythm on the monitor, but the one whose starved immune system is quietly losing a battle with pneumonia, a urinary infection, or an occult source of sepsis. Malnutrition impairs cell-mediated immunity, gut barrier integrity, and the febrile response, all of which can mask infection until it is advanced.

The study took care to address a potential confounder of the era: COVID-19. Because the study window overlapped the pandemic, the investigators repeated their analysis after excluding admissions coded for COVID-19, which made up only 1.1 percent of the cohort. The estimates were essentially unchanged, indicating that pandemic-era coding did not distort the central findings. The authors also disclosed that generative artificial intelligence tools were used for language editing and to cross-check diagnosis codes, with the authors verifying all content, and the study was exempt from institutional review board review because it analyzed de-identified, publicly available administrative records.

The practical message for hospital teams is a reordering of priorities. Inpatient vigilance, the authors conclude, should center on the severity of malnutrition, glycemic control, and infectious risk, with comorbidity weighted heavily, rather than on heart rate in either direction. That does not diminish the importance of cardiac monitoring, which remains standard for severely malnourished patients at risk of prolongation of the QT interval and sudden arrhythmic death, outcomes that administrative data may simply fail to capture. But it does suggest that the most actionable warnings of impending death in these patients are metabolic and infectious, not electrical. For a disorder whose mortality has stubbornly resisted decades of clinical attention, a finding this counterintuitive, drawn from tens of thousands of real hospitalizations, deserves to change what clinicians watch first when a starving patient arrives on the ward.

Subject of Research: Acute cardiac and metabolic complications associated with in-hospital mortality in adults hospitalized with anorexia nervosa in the United States

Article Title: Acute cardiac complications and in-hospital mortality in adults hospitalized with anorexia nervosa: a national analysis, 2016–2022

Article References: Le, Q., Al-Ahmad, M., Dangol, G., Maharjan, S., Ahmad, Z., & Bhandari, A. (2026). Acute cardiac complications and in-hospital mortality in adults hospitalized with anorexia nervosa: a national analysis, 2016–2022. Journal of Eating Disorders. https://doi.org/10.1186/s40337-026-01766-9

Image Credits: AI Generated

DOI: 10.1186/s40337-026-01766-9

Keywords: anorexia nervosa, in-hospital mortality, cardiac complications, malnutrition, hypoglycemia, sepsis, bradyarrhythmia, National Inpatient Sample, eating disorders, heart failure, sex differences, hospital medicine