Interleukin-6 has long fascinated hepatologists because it is a proximal cytokine, meaning it appears early in inflammatory signaling and helps orchestrate much of what follows. In cirrhosis, the liver is structurally scarred and functionally compromised, and episodes of decompensation, such as ascites, bleeding, jaundice, or hepatic encephalopathy, often mark the transition to a phase of illness with high short-term mortality. Inflammation is thought to be a key driver of these acute deteriorations, whether triggered by bacterial infection, alcoholic hepatitis, or sterile systemic inflammation. Yet despite decades of research, the question of whether measuring interleukin-6 at the bedside adds anything beyond well-established severity scores has remained unresolved. The new study was designed specifically to answer that question using both frequentist and Bayesian statistical frameworks.

The numbers from the cohort are striking. The median admission interleukin-6 level was 23.8 picograms per milliliter, a value far above what is seen in healthy individuals, reflecting the intense inflammatory state that accompanies decompensated cirrhosis. Nine percent of the patients died during their hospital admission, while overall mortality across the follow-up period reached 70.6 percent, underscoring how lethal advanced liver disease remains even in a specialized center. Against this backdrop, the researchers asked whether a single admission measurement of interleukin-6 could predict in-hospital death independently of age, sex, and established disease severity measures.

The answer was yes, but with important qualifications. After statistical adjustment for age, sex, and disease severity, each increment in interleukin-6 was associated with a 34 percent increase in the odds of dying in the hospital, with an adjusted odds ratio of 1.34 and a 95 percent confidence interval of 1.08 to 1.67. Crucially, the confidence interval excludes one, indicating a statistically significant association, but its lower bound sits close to the threshold of no effect, hinting at a signal that is real yet not overwhelming. Bayesian analysis reinforced this conclusion: the posterior probability that interleukin-6 truly predicts in-hospital mortality was 99.6 percent, a figure that will reassure statistically minded readers that the association is very unlikely to be a fluke of this particular dataset.

The researchers then went a step further and asked whether interleukin-6 improves prediction when layered on top of the scores clinicians already use. Adding the cytokine to conventional severity scores produced a modest but statistically significant improvement in discrimination for in-hospital mortality. The best-performing model combined an organ failure score with interleukin-6 and achieved an area under the receiver operating characteristic curve of 0.802, meaning the model distinguished survivors from non-survivors reasonably well. For context, an area under the curve of 0.5 represents chance-level prediction, while 1.0 represents perfect discrimination. A value of 0.802 is respectable for mortality prediction in acutely ill patients, though the incremental gain over existing scores was incremental indeed, not transformative.

Perhaps the most clinically interesting finding concerns etiology, the underlying cause of the liver disease. The prognostic power of interleukin-6 was substantially stronger in patients with alcohol-associated liver disease than in those with metabolic dysfunction-associated steatotic liver disease, the fatty liver condition now epidemic worldwide. In sub-cohort analyses, the etiology-specific effect persisted even after adjustment for infection and after the researchers excluded patients with severe alcohol-associated hepatitis, suggesting the signal is not simply a proxy for one dramatic inflammatory subtype. Mediation analysis revealed that infection accounted for 17.3 percent of the overall association between interleukin-6 and mortality, but the mediation fraction differed sharply by etiology: 23 percent in alcohol-associated liver disease compared with just 2 percent in metabolic dysfunction-associated steatotic liver disease. In other words, in alcohol-related cirrhosis, much of what interleukin-6 captures is infection-driven inflammation, whereas in fatty liver disease the cytokine’s signal operates largely through other pathways.

The study also mapped where the incremental value of interleukin-6 lives on the clinical landscape. Its advantage over MELD-3.0, the current standard severity score for end-stage liver disease, was confined to patients with alcohol-associated disease and, within that group, to those who had infections or intermediate MELD-3.0 scores. This is a meaningful refinement. Patients at the extremes of MELD-3.0, either very sick or relatively stable, are already well stratified by the score alone, and a cytokine measurement adds little. It is the ambiguous middle, where clinicians genuinely struggle to judge short-term risk, that interleukin-6 appears to illuminate, and only in the alcohol-associated subgroup.

Equally important is what the study found about time. The independent prognostic signal of admission interleukin-6 attenuated for longer-term outcomes, meaning the cytokine predicts death in the near term, during the hospitalization itself, but loses its independent predictive power as the follow-up horizon extends. This makes biological sense. Interleukin-6 is a snapshot of an acute inflammatory state, and acute states resolve, evolve, or are overtaken by other processes such as progressive liver failure, cancer, or cardiovascular disease. A marker measured once at admission should not be expected to forecast events months or years later, and the data confirm that expectation.

The authors are careful about what their findings do and do not support, and their caution is warranted. Admission interleukin-6 did not outperform established severity scores, and the study’s conclusions explicitly do not endorse routine measurement of the cytokine in unselected cirrhosis admissions. Widespread adoption would add cost and complexity for a benefit that, in most patients, is negligible. Instead, the study identifies a specific niche where interleukin-6-informed risk stratification merits further work: alcohol-associated cirrhosis, particularly when infection is present or MELD-3.0 scores fall in the intermediate range. The authors call for prospective validation and economic evaluation in exactly this setting before any change to clinical practice.

For the broader field, the study is a model of methodological transparency in a literature often criticized for overclaiming biomarker effects. By combining frequentist and Bayesian approaches, testing mediation by infection, stratifying by etiology, and explicitly bounding the time window in which the signal operates, the researchers have converted a simple question, does interleukin-6 predict death in cirrhosis, into a precise answer: modestly, in the hospital, in alcohol-associated disease, and mainly where infection or intermediate severity blurs the picture. That kind of specificity is what turns an interesting biomarker into a clinically usable one, and it gives trial designers a clear target population should interleukin-6-guided care ever be tested prospectively.

Subject of Research: Prognostic value of admission interleukin-6 in hospitalized patients with cirrhosis

Article Title: Admission interleukin-6 as a time-limited, etiology-dependent prognostic signal in hospitalized cirrhosis – A frequentist and Bayesian cohort study

Article References: Oommen, T. T., Philips, C. A., Theruvath, A. H., Ravindran, R., Sreemohan, A., Baby, A., Ahamed, R., Varghese, J., Philip, B. T., Tharakan, A., & Augustine, P. (2026). Admission interleukin-6 as a time-limited, etiology-dependent prognostic signal in hospitalized cirrhosis – A frequentist and Bayesian cohort study. PLOS One, 21(10), e0360005. https://doi.org/10.1371/journal.pone.0360005

Image Credits: AI Generated

DOI: 10.1371/journal.pone.0360005

Keywords: cirrhosis, interleukin-6, in-hospital mortality, MELD-3.0, alcohol-associated liver disease, metabolic dysfunction-associated steatotic liver disease, Bayesian analysis, inflammation, prognosis, infection, biomarker, hepatology