Design: secondary cohort analysis with Cox regression, restricted cubic splines and a prespecified interaction test
Population: 466 US adults with sepsis and respiratory failure or vasopressor-dependent shock
Primary outcome: 90-day mortality
Effect: HR 2.25 per sd (95% CI 1.76-2.88) without diabetes or atherosclerotic disease; 1.04 (0.83-1.30) with, interaction p<0.001

Baseline soluble tumour necrosis factor receptor 1 was measured in 466 adults with sepsis complicated by respiratory failure or vasopressor-dependent shock, and tested against 90-day mortality with an interaction term for diabetes or atherosclerotic cardiovascular disease. Half the cohort, 52.1%, had one or both.

In patients without either condition, the biomarker performed as expected: a hazard ratio of 2.25 per standard deviation (95% CI 1.76-2.88). In patients with diabetes or atherosclerotic disease, it did not - hazard ratio 1.04 (0.83-1.30), with the interaction highly significant. Their baseline levels were higher to begin with (median 6,547 against 4,883 pg/mL) despite similar illness severity and similar mortality, and the risk curve flattened above about 12,000 pg/mL. The effect was specific: interleukin-6 and angiopoietin-2 showed no such interaction, and obesity and hypertension without diabetes or atherosclerosis produced none either.

This biomarker is not in routine use anywhere, so nothing changes at the bedside tomorrow. The finding matters for what it says about the enterprise. Chronic inflammatory disease raises the baseline and appears to break the relationship between the marker and the outcome, in exactly the half of a septic population that carries the most comorbidity. A prognostic threshold derived in one comorbidity mix will not transfer to another - and in Indian intensive care, where the diabetes prevalence among septic adults is higher again, that is the relevant half.

Interpret any sepsis biomarker threshold against the comorbidity mix it was derived in.
Diabetes or atherosclerotic disease raised baseline levels without raising illness severity or mortality.
The interaction was specific to this receptor; interleukin-6 and angiopoietin-2 behaved the same in both groups.
This marker is not in routine clinical use - nothing changes in current practice.
Obesity and hypertension alone did not produce the same effect.

Why it matters

It suggests biomarker thresholds in sepsis may not be portable between populations with different burdens of chronic disease.

The statistics, in plain English

An interaction test asks whether an association differs between groups, and here it does so decisively - a hazard ratio of 2.25 in one group and 1.04 in the other, with the second confidence interval comfortably spanning 1.0. The plateau above 12,000 pg/mL comes from a spline model, which finds shape in the data flexibly and can also overfit it; with 466 patients split in two, the exact location of that plateau is not firm.