Head-to-head comparison of routinely ordered inflammatory markers, interleukin-6 and lactate for 28-day mortality in high-acuity ICU sepsis: a predominantly vasopressor-treated cohort

In intensive-care-unit (ICU) sepsis, C-reactive protein (CRP), procalcitonin (PCT), the white-cell count (WBC) and neutrophil-to-lymphocyte ratio (NLR) are commonly ordered and often interpreted prognostically. We compared their discrimination of 28-day mortality with interleukin-6 (IL-6) and lactate in the same patients. This single-centre retrospective cohort included 310 adults with Sepsis-3 sepsis requiring intensive care and an ICU stay of at least 24 h. Of these, 82.3% received vasopressors and 65.2% met a prespecified septic-shock approximation based on vasopressor use and first post-ICU venous lactate > 2 mmol/L. Single-marker analyses used the first value after ICU admission (median 0.6 h, IQR 0.5 to 4.5) and, in parallel, the worst value within 24 h. Discrimination was compared by area under the receiver-operating-characteristic curve (AUC; DeLong test) overall, within severity strata and by admission source. Incremental value of IL-6 and lactate over partially window-aligned clinical models was also assessed. 28-day mortality was 52.6%. Using first values, CRP (AUC 0.504, 95% CI 0.439 to 0.567) and PCT (0.549, 0.486 to 0.614) discriminated at or near chance. NLR had a below-chance point estimate (0.438, 0.371 to 0.501), but its 95% CI included 0.50. IL-6 (0.694, 0.631 to 0.756) and venous lactate (0.682, 0.622 to 0.742) discriminated better. Results were similar with worst 24-h values (all paired DeLong p at least 0.19). CRP and NLR were not significantly correlated with reconstructed SOFA (Spearman rho 0.094 and − 0.032); only 1.3% of first CRP values reached the assay ceiling. In the 92 patients not meeting the septic-shock approximation, IL-6 and lactate AUCs fell to 0.578 and 0.502, respectively. This attenuation meant that severity restriction could not be excluded as an explanation. Adding first IL-6 and lactate to the partially window-aligned clinical model increased AUC from 0.782 to 0.808 (likelihood-ratio p = 0.0004; DeLong p = 0.057). IL-6 and lactate showed greater prognostic discrimination than CRP, PCT and NLR in this high-acuity cohort with ICU stays of at least 24 h and warrant prospective evaluation. Assay censoring did not appear sufficient to explain the limited prognostic information carried by CRP and the neutrophil-based indices, whereas restriction to a narrow, high-severity case mix could not be excluded.