Incremental prognostic value of inflammatory biomarkers for 28-day mortality in patients with sepsis presenting to the emergency department: a retrospective 24-hour landmark cohort study

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Abstract Background Inflammatory biomarkers are associated with outcomes in sepsis, but prognostic association does not necessarily imply additional predictive information beyond clinical severity. We evaluated the incremental prognostic value of procalcitonin (PCT), the neutrophil-to-lymphocyte ratio (NLR), and the monocyte-to-lymphocyte ratio (MLR) at a 24-hour landmark. Methods This single-center retrospective cohort included 248 adults admitted through the emergency department with sepsis between January 2024 and April 2026. The primary analysis included 243 patients alive 24 h after arrival and used 50 multiply imputed datasets. Fixed-horizon logistic models compared a clinical base model comprising sex, albumin, lactate, and APACHE II with models that added continuous PCT alone or PCT, NLR, and MLR together. Internal validation used ordinary bootstrap optimism correction with 1000 patient-level resamples and multiple imputation nested within resampling; within each resample, separate MICE runs completed the incomplete bootstrap sample and the original cohort used to evaluate the bootstrap-fitted models (five imputations each). A 0-3-point equal-weight inflammatory score based on within-cohort Youden cut-offs was examined as a secondary exploratory analysis. Results Forty-six patients in the landmark cohort died by day 28. Exploratory pooled comparisons did not provide evidence that adding PCT improved the clinical base model ( P = 0.775) or that adding PCT, NLR, and MLR together improved it ( P = 0.428). The odds ratios per doubling were 0.943 (95% CI 0.817–1.089) for PCT, 1.350 (0.937–1.943) for NLR, and 0.866 (0.681–1.101) for MLR. Optimism-corrected AUCs were 0.749, 0.743, and 0.728 for M0, M1, and M2, respectively. The corrected AUC difference between M2 and M0 was −0.022 (95% CI −0.030 to 0.023). The apparent association of the fixed-cut-off score was no longer supported after repeating the complete cut-off search procedure (selection-adjusted P = 0.295). A first-laboratory sensitivity analysis retaining all 248 patients and 51 deaths yielded a similar pattern (exploratory pooled P = 0.951). Conclusions PCT, NLR, and MLR did not provide evidence of incremental predictive value in this exploratory single-center cohort. The within-cohort equal-weight inflammatory score was subject to substantial cut-off selection uncertainty and remains exploratory. Neither the model nor the score should guide clinical decisions before independent evaluation.