Association between iron metabolism biomarkers and all-cause mortality in critically ill patients with sepsis: a retrospective cohort study.
TL;DR
Iron metabolism biomarkers provide significant incremental prognostic value for 30-day mortality in sepsis patients and integrating them into machine learning models enhances risk stratification beyond traditional clinical scores, offering a clinically actionable tool for early identification of high-risk patients.
Abstract
Background
Iron metabolism disorders were frequently observed in sepsis, yet their association with outcomes remained incompletely understood. Therefore, this study aimed to evaluate the prognostic value of iron-related biomarkers and to develop models for predicting 30-day mortality in septic patients.
Methods
This retrospective cohort study analyzed 2729 sepsis patients from the Medical Information Mart for Intensive Care IV database to evaluate the prognostic value of iron metabolism biomarkers and to develop machine learning-based prediction models for 30-day mortality. We assessed associations using Kaplan-Meier analysis, multivariable Cox regression, and restricted cubic spline regression. Six machine learning models were developed and internally validated using 100 bootstrap resamples.
Results
The 30-day mortality rate was 19.57%. Ferritin levels above 680 ng/mL were associated with the strongest mortality risk increase, while transferrin above 160 mg/dL demonstrated a protective effect. Restricted cubic spline analysis revealed nonlinear relationships, with serum iron exhibiting an inverted U-shaped association between 35 and 328 μg/dL. The Light Gradient Boosting Machine model incorporating iron biomarkers achieved a validation area under the receiver operating characteristic curve of 0.782 (0.743-0.819), with significant reclassification improvement (Net Reclassification Improvement: 0.076, p = 0.006).
Conclusion
Iron metabolism biomarkers provide significant incremental prognostic value for 30-day mortality in sepsis patients. Integrating them into machine learning models enhances risk stratification beyond traditional clinical scores, offering a clinically actionable tool for early identification of high-risk patients.