Incremental Prognostic Value Of Neutrophil-To-Lymphocyte Ratio Beyond Serum Lactate For ICU Mortality Prediction: A Reanalysis Of MIMIC-IV Data

Abstract

Background: Serum lactate is an established prognostic biomarker in critically ill patients, reflecting tissue hypoperfusion and metabolic stress. The neutrophil-to-lymphocyte ratio (NLR) has emerged as a marker of systemic inflammation and immune dysregulation. Whether NLR provides incremental prognostic value beyond lactate for intensive care unit (ICU) mortality prediction remains uncertain. We reanalysed MIMIC-IV data to rigorously evaluate the independent and incremental contribution of early NLR measurements to lactate-based mortality prediction.

Methods: We conducted a retrospective cohort study using the Medical Information Mart for Intensive Care IV (MIMIC-IV) database from Beth Israel Deaconess Medical Centre. We included adult patients (≥18 years) admitted to the ICU with serum lactate and complete blood count measurements obtained within the first 24 hours of ICU admission. The primary outcome was in-ICU mortality. We calculated NLR from absolute neutrophil and lymphocyte counts. We constructed multivariable logistic regression models with log-transformed, standardised predictors and evaluated discrimination using receiver operating characteristic (ROC) curve analysis with bootstrap-derived confidence intervals (100 resamples). We assessed incremental value using DeLong tests comparing area under the curve (AUC) estimates. We used bootstrap resampling with 100 iterations to estimate optimism-corrected performance.

Results: Of 65,355 eligible first ICU admissions, 122 patients met inclusion criteria with complete early lactate and NLR measurements. The cohort had a median age of 63 years (IQR 53–71) and experienced 15 in-ICU deaths (12.3%). Non-survivors had significantly higher lactate levels (median 3.60 vs 1.90 mmol/L, p=0.006) and older age (68 vs 62 years, p=0.037) compared to survivors. NLR was elevated in non-survivors but did not reach statistical significance (13.41 vs 7.21, p=0.145). In multivariable logistic regression, lactate remained strongly predictive of mortality (OR 2.50 per 1-SD increase, 95% CI 1.38–4.52, p=0.003), while NLR showed no independent association (OR 1.23, 95% CI 0.69–2.19, p=0.481). Lactate alone achieved an AUC of 0.719 (95% CI 0.566–0.830). Adding NLR increased AUC to 0.752 (95% CI 0.619–0.860), representing a non-significant improvement (ΔAUC +0.033, p=0.155). After adjustment for age and Charlson comorbidity index, the combined model achieved an AUC of 0.783 (95% CI 0.658–0.896). Bootstrap internal validation yielded an optimism-corrected AUC of 0.704 with excellent calibration (slope 1.00, intercept 0.00).

Conclusions: In this reanalysis of MIMIC-IV data, serum lactate measured within 24 hours of ICU admission demonstrated robust independent association with in-ICU mortality. Early NLR did not provide statistically significant incremental prognostic value beyond lactate alone. The high exclusion rate due to missing laboratory data (0.19% of eligible ICU admissions) and absence of severity-of-illness scores limit generalizability. These findings suggest that while NLR may capture inflammatory processes relevant to critical illness, its clinical utility for mortality prediction in the presence of lactate measurements requires further validation in larger, prospectively designed cohorts with comprehensive severity adjustment.