The Geriatric Nutritional Risk Index and its association with all-cause mortality in cancer patients with sepsis: a dual-center retrospective cohort study.

Background: The Geriatric Nutritional Risk Index (GNRI) is a validated nutritional assessment tool predictive of outcomes in elderly patients, yet its prognostic value in adult cancer patients with sepsis remains unexplored. Methods: This retrospective cohort study included adult cancer patients with sepsis from two sources: 523 patients from Guangxi Medical University Cancer Hospital ICU (2013-2024) and 4,447 patients from the MIMIC-IV database (2008-2019). Cox multivariable regression was used to analyze the association between GNRI and 28-day/60-day all-cause mortality during ICU stay. Restricted cubic splines (RCS), proportional hazards (PH) assumption testing, and landmark analyses were performed to assess linearity and temporal stability. Predictive performance was evaluated using area under the receiver operating characteristic (AUROC) curve, calibration curves with bootstrap resampling, decision curve analysis (DCA), integrated discrimination improvement (IDI), and net reclassification improvement (NRI). Internal cross-validation was conducted to validate model stability and correct for overfitting. Results: In Cox multivariable regression analysis, GNRI levels were independently associated with reduced mortality in both cohorts. Restricted cubic spline (RCS) curves demonstrated a linear dose-response relationship between GNRI and mortality. PH testing and landmark analysis confirmed the time-independent prognostic value of GNRI. Subgroup analyses confirmed the robustness of these findings. The model achieved excellent discriminative ability (AUC 0.906 in the Guangxi cohort and 0.810 in the MIMIC-IV cohort for 28-day mortality). Calibration curves showed excellent agreement between predicted and observed probabilities. DCA confirmed favorable net clinical benefit across a wide range of risk thresholds. IDI and NRI verified that GNRI significantly improved risk classification. Internal cross-validation showed stable optimism-corrected C-indices with minimal overfitting. Conclusion: Among adult patients with cancer and sepsis, a lower GNRI is associated with increased short- and medium-term all-cause mortality. GNRI showed excellent discriminative ability, good calibration, favorable clinical net benefit, significant improvement in risk classification, and robust stability in internal validation, with consistent performance in external validation. GNRI may serve as a simple, reliable prognostic tool for risk stratification and early nutritional intervention in this high-risk population.