Joint trajectories of nutritional risk and fluid balance and prognosis in critically ill adults: a dual-trajectory modeling study using MIMIC-IV and eICU.

Background: Nutritional risk and fluid balance evolve rapidly during critical illness, but their joint longitudinal patterns and prognostic relevance remain incompletely characterized. This study identified early joint trajectories of GNRI-derived nutritional risk and fluid balance and evaluated their associations with mortality in critically ill adults. Methods: This retrospective dual-database study used MIMIC-IV as the development cohort and eICU as an independent reproducibility cohort. Adult ICU patients with an ICU stay >=3 days and at least three paired daily measurements of GNRI-derived nutritional risk and weight-standardized fluid balance during ICU days 1-7 were included. Nutritional risk was defined as 98 minus daily GNRI. Group-based multi-trajectory modeling identified joint phenotypes. Multivariable logistic and Cox regression models evaluated associations with hospital and short-term mortality. Sensitivity analyses included landmark analyses, extended covariate adjustment, proportional-hazards diagnostics, posterior-classification assessment, and incremental prediction evaluation. Results: The analysis included 1,243 patients from MIMIC-IV and 9,912 from eICU. Three reproducible joint trajectory phenotypes were identified: persistent moderate-to-high nutritional risk with gradual deresuscitation (Group A), initially severe nutritional risk with high early fluid load and rapid decline (Group B), and worsening nutritional risk with mild-to-moderate deresuscitation (Group C). Group B had the highest crude mortality in both cohorts, with hospital mortality of 40.0% in MIMIC-IV and 34.7% in eICU. In fully adjusted models, Group B was associated with higher 30-day mortality in MIMIC-IV (HR 1.59, 95% CI 1.12-2.28) and higher hospital mortality in eICU (OR 1.95, 95% CI 1.55-2.45). Group C was also associated with increased 90-day mortality in MIMIC-IV (HR 1.37, 95% CI 1.10-1.72) and hospital mortality in eICU (OR 1.29, 95% CI 1.14-1.46). Conclusion: Early joint trajectories of GNRI-derived nutritional risk and fluid balance identified reproducible prognostic phenotypes in selected critically ill adults with sufficient repeated measurements. The phenotype combining severe early nutritional risk with high initial fluid load showed the highest mortality risk. Incremental prediction analyses indicated modest additional prognostic information beyond baseline GNRI and early fluid balance, supporting dynamic nutrition-fluid phenotyping as a complement to conventional severity assessment.