Development and Validation of the Composite Renal–Inflammatory–Nutritional–Metabolic Index for Predicting 28-Day Mortality in Critically Ill Patients

Background/Objectives: This study aimed to develop and internally validate the RIN-IM model, which integrates routinely measured biomarkers reflecting renal dysfunction, systemic immune-inflammatory activation, albumin-related alterations during acute illness, and metabolic stress, for prediction of 28-day mortality in critically ill adults. Methods: This retrospective cohort study included 1000 adults admitted to the intensive care unit (ICU) between January 2020 and June 2025. The first available demographic, clinical, and laboratory measurements obtained within 24 h of ICU admission were analyzed. The five prespecified RIN-IM predictors—serum creatinine, serum albumin, SII, PIV, and the TyG index—were initially evaluated as continuous variables; SII and PIV were natural-log-transformed, and potential nonlinear associations were examined using restricted cubic splines. Ridge-penalized logistic regression was used to address coefficient instability arising from correlation between SII and PIV. After evaluation of the primary continuous model, a simplified point-based score was derived by assigning 0–2 points to clinically ordered categories for each component, producing a total score of 0–10. Scores were categorized as low- (0–3), intermediate- (4–6), or high-risk (7–10). The primary outcome was 28-day all-cause mortality. Model performance was assessed using discrimination, calibration, Brier score, formal paired AUC comparisons, incremental-value testing, and decision-curve analysis. The complete development procedure was internally validated using 1000 bootstrap resamples. Results: Overall, 300 of 1000 patients (30.0%) died within 28 days. Mortality increased across the RIN-IM risk categories, from 7.6% (32/420) in the low-risk group to 27.2% (98/360) in the intermediate-risk group and 77.3% (170/220) in the high-risk group (p < 0.001). After adjustment for the clinical reference variables, each one-point increase in the simplified RIN-IM Score was associated with higher odds of 28-day mortality (adjusted OR, 1.68; 95% CI, 1.54–1.84; p < 0.001). The primary continuous model had an apparent AUC of 0.907 (95% CI, 0.885–0.929) and an optimism-corrected AUC of 0.898. The corresponding values for the simplified score were 0.890 (95% CI, 0.861–0.919) and 0.878, respectively. The apparent difference between the continuous and simplified models was 0.017 (95% CI, 0.007–0.028; p = 0.001). The simplified RIN-IM Score had higher discrimination than APACHE II (ΔAUC, 0.070; 95% CI, 0.038–0.102; p < 0.001) and SOFA (ΔAUC, 0.090; 95% CI, 0.057–0.123; p < 0.001). Adding RIN-IM to the clinical reference model increased the apparent AUC from 0.821 to 0.923 (ΔAUC, 0.102; 95% CI, 0.078–0.126; p < 0.001) and improved the Brier score from 0.160 to 0.112. Conclusions: In this single-center cohort, the RIN-IM model showed good discrimination and calibration for predicting 28-day mortality. The simplified score retained most of the predictive information of the continuous model and may complement established clinical severity scores using routinely available early laboratory measurements. External validation and clinical-impact studies are required before routine implementation.