Bayesian Analysis of Modified NUTRIC Score for Mortality Prediction in Critically Ill Patients

BACKGROUND: Malnutrition is a critical determinant of outcomes in critically ill patients, significantly influencing mortality rates. The Modified Nutrition Risk in Critically Ill (mNUTRIC) score is a widely utilized tool for assessing nutritional risk and predicting mortality in intensive care unit (ICU) settings. However, heterogeneity in study designs, cutoff thresholds, and clinical contexts has contributed to inconsistent conclusions regarding its diagnostic accuracy. Addressing these challenges, this study leverages Bayesian analysis to provide a comprehensive evaluation of the mNUTRIC score's predictive performance. The primary objectives were to estimate its diagnostic accuracy for mortality prediction, evaluate its performance across diverse thresholds and mortality timeframes, and address the heterogeneity observed in existing studies. METHODS: A Bayesian meta-analytic framework was employed to synthesize data from 31 studies encompassing 13,271 critically ill patients. Posterior distributions were generated for key diagnostic metrics, including sensitivity, specificity, area under the curve (AUC), and likelihood ratios. Subgroup analyses were conducted to investigate the influence of different cutoff thresholds (<5 and ≥5) and mortality timeframes (28-day mortality vs. longer-term mortality). Heterogeneity was assessed using hierarchical models, and publication bias was evaluated with statistical tests to ensure robustness. RESULTS: The mNUTRIC score demonstrated high sensitivity (0.78, 95% CrI: 0.74–0.82) and moderate specificity (0.67, 95% CrI: 0.63–0.71), yielding an overall AUC of 0.80 (95% CrI: 0.76–0.83). These findings indicate that the score is a reliable tool for predicting mortality in critically ill patients. Subgroup analyses revealed consistent performance across varying thresholds (<5 and ≥5) and mortality timeframes, underscoring the mNUTRIC score's adaptability to different ICU settings. Hierarchical modeling identified moderate heterogeneity, attributable to study-level variability in design and population characteristics, but did not significantly impact the overall conclusions. No evidence of publication bias was observed (p=0.8). CONCLUSION: This Bayesian analysis reaffirms the mNUTRIC score as a robust and adaptable predictor of mortality in critically ill patients. Its consistent diagnostic performance across diverse settings and thresholds highlights its utility for guiding nutritional interventions and optimizing ICU resource allocation. These findings support the mNUTRIC score's integration into routine clinical practice as a valuable tool for risk stratification. Future research should prioritize validating its efficacy in underrepresented populations and exploring its impact on long-term clinical outcomes and quality of life in ICU survivors. Such efforts will further enhance the evidence base for the mNUTRIC score's role in improving critical care delivery.