Background: The prognostic abilities of malnutrition assessment tools for the critically ill are still controversial. This study aimed to assess the predictive ability of MUST, NRS-2002, and NUTRIC tools to predict malnutrition risk for enterally fed, mechanically ventilated patients in intensive care.
Methods: In a multicenter, prospective, observational study, patients from five ICU units in Jordan were observed at two stages. During the first 24 hours of admission, MUST, NUTRIC, and NRS-2002 scores were obtained in addition to the demographic and admission characteristics. In the assessment stage, on day 6th of admission forward, the Bioelectrical Impedance Analysis (BIA), including body compositions and Phase Angle (PhA) were assessed. Machine Learning (ML), structural equation modeling (SEM), and area under the curve (AUC) were used for measuring malnutrition estimates.
Results: A total of 709 patients were observed. At admission, NUTRIC, MUST, and NRS-2002 were congruent in identifying high malnutrition risk (45.1%, 46.4%, and 53.6%, respectively). In reference to PhA, MUST and NRS-2002 scored higher for sensitivity (77.6%, and 76.8%, respectively) and specificity (93.3%, and 90.7%, respectively). They reported acceptable correlation estimates by SEM (0.65, and 0.70, respectively), and ML (0.90, and 0.91, respectively). Further, MUST was the best to discriminate malnutrition, followed by NRS-2002 and NUTRIC (AUC= 0.76, 0.64, and 0.53, respectively).
Conclusions: Alongside validated BIA technology, MUST and NRS-2002 functioned as reliable prognostic indicators of malnutrition risk in the ICU.