External validation of the feeding intolerance prediction model (NOFI) in critically ill patients: A post hoc analysis of a large-scale randomized controlled trial.

Background: Feeding intolerance (FI) is associated with poor prognosis of critically ill patients. This study aims to externally validate the performance of a FI model (NOFI), which was developed using primary diagnosis, Acute Physiology and Chronic Health Evaluation II (APACHE II) score, and acute gastrointestinal injury (AGI) grade to predict FI occurrence in critically ill patients. Methods: The NOFI model was developed using a retrospective cohort and was preliminarily validated in a prospective study. A post hoc analysis was then conducted using data from a multi-center randomized controlled trial to select patients who initiated early enteral nutrition as an external validation dataset. The predictive performance of NOFI was evaluated in terms of discrimination, calibration, and clinical utility. FI risk stratification was defined as follows: The NOFI prediction probability >0.85 (high risk), 0.30-0.85 (middle risk), or <0.30 (low risk). Multivariate Cox regression was performed to explore the association between FI risk and 28-day mortality. Results: Among the 1545 patients included, 856 developed FI. The area under the receiver operating characteristic curve (AUROC) for NOFI was 0.723 (95% confidence interval [CI]: 0.697 to 0.748). The prediction model was able to accurately predict FI, with a sensitivity of 75.8%, specificity of 61.1%, positive predictive value of 70.8%, and negative predictive value of 67.0%. FI risk stratification based on NOFI was associated with 28-day mortality ( Conclusions: NOFI demonstrated reasonable performance in the prediction of FI. Patients at high risk for FI as stratified by the NOFI had a higher risk of 28-day mortality.