Objective To analyze the risk factors of enteral nutrition feeding intolerance in ICU patients based on machine learning algorithms.

Methods Data of comprehensive ICU patients who received enteral nutrition support from a tertiary grade B hospital from January 2022 to December 2022 were retrospectively collected. According to whether feeding intolerance occurred or not, they were divided into intolerance group (n = 199) and tolerance group (n = 127). Univariate logistic regression was used to filter input indicators. Logistic regression, random forest, decision tree, Naive Bayes and eXtreme Gradient Boosting (XGBoost) algorithms based on machine learning were used to construct predictive models. From January to February 2023, 46 ICU patients supported by enteral nutrition in the same hospital were selected to verify the model. The models constructed by the five methods were compared for predictive value of ENFI in ICU patients. The risk factors were screened based on the best predictive model.

Results The incidence of ENFI in ICU patients was 61.04% (199/326). Among the five machine learning algorithm models, the random forest model had the best predictive performance (AUC = 0.831). The analysis results of the random forest model showed that albumin, length of ICU stay, and Acute Physiology and Chronic Health Evaluation II (APACHE II) score were important influencing factors for ENFI in ICU patients.

Conclusion The incidence of ENFI in ICU patients is relatively high. Clinical medical staff should pay attention to patients with low albumin levels, prolonged ICU stays, and higher APACHE II scores, timely supplement protein-based drugs, and manage feeding tolerance well, in order to reduce the incidence of ENFI.