To explore the risk factors of enteral feeding intolerance in critically ill patients, build a risk prediction model and verify it, in order to provide reference for early identification and screening of high-risk groups

Based on the previous literature study, the risk factors of enteral feeding intolerance in critically ill patients were obtained, and the general demographic, disease and treatment information of patients were collected. Four machine learning algorithms, namely traditional logistic regression, random forest, support vector machine and naive Bayes, were used to construct risk prediction models, and the optimal model was selected and verified by comparing the model performance

Inclusion Criteria:

• The nutritional status of the patient is assessed by the doctor and the need for enteral nutrition is determined;Patients with first nasogastric tube insertion and successful enteral feeding;There was no gastrointestinal discomfort before enteral nutrition;Patients (or family members) authorized to participate in this study and signed informed consent

Exclusion Criteria:

• Previous gastrointestinal disease or gastrointestinal surgery;enteral nutrition through gastrostomy tube or jejunostomy tube;discontinuation or suspension of enteral nutrition for special reasons other than feeding intolerance;Patients with serious lack of clinical data

The sample size calculation method of this study was based on the results of previous literature studies to obtain enteral nutritional feeding intolerance According to the sample size estimation method of N= [(risk variable) ×(5 ~ 10)]

,the incidence of feeding intolerance was 41.3%,so at least 269 samples are needed for further examination.Considering a 10% loss of follow-up rate, the estimated sample size is 299 cases, the data collected in this part of the study as the model Develop the data set.Using the ratio of modeling group: test group = 7:3, and considering 10% loss of follow-up rate, the minimum sample of test group was 143 cases. Therefore, a total of at least 442 samples are needed in this study.