Huijiao Chen ¹ Jin Han ¹ Jing Li ¹ Jianhua Xiong ¹ Dong Wang ¹ Mingming Han ¹ Yuehao Shen ^1* Wenli Lu ^2*

• ¹ Tianjin Medical University General Hospital, Tianjin, China
• ² Tianjin Medical University, Tianjin, Tianjin Municipality, China

Objective: To thoroughly examine studies on feeding intolerance risk prediction models for enteral nutrition patients. Design: Conducted a systematic review and meta-analysis of observational studies. Methods: Searched databases including CNKI, Wanfang, VIP, SinoMed, PubMed, Web of Science, Cochrane, CINAHL, and Embase from inception to August 12, 2024. Extracted data on study design, subjects, follow-up, data sources, outcomes, sample size, missing data handling, continuous variable methods, variable selection, predictors, model development, performance, and presentation. Evaluated applicability and bias risk using PROBAST. Results: Included 18 models from 14 studies. One study used multiple machine-learning techniques; others used logistic regression. FI incidence in enteral nutrition was 32.4%-63.1%. Top predictors: APACHE II, age, albumin, intra-abdominal pressure, mechanical ventilation. AUC ranged from 0.70 to 0.921. All studies had high bias risk due to inappropriate data sources and inadequate reporting. Conclusion: Studies reported discriminatory power in FI prediction models, but PROBAST deemed high bias risk. Future research should focus on innovative models with larger, diverse samples, strict designs, and multicenter validation.