Abstract: Objective  The purpose of this article was to analyze the existing evidence on the application of machine learning-based chronic kidney disease mortality risk prediction models in chronic kidney disease-related fields.  Methods  Literature search was conducted using subject words combined with free words in six databases: Wanfang Medicine, China National Knowledge Infrastructure, VIP, Pubmed, Embase, and Web of Science. The included articles contain one or more prediction models based on artificial intelligence.  Result  A total of 918 articles were retrieved, and ultimately 15 articles were included in the analysis. The usage rate of artificial intelligence-based risk prediction models (random forest, support vector machine, and extreme gradient boosting) was the highest (n=11, 73.3%), and the performance of this type of risk prediction model was better. Among them, the AUCs of the extreme gradient boosting models were all greater than 0.8 (95% CI: 0.768～0.832).  Conclusion Prediction models based on machine learning are generally better than traditional prediction methods in predicting death risk in patients with chronic kidney disease.

Key words: Chronic kidney disease, Machine learning, Artificial intelligence, Mortality risk, Predictive modelling