Abstract
BACKGROUND: We used the Predictive Model Bias Risk Assessment tool (PROBAST) tool to systematically evaluate the existing models worldwide, in order to provide a reference for clinical staff to select and optimize DFU recurrence risk prediction models. METHODS: Literature on DFU recurrence risk prediction model construction published in CNKI, China Biomedical Literature Database, Vipu China Knowledge, China Biomedical Literature Database, Vipu Chinese Journal Service Platform, Wanfang Data Knowledge Service Platform, Embase, PubMed, Web of Science, Cochrane Library and other databases were systematically searched. The search period was until January 29, 2024, encompassing all relevant studies published up to that date. Literature screening and data extraction were conducted by two researchers, and the PROBAST was used to evaluate the bias risk and applicability of the included literature. RESULTS: Finally, 9 literatures were included, 13 prediction models were established, and the area under the AUC or C-index ranged from 0.660 to 0.943. Nine models were validated internally and one model was validated externally. All the models constructed in the included literature are of high-risk bias, and the applicability of the models is reasonable. Common predictors in the prediction model were Wagner scale, glycosylated hemoglobin, and diabetic peripheral neuropathy. CONCLUSION: Although most of the existing DFU risk prediction models have good prediction performance, they all have high risk of bias. It is suggested that researchers should update the existing models in the future, and future modeling studies should follow the reporting norms, so as to develop a scientific, effective and convenient risk prediction model that is more conducive to clinical practice.