Predictive Models for Acute Kidney Injury in Acute Pancreatitis: A Systematic Review and Meta-Analysis

Haoran ZhuShengteng GuoQinghua Wang^*

• Binzhou Medical University, Binzhou, China

Background:The utilisation of predictive models facilitates the identification of patients at risk, thereby enabling the implementation of individualised interventions. Despite the growing use of predictive models to estimate the likelihood of AKI in AP, concerns persist regarding their effectiveness in clinical settings and the rigor and relevance of forthcoming research. The objective of this study is to systematically review and evaluate predictive models for AKI in AP. Methods: A comprehensive search of relevant databases was conducted, encompassing China National Knowledge Infrastructure (CNKI), Wanfang, VIP, Chinese Medical Association, PubMed, Web of Science, Scopus, and Cochrane Library, with the search extending from database inception to 26 November 2024. The data from a number of selected studies was extracted using the CHARMS form, whilst the quality of predictive modelling studies was assessed using the PROBAST tool. A meta-analysis of AUC for predictive models and relevant predictors (≥2) was conducted using Stata 17.0 and MedCalc software. Results: The total number of studies included in the review was 17, with a total of 9,949 patients and 37 predictive models. Of these, 32 models underwent internal validation, with an area under the curve (AUC) > 0.7. The overall risk of bias was high across all 17 studies, yet the overall applicability was deemed satisfactory. The results of the meta-analysis indicated that the pooled AUC for internal validation across 20 predictive models for AKI in AP was 0.790 (95% CI = 0.761– 0.818); and the pooled external validation AUC for five models was 0.766 (95% CI = 0.684–0.845). The overall risk of bias was high across all 17 studies, with significant heterogeneity observed. However, the overall applicability was deemed satisfactory. Conclusion: The predictive model for AKI complicating AP demonstrates moderate predictive efficacy. Nevertheless, given the elevated risk of bias in the majority of studies and the absence of adequate external validation, its clinical applicability merits further investigation. Systematic review registration: https://www.crd.york.ac.uk/PROSPERO/view/CRD420251008769, identifier CRD420251008769.