Scoping Review of Prognostic Prediction Models for Endovascular Abdominal Aortic Aneurysm Repair

Numerous prediction models have been developed to support the challenge of preoperative risk stratification and postoperative follow-up for endovascular repair (EVAR) of infrarenal abdominal aortic aneurysms. This scoping review aims to determine which models have clinical utility based on predictive performance.

An a-priori protocol based on the Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews framework was developed. PubMed, Embase, and Cochrane Library were searched for studies published between 2000 and 2022. Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies guidelines, Prediction model Risk of Bias Assessment Tool and Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) tools were adapted for data extraction.

Among 8,055 screened citations, 52 development and 48 external validation studies met the eligibility criteria, identifying 98 models. The Vascular Study Group of New England (VSGNE) model outperformed popular models such as Glasgow Aneurysm Score, Giles, and Vascular Governance North West for predicting short-term mortality. The Vascular Quality Initiative 1-year survival model model showed superior performance for long-term mortality but requires external validation. Several models predicting other complications (Müller-Wille et al., Kaladji et al., Mid-Essex Artificial Neural Network) showed strong discrimination but require further validation. In the ruptured endovascular aneurysm repair (rEVAR) subgroup, the Dutch Aneurysm Score model best predicted short-term mortality but performance was variable between studies. TRIPOD adherence averaged 63.59% and 51.58% for development and validation studies respectively. Risk of bias was highest in analysis domains.

Short-term mortality in elective EVAR could be reliably predicted using the VSGNE model. However, clinically relevant models for rEVAR, long-term mortality and reintervention lack external validation, and issues such as inadequate handling of missing data, limited use of shrinkage methods, and poor calibration reporting hinder their clinical utility. Robust multicenter studies focused on externally validating existing models are essential to address these limitations.