Preoperative Computed Tomography Aortic Calcification and Attenuation of the Vertebral Body Predict Fractures Post-Kidney Transplantation Using Machine Learning

Abstract

Introduction

Post-transplant fractures are associated with increased mortality and reduced graft survival. Although computed tomography (CT)-derived metrics, such as vertebral trabecular bone attenuation and abdominal aortic calcification have been reported as useful for assessing osteoporosis, their utility in predicting post-transplant fractures in kidney transplant recipients (KTRs) has not been explored. This study examined pre-transplant factors, including CT imaging parameters, associated with post-transplant fractures using machine learning techniques.

Methods

This retrospective study included 111 primary living-donor KTRs who underwent preoperative CT at our hospital between 2009 and 2023. Aorto-iliac calcification was quantified using the Agatston score. Vertebral trabecular bone CT attenuation values (Hounsfield units [HU]) were recorded at T12 to L5 levels. Machine learning with SHapley Additive exPlanations (SHAP) was used to identify the top predictors of fracture.

Results

The median follow-up period was 69 months. Twenty-four recipients experienced fractures: hip (n = 3), leg (n = 6), arm (n = 5), rib (n = 3), and vertebrae (n = 9). The L1 vertebral CT attenuation value and aortic segment calcification score emerged as the top predictors of fracture. Based on SHAP cutoffs (L1 CT = 104 HU; Agatston score = 504), recipients were categorized into Group 1: high L1/low calcification; Group 2: high L1/high calcification; and Group 3: low L1/high calcification. Compared to Group 1, fracture risk was higher in Group 2 (Subdistribution hazard ratio [SHR] 3.06; p = 0.034) and markedly higher in Group 3 (SHR 11.66; p < 0.0001).

Conclusion

Pre-transplant lower vertebral trabecular bone attenuation and higher aortic calcification on CT are strongly associated with post-transplant fractures.