Clinical and Radiological Fusion: A New Frontier in Predicting Post-Transplant Diabetes Mellitus

Area of Science:

• Nephrology
• Endocrinology
• Radiology
• Artificial Intelligence

Background:

• Post-Transplant Diabetes Mellitus (PTDM) is a common complication after kidney transplantation.
• Existing risk stratification models for PTDM often lack precision.
• Novel approaches are needed to identify kidney transplant recipients at high risk for PTDM.

Purpose of the Study:

• To develop and validate a predictive model for PTDM using integrated clinical and radiological data.
• To assess the utility of deep learning-based body composition analysis from CT scans in PTDM prediction.
• To compare the performance of the integrated model against a clinical-only model.

Main Methods:

• Retrospective analysis of 2,005 non-diabetic kidney transplant recipients across three Mayo Clinic sites.
• Integration of clinical data with deep learning analysis of pre-transplant CT scans to quantify body composition (adipose tissue, muscle mass).
• Multivariable logistic regression and C-statistic calculation to evaluate predictive performance.

Main Results:

• 16.7% of recipients developed PTDM within one year.
• PTDM patients were older, had higher BMI, triglycerides, and differed in race and sex.
• Key predictors included age, family diabetes history, race, and visceral adipose tissue (VAT) area.
• The integrated model (C-statistic 0.724) outperformed the clinical-only model (C-statistic 0.68).

Conclusions:

• A predictive model integrating clinical data and CT-based body composition analysis accurately identifies kidney transplant recipients at risk for PTDM.
• Visceral adipose tissue (VAT) is a significant predictor of PTDM.
• This enhanced predictive capability can facilitate timely interventions and potentially improve patient outcomes, including reduced mortality.