Predicting Adverse Outcomes in Kidney Transplant Recipients Using an Interpretable Model Based on Shear-Wave...

Area of Science:

• Nephrology
• Medical Imaging
• Machine Learning

Background:

• Renal fibrosis is a key factor in kidney transplant allograft deterioration.
• Noninvasive methods are needed to monitor kidney transplant recipients.
• Shear-wave elastography (SWE) shows potential for assessing renal fibrosis.

Purpose of the Study:

• To develop an interpretable machine learning model for predicting allograft deterioration in kidney transplant recipients.
• To evaluate the predictive performance of SWE features in this population.

Main Methods:

• A prospective cohort study included 396 kidney transplant recipients.
• Shear-wave elastography (SWE) examinations were performed.
• An XGBoost model was developed using SWE features and laboratory data to predict a composite outcome of 40% eGFR decline or end-stage kidney disease.

Main Results:

• The XGBoost model incorporating SWE features achieved high predictive performance (AUC 0.870 in training, 0.868 in validation).
• Higher medullary or cortical tissue stiffness on SWE was associated with worse prognosis.
• Elevated medullary SWE (>10 kPa) independently predicted adverse outcomes (aOR 2.68).

Conclusions:

• Combining SWE parameters with laboratory data significantly enhances the prediction of allograft function decline.
• An interpretable XGBoost model using noninvasive SWE shows promise for guiding patient monitoring in kidney transplantation.