Background: After kidney transplantation, clinicians often face uncertainty when interpreting whether a decline in eGFR is within the patient's expected range of fluctuation or if it signals a significant deviation requiring further investigation. This study aims to assess whether an enhanced machine learning model can reliably predict the patient-specific expected eGFR range post-transplant, acting as an early warning system and facilitating timely interventions, such as biopsies to prevent early graft rejection and immunosuppression adjustments.
Materials and Methods: This study evaluates the efficacy of a Gradient Boosting model in predicting post-transplant GFR. It included 75 pediatric patients aged 1-18 years who underwent kidney transplantation between 2016 and 2023 at Baskent University Hospital, Ankara, Turkey. The dataset consisted of 2,609 eGFR measurements, as well as patient demographics and transplant-related data. Key features like 'Days to Transplantation' (GFR values before the transplantation), 'Days from Transplantation' (GFR values up to 30 days post-transplantation), patient age, gender, and donor types were algorithmically selected for their predictive value. The model was specifically developed to predict GFR values during the first three months post-transplantation.
Results: The median age of the patients was 13.8 years (IQR: 8.6-16.3), with males constituting 62.7% and females 37.3%. The machine learning model achieved a Mean Absolute Error (MAE) of 4.06 mL/min/1.73 m², demonstrating its precision in estimating eGFR values. Furthermore, the model attained a Root Mean Squared Error (RMSE) of 7.58 mL/min/1.73 m², underscoring its effectiveness in capturing the variability of eGFR post-transplantation. Cross-validation procedures reinforced the model's robustness, yielding a Cross-Validated RMSE (CV RMSE) of 8.90 mL/min/1.73 m².
Conclusions: Gradient Boosting's application underscores the promise of advanced machine learning techniques in refining GFR prediction after kidney transplantation. With its augmented precision, the model can support clinicians in making informed decisions regarding early biopsies and interventions, thus highlighting the vital role of sophisticated analytical methods in medical prognosis and the monitoring of pediatric patient care.