Forecasting of Patient-Specific Kidney Transplant Function With a Sequence-to-Sequence Deep Learning Model.

Area of Science:

• Nephrology
• Artificial Intelligence
• Biomarker Analysis

Background:

• Kidney transplant recipients exhibit variable estimated glomerular filtration rate (eGFR) trajectories.
• Distinguishing between graft dysfunction and normal fluctuations is challenging using current methods.

Purpose of the Study:

• To develop and validate a deep learning model for predicting patient-specific eGFR reference ranges post-kidney transplant.
• To assess the model's accuracy in identifying deviations from expected eGFR trajectories.

Main Methods:

• A sequence-to-sequence deep learning model was trained on eGFR data from a large derivation cohort (933 patients, 100,867 measurements).
• The model's predictive performance was evaluated in two independent test cohorts (1,170 patients, 39,999 measurements).
• Model predictions were compared against conventional autoregressive integrated moving average (ARIMA) models.

Main Results:

• The deep learning model accurately predicted patient-specific eGFR trajectories within 3 months post-transplant (RMSE: 6.4-8.9 mL/min/1.73 m2).
• Sequence-to-sequence model predictions significantly outperformed ARIMA models across various input/output scenarios.
• The model demonstrated robust performance in independent validation cohorts from Belgium and Germany.

Conclusions:

• A validated deep learning model can forecast individual kidney transplant function.
• Patient-specific eGFR predictions can aid clinicians in identifying significant graft function changes.
• This approach offers a more personalized method for monitoring kidney transplant recipients than population-based reference ranges.