ABSTRACT
Outcome prediction for live-donor kidney transplantation improves clinical and patient decisions and donor selection. However, the concurrently used models are of limited discriminative or calibration power and there is a critical need to improve the selection process. We aimed to assess the value of various artificial intelligence (AI) algorithms to improve the risk stratification index. We evaluated pre-transplant variables among 66 914 live-donor kidney transplants (performed between 01/12/2007-01/06/2021) from the United Network of Organ Sharing database, randomized into training (80%) and test (20%) sets. The primary outcome measure was death-censored graft survival. We tested four machine learning models for discrimination (time-dependent concordance index, CTD, and area under the ROC curve) and calibration (integrated Brier score, IBS). We used decision curve analysis to assess the potential clinical utility. Among the models, the deep Cox mixture model showed the best discriminative performance (AUC = 0.70, 0.68, and 0.68 at 5, 10, and 13 years post-transplant, respectively). CTD reached 0.70, 0.67, and 0.66 at 5, 10, and 13 years post-transplant. The IBS score was 0.09, indicating good calibration. In comparison, applying the Living Kidney Donor Profile Index (LKDPI) on the same cohort produced a CTD of 0.56 and an AUC of 0.55-0.58 only. Decision curve analysis showed an additional net benefit compared to the LKDPI, 'Treat all' and 'Treat None' approaches. Our AI-based deep Cox mixture model, termed Live-Donor Kidney Transplant Outcome Prediction outperforms existing prediction models, including the LKDPI, with the potential to improve decisions for optimum live donor selection by ranking potential transplant pairs based on graft survival. This model could be adopted to improve the outcomes of paired exchange programs.
ABSTRACT
In kidney transplantation, pairing recipients with the highest longevity with low-risk allografts to optimize graft-donor survival is a complex challenge. Current risk prediction models exhibit limited discriminative and calibration capabilities and have not been compared to modern decision-assisting tools. We aimed to develop a highly accurate risk-stratification index using artificial intelligence (AI) techniques. Using data from the UNOS database (156,749 deceased kidney transplants, 2007-2021), we randomly divided transplants into training (80%) and validation (20%) sets. The primary measure was death-censored graft survival. Four machine learning models were assessed for calibration (integrated Brier score [IBS]) and discrimination (time-dependent concordance [CTD] index), compared with existing models. We conducted decision curve analysis and external validation using UK Transplant data. The Deep Cox mixture model showed the best discriminative performance (area under the curve [AUC] = 0.66, 0.67, and 0.68 at 6, 9, and 12 years post-transplant), with CTD at 0.66. Calibration was adequate (IBS = 0.12), while the kidney donor profile index (KDPI) model had lower CTD (0.59) and AUC (0.60). AI-based D-TOP outperformed the KDPI in evaluating transplant pairs based on graft survival, potentially enhancing deceased donor selection. Advanced computing is poised to influence kidney allocation schemes.
