A Hybrid Model Combining Survival Analysis, Knapsack Optimization and Supervised Learning to Extrapolate the Evolution of Kidney Transplantation Patients from Donors with Expanded Criteria After Controlled Circulatory Death
F. Santos Arteaga1, D. Di Caprio2, O. Bestard3, N. Montero4, F. Moreso3, M. Crespo5, C. Facundo6, J. Reinoso-Moreno7, D. Cucchiari7, B. Bayes7, E. Poch7, J. M. Campistol7, F. Oppenheimer7, F. Diekmann7, I. Revuelta7
1Faculty of Economics and Management, Universidad Complutense de Madrid, Madrid, Spain, 2Department of Economics and Management, University of Trento, Trento, Italy, 3Department of Nephrology and Kidney Transplant, Hospital Universitari Vall Hebrón, Barcelona, Spain, 4Department of Nephrology and Kidney Transplant, Hospital Universitari de Bellvitge, Barcelona, Spain, 5Department of Nephrology and Kidney Transplant, Hospital del Mar, Barcelona, Spain, 6Department of Nephrology and Kidney Transplant, Fundació Puigvert, Barcelona, Spain, 7Department of Nephrology and Kidney Transplant, Hospital Clinic, Barcelona, Spain
Meeting: 2022 American Transplant Congress
Abstract number: 1802
Keywords: Circulatory Death, Kidney transplantation, Prediction models
Topic: Clinical Science » Organ Inclusive » 72 - Machine Learning, Artificial Intelligence and Social Media in Transplantation
Session Information
Session Name: Machine Learning, Artificial Intelligence and Social Media in Transplantation
Session Type: Poster Abstract
Date: Tuesday, June 7, 2022
Session Time: 7:00pm-8:00pm
Presentation Time: 7:00pm-8:00pm
Location: Hynes Halls C & D
*Purpose: Kidney transplantation (KT) with expanded criteria donors (ECD) after controlled circulatory death (cDCD) in high-risk patients is being debated. We categorize patients via a hybrid model based on their evolution through transplantation, identify main variables leading to negative outcomes per patient, and generate clusters of patients and donors defined by their characteristics relative to those determining transplant outcomes
*Methods: Multicenter, retrospective study. Recipient CV events and age used to categorize scenarios in cDCD/DBD KT. We combine survival analysis, knapsack optimization and machine learning(ML) techniques. Variables of greater significance are identified via survival and used to compute a knapsack optimization-based index and a summary metric per donor-recipient pair. A battery of ML techniques and an Artificial Neural Network(ANN) are applied to extrapolate the performance of donor-recipient pairs based on initial characteristics. Study approved by all Local Ethical Committees
*Results: All consecutive deceased-donor KT performed in 5 transplant centers (Jan/2013-Dec/2017; FU:Dec/2019). Out of 1,281 patients, 1,161 were included (DBD:74.3%;cDCD:25.7%). Similar mean age in donors and recipients (61.8±14.4; 60±12.5 years); Hypertension(89.1%), diabetes(28.1%) and previous CV events(20.1%) in recipients, mostly first KT(86%) and low pretransplant DSAs rate(6.03%). Direct implementation of survival analysis delivers rates of observed to predicted deaths at three years below 50% for all high-risk categories. Our accuracy percentages are higher than 75% for the set of donor-recipient pairs generated, with k-nearest neighbors and Gaussian support vector machines performing above 80% together with the ANN
*Conclusions: The knapsack optimization process incorporates a matching procedure where each potential donor-recipient pair is considered as part of both the optimization problem and the subsequent supervised learning process. This quality allows to define average performances per patient or select the best potential paired alternative
To cite this abstract in AMA style:
Arteaga FSantos, Caprio DDi, Bestard O, Montero N, Moreso F, Crespo M, Facundo C, Reinoso-Moreno J, Cucchiari D, Bayes B, Poch E, Campistol JM, Oppenheimer F, Diekmann F, Revuelta I. A Hybrid Model Combining Survival Analysis, Knapsack Optimization and Supervised Learning to Extrapolate the Evolution of Kidney Transplantation Patients from Donors with Expanded Criteria After Controlled Circulatory Death [abstract]. Am J Transplant. 2022; 22 (suppl 3). https://atcmeetingabstracts.com/abstract/a-hybrid-model-combining-survival-analysis-knapsack-optimization-and-supervised-learning-to-extrapolate-the-evolution-of-kidney-transplantation-patients-from-donors-with-expanded-criteria-after-contr/. Accessed December 12, 2024.« Back to 2022 American Transplant Congress