Bayesian Modeling of Pretransplant Variables Accurately Predicts Kidney Graft Survival

Abstract

Introduction: Machine learning can enable the development of predictive models that incorporate multiple variables for a systems approach to organ allocation. We explored the principle of Bayesian Belief Network (BBN) to determine whether a predictive model of graft survival can be derived using pretransplant variables. Our hypothesis was that pretransplant donor and recipient variables, when considered together as a network, add incremental value to the classification of graft survival. Methods: We performed a retrospective analysis of 5,144 randomly selected patients (age ≥18, deceased donor kidney only, first-time recipients) from the United States Renal Data System database between 2000 and 2001. Using this dataset, we developed a machine-learned BBN that functions as a pretransplant organ-matching tool. Results: A network of 48 clinical variables was constructed and externally validated using an additional 2,204 patients of matching demographic characteristics. This model was able to predict graft failure within the first year or within 3 years (sensitivity 40%; specificity 80%; area under the curve, AUC, 0.63). Recipient BMI, gender, race, and donor age were amongst the pretransplant variables with strongest association to outcome. A 10-fold internal cross-validation showed similar results for 1-year (sensitivity 24%; specificity 80%; AUC 0.59) and 3-year (sensitivity 31%; specificity 80%; AUC 0.60) graft failure. Conclusion: We found recipient BMI, gender, race, and donor age to be influential predictors of outcome, while wait time and human leukocyte antigen matching were much less associated with outcome. BBN enabled us to examine variables from a large database to develop a robust predictive model.

Predictive Score for Posttransplantation Outcomes

“…Despite previous studies having recognized pretransplantation risk factors for kidney allograft loss or mortality 15,17–25 , only few previous prediction score based solely on only these variables, 29,32,34,35 and none of them has been developed in the 21 st century and focused on both graft and patients’ survival. Additional calculations have been performed to calculate life years from transplant, 27,28 and scores have been developed for predicting coronary heart disease, 30 graft function at 1-year 33 or survival after discharge.…”

“…Additional calculations have been performed to calculate life years from transplant, 27,28 and scores have been developed for predicting coronary heart disease, 30 graft function at 1-year 33 or survival after discharge. 31 Only few previous scores have been developed based on data from 21 st century, 26,33,35,39–41 however, none of them focused on both graft and patients’ survival. Moreover, only a few efforts have been made to describe risk scores for use as prognostic tools to individualize risk of allograft loss or mortality in incident or prevalent transplant recipients.…”

“…56 Most importantly, our prediction model has yet to be externally validated in other cohorts. To the best of our knowledge, only few previously developed prediction scores were externally validated, 35,38,39,57 and only 2 these scores was validated in a different center. 39,57 Moreover, neither of these externally validated scores were developed and validated in patients in the United States at the 21 st century.…”

An adjustable predictive score of graft survival in kidney transplant patients and the levels of risk linked to de novo donor-specific anti-HLA antibodies

“…They are never updated, even if the patient’s prognosis is altered. The performance of these scores is usually evaluated with respect to shorter term graft survival and at a single time point [3–5,10,25]. In this study, we used the non-parametric RSF method which has several advantages compared to regression approaches among which it does not test the goodness of fit of data to a hypothesis, but seeks a model that explains the data [26].…”

“…Serum creatinine (Scr) and estimated glomerular filtration rate (GFR) are not sufficiently reliable predictors for long-term risk of graft loss or patient death [1]. In the last decade, predictive models of graft survival based on large panels of data collected in the donor [2], in the recipient before transplantation [3], and/or in the first year post-transplantation [4,5] have been proposed. A limitation of these models is that they do not take into account the onset of adverse events over time, which modify graft outcome.…”