Identification of driven risk factors for HLA-DR in kidney transplantation
Yunwei Zhang1,2, Jean Yang1,2, Samuel Mueller1, Germaine Wong3,4,5,6.
1School of Mathematics and Statistics, Faculty of Science, The University of Sydney, Sydney, Australia; 2Charles Perkins Centre, The University of Sydney, Sydney, Australia; 3Centre for Kidney Research, Sydney, Australia; 4Kids Research Institute, Sydney, Australia; 5The Children’s Hospital at Westmead, Sydney, Australia; 6Sydney School of Public Health, The University of Sydney, Sydney, Australia
Introduction: Optimising graft and patient survival are the key priorities for patients, caregivers and health professionals. Current deceased donor kidney allocation priorities HLA DR matching because there is considerable evidence to show avoiding DR mismatch will reduce the risk of allograft rejection and improve overall survival. However, the specific factors which may modify the impact of HLA-DR matching and long-term recipient outcomes are unknown.
Materials and Methods: Using data from the Australian and New Zealand dialysis and transplant registry (ANZDATA, 2006 – 2017, n = 3730), we developed a combined robust risk factor selection/ bootstrapping and Cox regression modelling approach to identify the factors that may modify the impact of HLA-DR matching on allograft outcomes including age, height, weight, waiting time on the waiting list, ethnic, lung function characteristics variables from donors and age, height, weight, creatinine level characteristics variables from recipients and transplant cold ischaemia time. We also compare our strategies with various unsupervised learning approaches including self-organising maps to partition the data into different subgroups. A second stage of machine learning approaches including but not limiting to random forest or logistic regression are built for detecting the characteristic of cohort where HLA-DR is an important risk factor.
Results and Discussion: A total number of 3692 participants were included in the modelling and 11 variables are identify including know key variables such as recipient’s age, recipient’s waiting time, donor's age, donor's weight and more. We then use a machine learning approaches such as random forest to identify variables / characteristic of cohort where HLA-DR is an important risk factor. Here, the explained variance of the out-of-bag error for the forest is 1.8%. The tree model (using those aforehead mentioned 11 variables) suggests that HLA-DR is particularly significant for long and short-term graft survival prediction for deceased donor-recipient pair with younger donor age, lower BMI and lower recipient BMI.
Conclusion: in general, HLA-DR is an essential immunological compatibility measure for long and short-term post-transplant graft survival prediction, which also indicates that it should be considered in the renal allocation algorithm as well. HLA-DR is particularly important for “fit” recipient-donor pairs makes sense clinically as both the donors and recipients are in good physical condition, and therefore the determining factor for post-transplant life years should be the matching factor. Future implementation will be the precise individualised graft and recipients’ survival prediction model building. This framework will dynamically adjust itself using different risk factors for different recipient.
There are no comments yet...