Abstract: FR-PO886
The Association Between Location of Eplet Mismatches, De Novo Donor Specific Antibodies, and Acute Rejection in Simultaneous Pancreas-Kidney Transplant Recipients Using Novel Machine Learning Methods
Session Information
- Transplantation: Translational and Transplant Pathology
October 26, 2018 | Location: Exhibit Hall, San Diego Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Transplantation
- 1802 Transplantation: Clinical
Authors
- Sharma, Ankit, Centre for Kidney Research, Westmead, New South Wales, Australia
- Coorey, Craig, Centre for Kidney Research, Westmead, New South Wales, Australia
- Taverniti, Anne Therese, Centre for Kidney Research, Westmead, New South Wales, Australia
- Nankivell, Brian John, Westmead Hospital, WESTMEAD, New South Wales, Australia
- Chapman, Jeremy R., Westmead Hospital, WESTMEAD, New South Wales, Australia
- Craig, Jonathan C., Centre for Kidney Research, Westmead, New South Wales, Australia
- Lim, Wai Hon, Sir Charles Gairdner Hospital, Nedlands, Western Australia, Australia
- Yang, Jean, University of Sydney, Sydney, New South Wales, Australia
- Wong, Germaine, Centre for Kidney Research, Westmead, New South Wales, Australia
Background
To determine the association between HLA class and location of eplet mismatches with de novo donor specific antibodies (dnDSA) and acute rejection in simultaneous pancreas-kidney (SPK) transplant recipients.
Methods
The cohort consisted of SPK recipients (n=170) transplanted in New South Wales, Australia between 2005 and 2017. Using machine learning models (random forest), we determined the association between the number and location of class specific eplet mismatches, and adverse allograft outcomes (dnDSA formation and acute rejection (acute cellular (ACR) and (antibody-mediated (AMR)). One hundred times 5-fold cross validation was conducted using R.
Results
The cohort included 93 (55%) males, with mean age at transplant of 38.7 years (SD 7.0) and median follow up time of 5.0 years (IQR: 2.1, 7.1). The median total number of class I and II eplet mismatches were 17 (IQR: 13-22) and 34 (IQR: 22-47), respectively. The most important (highest Mean Decrease Accuracy) class I eplet mismatches for predicting dnDSA and any acute rejection corresponded to locations 102HV, 149TAH, 152RE, 163RW, 21H, 44RM, 44RMA and 82LR; and for class II mismatches at 76L. For the prediction of class I and II dnDSA, the median AUC and balanced error rate from the location of class I eplet mismatches were 57.8% (SD: 3.5%) and 47.0% (SD: 1.6%); and for class II were 53.5% (SD: 2.6%) and 50.9% (SD: 2.5%), respectively. The median AUC and balanced error rate for predicting any acute rejection from class I eplet locations were 54.3% (SD: 2.7%) and 50.2% (SD: 2.7%); and for class II were 56.7% (SD: 3.3%) and 43.9% (SD: 2.7%), respectively.
Conclusion
In this cohort of SPK recipients, the location of class I mismatches best classify recipients with dnDSA and any acute rejection.