MACHINE LEARNING PREDICTION FOR DE NOVO DONOR SPECIFIC ANTIBODIES (dnDSA), AMR AND GRAFT LOSS IN SIMULTANEOUS KIDNEY PANCREAS (SPK) TRANSPLANT RECIPIENTS

A SHARMA1,2, C COOREY1,2, A TAVERNITI1, B NANKIVELL3, J CHAPMAN3, J CRAIG1,2, P O’CONNELL3, H PLEASS3, W LIM4, J YANG5,6, G WONG1,2,3
1Centre for Kidney Research, Westmead, Australia, 2School of Public Health, University of Sydney, Sydney, Australia, 3Centre for Transplant and Renal Research, Westmead, Australia, 4Department of Renal Medicine, Perth, Australia, 5School of Mathematics and Statistics, University of Sydney, Sydney, Australia, 6Charles Perkins Centre, Sydney, Australia

Aim: To develop a prediction model for dnDSA and allograft loss based on the number and location of eplet mismatches in SPK transplant recipients.
Methods: 170 SPK transplant recipients (2005-2017) were assessed using data from ANZDATA registry and NOMS. Adjusted logistic regression were conducted to determine associations between class specific eplet HLA mismatches, dnDSA production, AMR and graft loss. Machine learning models (random forests) were used to predict dnDSA, AMR and allograft loss based on location of eplet mismatches.
Results: The cohort included 92 (54%) males with mean age 38.5 years (SD 6.9) and median follow up time 6.6 years (IQR: 3.9-11.0). The most common class I and II eplet mismatches were at 156RA (35%) and 70D (55%), respectively, with 33 (19%) and 52 (31%) recipients developing Class I and II dnDSA. An increased risk of dnDSA production was observed per 10 increase in Class I (adjusted odds ratio (aOR) 2.0, 95%CI 1.2-3.3, P=0.01), and Class II eplet mismatches (aOR 1.3, 95%CI 1.1-1.5, P=0.01). The presence of Class II dnDSA was associated with increased risk of AMR (aOR 3.9, 95%CI 1.6-9.5, P<0.01). The top eplet mismatches associated with class II dnDA production were located at 45EV and 46VY, corresponding to HLA DQB1*03:01 and DQB1*02:01. Random forest model with the location of the top 10 class specific eplet mismatches achieved a mean cross-validation accuracy of 63.9% and 61.2% for class I and II dnDSA production.
Conclusions: The number of eplet mismatches was associated with dnDSA production at class I and II, and the location of eplet mismatches at 45EV and 46VY on HLA DQB1*03:01 and DQB1*02:01 may predict Class II dnDSA after SPK transplant.


Biography:
Nephrologist and PhD candidate (University of Sydney)

Categories