The dynamic field of renal transplantation is characterised by limited access to data, the high cost of equipment and underlying complexity of immunological responses to transplants. Traditional clinical studies often fail to predict the postoperative outcome as they utilize standard statistical analysis requiring large number of participants. An alternative approach, combining traditional statistical tools with the development and application of machine learning algorithms, can open new perspectives for preoperative risk assessment and provide access to safe transplantation for many currently untranslatable patients. Our study demonstrates a novel data-driven approach based on classification decision trees (DTs) for prediction of acute antibody mediated rejection in the early post-transplant period after HLA-antibody incompatible transplantation. The available clinical dataset featured 15 potential predictor variables including pre-treatment DSA IgG subclass levels across 80 observation samples, out of which 46 belonged to rejector group and 34 represented non-rejector controls.

In order to compensate for high volatility in performance due to small number of training samples, 600 separate DTs were investigated. We demonstrated that the DT approach can successfully identify the optimal hierarchy of parameters associated with acute graft rejection. DT prediction was based on six out of the fifteen variables including three most important: the highest MFI DSA pre-treatment level, total IgG4 MFI pre-treatment level and number of HLA mismatches. The model has also provided specific levels of DSAs which associated with early graft rejection. The DT performance was evaluated in terms of classification accuracy, sensitivity and specificity, for the training and test datasets separately. The best performing model achieved 86.7% accuracy during the training phase and correctly classified 85% of test cases. The DT model provides an accurate prediction tool for antibody mediated rejection of renal transplants, whilst simultaneously estimating the highest risk factors and DSAs MFI levels associated with the increased risk of rejection in the early post-transplant period. The model can be a useful tool for identification of the risk factors for safe transplantation.

« Back to 2015 American Transplant Congress