Post-Transplant Diabetes Outcomes Prediction Through Machine Learning and Deep Phenotyping

Chronic Kidney disease (CKD) is a major global health burden and represents one of the most common non-communicable diseases. In Europe, CKD affects over 50 million people, representing approximately 10% of the adult population. Importantly, the presence of CKD is a significant economic burden on healthcare systems, with an estimated cost of 140 billion annually in Europe. Kidney transplantation represents the best treatment of end stage renal disease (ESRD) in terms of mortality, morbidity and quality of life. In addition, this therapeutic approach to ESRD considerably reduces the cost of renal replacement therapy. Post-transplant diabetes is a common metabolic complication of kidney transplantation. Up to 40% of kidney graft recipients present within the first 5 years after transplantation a diagnosis of de novo diabetes and another 30% are characterized by an impaired glucose tolerance (IGT). In addition, 20% of patients with IGT will eventually develop a post- transplant diabetes.

Immunosuppressive therapy represents the main culprit with its deleterious effects on either insulin resistance (corticosteroids, mTOR inhibitors) or insulin synthesis (tacrolimus). Behind the role of immunosuppressive therapy, other relevant risk factors are recipients' age and pre- and post-transplant BMI. A great amount of registry data and a recent meta- analysis on retrospective studies clearly indicate the detrimental effect of post-transplant diabetes on the main clinical outcome of kidney transplantation, recipients' mortality and graft loss. The excess mortality observed in this setting is mainly due, as expected, to an increase in cardiovascular death. Although the link between diabetes and cardiovascular mortality is well known and its mechanisms are mostly clear in the general population, we have a significant lack of information in this specific setting, where post- transplant diabetes act on the top of several other cardiovascular risk factors, often present in the transplant population. Thus, our ability to stratify the risk and to intervene accordingly, to prevent cardiovascular events in kidney graft recipients with post-transplant diabetes is significantly limited. This lack of knowledge will inevitably lead to an overtreatment of patients potentially at lower risk and to an under-treatment of graft recipients potentially at very high risk. On the other hand, when we consider the issue of graft loss, we inevitably focus our attention on the immunological mechanisms linked to the alloimmune response of the recipients against the graft and, subsequently on the modulation of immunosuppression. However, in the last few years we are realizing that the risk factor for ESRD that are well known in the general population have a significant prognostic weight also in the prediction of graft loss in kidney transplantation. We are well aware that among these risk factors the diabetes is still one of the most important. Although, also in this case we lack information on how the diabetic milieu interacts with transplant-specific ESRD risk factors to determine the fate of the graft.

In addition, for ESRD, our inability to stratify each patient risk will significantly limit our therapeutic intervention. Thus, the aim of the present project is to fill this gap of knowledge with an approach based on deep phenotyping of patients with post-transplant diabetes associated with a system biology strategy. This methodology will allow us to identify potential molecular markers at the urine, serum or renal tissue levels that will associate with clinical, imaging or histological features known to predict either cardiovascular mortality or ESRD. With the help of the artificial intelligence, we will then build the prototype of a predictive model for both cardiovascular mortality and graft loss to be then validated in a dedicated prospective study.