(219f) A Computational Predictive Framework Towards Individualized Risk Assessment of Kidney Transplantation Failure.

Kidney transplantation is the treatment needed for end-stage renal disease, whose success depends on several factors such as cancer, serious infections, severe cardiovascular diseases, and liver disease [1,2]. After receiving a kidney transplant, patients must take immune-suppressing medications to help their immune system to accept the new organ. More precisely, after kidney transplantation, patients might experience either B, or T, or both B and T cell rejection. Among them, T cell rejection is quite common; both B and T cells rejections are less often, while B cell rejections are rare and occur after a long time. Appropriate therapeutic intervention may successfully cure these complications but in many cases, only leads to partial or no improvement. Hence, despite all previous diagnostic and therapeutic progress, the mean kidney graft survival is limited to approximately 15 years [3,4].

Glomerular filtration is the process by which the kidneys filter the blood, removing excess wastes and fluids, and indicates kidney function. The physicians estimate the Glomerular filtration rate (GFR) based on the blood creatinine, age, body size and gender. The GFR levels help a doctor to plan treatment for the patient. The lower the GFR number is, the worse kidney function is. Thus, when the diagnosis becomes early, the kidney failure progress could be decreased or be stopped [5]. Chronic kidney disease is diagnosed when the GFR is below 60 for three months or more or GFR is above 60 with kidney damage (high levels of albumin patients' urine). There are five kidney disease stages. In stage 1, the GFR is 90 or higher, and the kidney function is normal. The GFR in stage 2 ranges from 60 to 89, and there is a mild loss of kidney function. Stage 3 is divided into Stage 3a which is indicated by GFR values ranging from 45 to 59 with mild to moderate kidney function, and Stage 3b where the GFR reaches 30, with moderate to severe kidney duction. In stage 4, the loss in kidney function is severe and its GFR values are within 15 to 29. Lastly, in stage 5, the GFR drops from 15, and kidney failure appears [6].

Management of patients with chronic kidney diseases with or without transplantation is a very challenging task. During chronic diseases, there is a dynamic disease process of injury and repair responses inherent to the specific disease present. These occurring responses may be simulated using a set of interacting and disease-modifying variables, whose values are based on the conditions, the events, and potential interventions. Validating the variable complex interplay is the key to personalized patient treatment in terms of diagnostic measures and therapeutic decisions. Of course, it is required the identification of the relevant causative and non-causative risk factors, their potential interaction, estimation of putative therapeutic effects concerning the clinical disease course and mid- and long-term outcomes [7].

Common mathematical approaches for predicting GFR are regression analysis and Cox proportional hazard analysis, which numerically estimate the outcome of GFR in a given study population at a specific point in time [8]. By nature, these approaches cannot truly capture the dynamic disease processes. Also, appropriate appreciation of potential interactions between different factors in the model is challenging. Thus, such mathematical models differ fundamentally from the holistic perspective of the clinician, who tries to integrate all information about the disease course and possible relationships between various factors up to the current time of patient assessment.

By using sparse medical data, this work proposes a novel framework that can predict the dynamics of the GFR with a mathematical model and apply uncertainty analysis to predict the risk of kidney failure up to two years. The mathematical model used was similar to a dumpling oscillator with a stability threshold and a dumpling factor. Two patient datasets were used. The training set with the validation cohort consists of 1700 patients. Briefly, the initial steps of our methodology were the feature selection in the patients' unmodeled data, the regression of GFR in a year, and the parameter estimation in non-rejection cases. The model analysis found the personalized stability threshold whose values were within the literature range. Based on the final personalized GFR range values, model parameter values were estimated, and uncertainty analysis was performed with the distribution of the dumpling factor. This procedure was applied for patients without a rejection, with one rejection, and with multiple rejections. By comparing the simulated values of the uncertainty analysis with the actual data, and in different personalized threshold ranges, the methodology achieved an accurate prediction of the success of transplantation, with an area under the curve (AUC) close to 0.80.

This work shows that the combination of machine learning, mathematical modeling, and big data analysis are undoubtedly valuable for predicting kidney transplantation outcomes. Subsequently, by capturing the personalized GFR dynamics, the medical community will proceed efficiently to the personalized treatment of the patients.

1. National Institute of Diabetes and Digestive and Kidney Diseases [U.S.]. Division of Kidney Urologic and Hematologic Diseases., Urban Institute. Renal Research Program., University of Michigan., USRDS Coordinating Center. United States Renal Data System. Annual Data Report. 2008;v.
2. Humar A, Matas AJ. Surgical complications after kidney transplantation. Semin Dial. 2005;18(6):505–10.
3. Luque S, Lúcia M, Melilli E, Lefaucheur C, Crespo M, Loupy A, et al. Value of monitoring circulating donor-reactive memory B cells to characterize antibody-mediated rejection after kidney transplantation. Am J Transplant. 2019;19(2):368–80.
4. Matz M, Fabritius K, Lorkowski C, Dürr M, Gaedeke J, Durek P, et al. Identification of T cell-mediated vascular rejection after kidney transplantation by the combined measurement of 5 specific MicroRNAs in blood. Transplantation. 2016;100(4):898–907.
5. Bakris GL, Williams M, Dworkin L, Elliott WJ, Epstein M, Toto R, et al. Preserving renal function in adults with hypertension and diabetes: a consensus approach. National Kidney Foundation Hypertension and Diabetes Executive Committees Working Group. Am J Kidney Dis [Internet]. 2000;36(3):646–61. Available from: http://www.ncbi.nlm.nih.gov/pubmed/10977801
6. Thomas C, Thomas L. Renal failure--measuring the glomerular filtration rate. Dtsch Arztebl Int (Internet). 2009;106(51–52):849–54. Available from: http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=2803612&tool=p...
7. Ng K, Kartoun U, Stavropoulos H, Zambrano JA, Tang PC. Personalized treatment options for chronic diseases using precision cohort analytics. Sci Rep. 2021;11(1).
8. Van Den Brand JAJG, Dijkstra TMH, Wetzels J, Stengel B, Metzger M, Blankestijn PJ, et al. Predicting kidney failure from longitudinal kidney function trajectory: A comparison of models. PLoS One. 2019;14(5).