Verena Gotta (1,2), Andrew Atkinson(1,3), Olivera Marsenic(4), and Marc Pfister (1)
(1) University Children’s Hospital Basel, University of Basel, Basel, Switzerland, (2) University Hospital Basel, Basel, Switzerland (3) University Hospital Bern, Bern, Switzerland (4) Stanford University School of Medicine, Lucile Packard Children's Hospital, Stanford, CA
Objectives: Mortality remains higher in pediatric and adult patients on chronic hemodialysis (HD) than after kidney transplantation[1]. As the etiology of underlying kidney disease frequently differs in pediatric patients compared to adults, disease-related risk factors of mortality are not well described in the small population of patients that started HD in childhood. We aimed to develop a time-to-event model characterizing the relationship of disease-related factors with survival on chronic HD, based on key factors associated with mortality previously identified using a machine-learning (ML) analysis[2].
Methods: Data used for this retrospective cohort analysis originate from patients <30 years on chronic HD treatment since childhood (≤18 years), having received thrice-weekly HD between 2004 and 2016 in outpatient DaVita dialysis centres[3,4]. Parametric survival analysis was performed using a Weibull model. The developed unadjusted baseline Weibull model (null model) was used to investigate disease-related laboratory variables as potential predictors of survival. Considered variables (n=20) were identified in a preceding exploratory ML analysis (random forest investigating 5-year mortality, n=11 disease-related predictors)[4], and were challenged by further variables reported in literature (n=9 additional potential predictors). Uni- and multivariable regression analysis was performed with forward selection and backward deletion and P<0.05 as inclusion/exclusion criteria, considering both linear and non-linear relationships for continuous variables.
Results: A total 1523 patients with complete information on prespecified evaluated variables were included in the analysis. The baseline Weibull model predicted 5-, 10- and 15-year survival of 94%, 85% and 75%, respectively, which matched non-parametric Kaplan-Meier estimates well. Estimated scale (σ) and shape (α) parameters were σ=37.7 years (time when predicted survival = 40%) and α=1.37 (indicating with α>1 increasing hazard of death over time). Eighteen out of twenty disease-related variables showed a significant (P<0.05) association with survival in univariable analyses, among 10 out of 11 disease-related variables identified in previous ML analysis. Six variables (relationship with log-hazard) were retained in the final multivariable model, all were part of the previous ML model: red blood cell distribution width (cubic), lymphocytes (linear), lactate-dehydrogenase (piece-wise linear), albumin (linear), creatinine (log-linear), and post-dialysis blood pressure (linear) (all P≤0.01).
Conclusions: The developed disease model suggests that factors related to nutrition, inflammation/metabolism, anemia and cardiovascular risk are associated with survival on chronic HD in patients starting dialysis in childhood. Evaluation of drug- and HD treatment effects on those parameters, combined with disease-model simulations, could help to evaluate and optimize treatment in this patient population, in whom randomized investigations are limited by small patient numbers. As mortality on HD increases over time, the ultimate goal of early kidney transplantation remains undisputed.
References:
[1] Chesnay et al. Pediatr Nephrol 2018
[2] Gotta V and Tancev G, et al. ASN Kidney week 2019. Abstract: TH-PO789
[3] Gotta V, et al. Nephrol Dial Transplant. 2018 Apr 18
[4] Gotta V, et al. ASAIO J. 2020 Jan;66(1):115-123
Reference: PAGE () Abstr 9538 [www.page-meeting.org/?abstract=9538]
Poster: Oral: Drug/Disease Modelling