Identifying scenarios of benefit or harm from kidney transplantation during the COVID-19 pandemic: A stochastic simulation and machine learning study

Abstract

Clinical decision-making in kidney transplant (KT) during the coronavirus disease 2019 (COVID-19) pandemic is understandably a conundrum: both candidates and recipients may face increased acquisition risks and case fatality rates (CFRs). Given our poor understanding of these risks, many centers have paused or reduced KT activity, yet data to inform such decisions are lacking. To quantify the benefit/harm of KT in this context, we conducted a simulation study of immediate-KT vs delay-until-after-pandemic for different patient phenotypes under a variety of potential COVID-19 scenarios. A calculator was implemented (http://www.transplantmodels.com/covid_sim), and machine learning approaches were used to evaluate the important aspects of our modeling. Characteristics of the pandemic (acquisition risk, CFR) and length of delay (length of pandemic, waitlist priority when modeling deceased donor KT) had greatest influence on benefit/harm. In most scenarios of COVID-19 dynamics and patient characteristics, immediate KT provided survival benefit; KT only began showing evidence of harm in scenarios where CFRs were substantially higher for KT recipients (eg, ≥50% fatality) than for waitlist registrants. Our simulations suggest that KT could be beneficial in many centers if local resources allow, and our calculator can help identify patients who would benefit most. Furthermore, as the pandemic evolves, our calculator can update these predictions.

Keywords: Scientific Registry for Transplant Recipients (SRTR); clinical research/practice; infection and infectious agents; kidney transplantation/nephrology.

FIGURE 1

State diagram of the Markov model. Patients who select “delay” and survive through n months after the delay transition from a “Waitlist” state to a "KT" state. Other state transition probabilities are determined by Markov model parameters and/or Poisson models of waitlist and post-KT mortality. CoV, coronavirus 2; KT, kidney transplant [Color figure can be viewed at wileyonlinelibrary.com]

FIGURE 2

Overall predicted waitlist and post-DDKT/post-LDKT mortality in absence of COVID-19, based on Poisson regressions used for input to the Markov model. COVID-19, corona virus disease 2019; DDKT, deceased donor kidney transplant; LDKT, living donor kidney transplant [Color figure can be viewed at wileyonlinelibrary.com]

FIGURE 3

Examples of simulated survival curves for immediate DDKT vs delay. A, White male patient, private insurance, age 20-29, 1-2 y on dialysis, no DM, no prior KT. Community COVID-19 acquisition risk is “medium-lingering,” acquisition risk is 50% higher for dialysis patients, and probability of nosocomial acquisition at KT is 10%; “medium” waitlist and post-KT CFR; 6-mo delay. A, “Medium” waitlist and post-KT CFR, 6-mo delay. B, Like (A) except 24-mo delay. C, Like (A) except “high” waitlist CFR. D, Like (A) except “very high” post-KT CFR. E, Like (A) except community COVID-19 acquisition risk is “high-lingering.” F, Like (A) except age 70+. CFR, case fatality rate; COVID-19, corona virus disease 2019; DDKT, deceased donor kidney transplant; DM, diabetes mellitus; KT, kidney transplant [Color figure can be viewed at wileyonlinelibrary.com]

FIGURE 4

Distribution of life-months gained over 5 years from immediate vs delayed KT: effect of varying a single parameter. All distributions are for a white male patient with private insurance and no prior KT. CoV, coronavirus 2; COVID-19, coronavirus disease 2019; KT, kidney transplant [Color figure can be viewed at wileyonlinelibrary.com]

FIGURE 5

CART summarizing estimated LMG5 from immediate transplant vs delay, based on epidemic characteristics and length of delay. Details of CFRs and COVID-19 risk parameters appear in Table 1. CART, classification and regression tree; CFRs, case fatality rates; COVID-19, coronavirus disease 2019; KT, kidney transplant; LMG5, life-months gained over the first 5 years