Perspective
Risk, Prognosis, and Unintended Consequences in Kidney Allocation
Benjamin E. Hippen, M.D., J. Richard Thistlethwaite, Jr., M.D., Ph.D., and Lainie Friedman Ross, M.D., Ph.D.
March 16, 2011 (10.1056/NEJMp1102583)
ArticleReferences
The gap between the supply and demand of transplantable kidneys is growing, leaving policymakers eager to maximize the benefit of every kidney transplanted. Recently, a proposal for changing the way kidneys from deceased donors are allocated was proffered for public comment by the Kidney Committee of the Organ Procurement and Transplantation Network (OPTN).1 Instead of continuing to use a patient's waiting time as the core determinant of allocation priority, the new system employs a “risk quantification score” called the Kidney Donor Profile Index (KDPI) combined with a calculated Estimated Post-Transplant Survival (EPTS) score in an attempt to quantify risk factors for graft failure and recipient death, respectively. These scores are mathematical models based on a retrospective analysis of data collected by the Scientific Registry of Transplant Recipients on donor and recipient characteristics over the past several years. The models have never been prospectively validated, nor does the proposal outline an intention to validate them.
Briefly, the proposed system would preferentially allocate the top quintile of donor kidneys with the best expected graft survival to the top quintile of transplant candidates with the longest predicted post-transplantation survival (see table
Key Proposed Changes in Allocation of Kidneys from Deceased Donors to Adult Candidates.
). For the remaining donors and recipients, there would be broad age matching within a 30-year age range.1 The touted advantages of this proposal over the current system are additional years of survival gained from each kidney transplanted by avoiding the “premature” death of patients with functioning grafts (which occurs when a kidney that is predicted to be of good quality is transplanted into a candidate with a shorter predicted lifespan). However, the proposal makes clear that allocating top-quintile organs to top-quintile candidates will come at the expense of the overall opportunities for older candidates on the waiting list to be offered a kidney from a deceased donor.1
The intuitive appeal of a proposal to maximize the overall survival of patients after kidney transplantation is obvious. However, a significant change in allocation policy must be based on good data and sound methods of data analysis. As Ware has pointed out, the use of risk factors as prognostic tools for the purpose of prospective individual risk stratification often yields disappointing results.2 This is because much depends on how various risk factors in donors and recipients are distributed across sample populations that do or do not have graft loss. If a single risk factor — or an elaborate combination of donor and recipient factors such as those captured in the KDPI and EPTS score — is distributed substantially similarly in the population of patients whose grafts survive and in the population of those who experience graft loss, then a model that uses it to distinguish graft survival from early graft failure will have low sensitivity in making predictions regarding individual donor kidneys (in the case of the KDPI) or individual candidates (in the case of the EPTS score). Any attempt to increase sensitivity in the model will yield an unacceptably high frequency of false positives.3
A glance at the receiver-operating-characteristic (ROC) curves and concordance (C) statistics of the KDPI model suggests how it might fare as a prognostic tool. For a binary outcome (graft survival vs. graft loss), a C-statistic of 0.5 represents a prediction no more accurate than chance. Across all patient groups, the KDPI had a C-statistic of 0.62,4 but when the lowest- and highest-risk quintiles were compared, the C-statistic improved to 0.78. So the KDPI can discriminate relatively well between a donor kidney at the very high extreme of risk for graft failure and one at the very low extreme, but comparing low-risk donor kidneys to kidneys in the “muddy middle” will carry a higher probability of error.
Unless a variable (or combination of variables, such as donor age, recipient age, presence or absence of diabetes, and presence or absence of hypertension) strongly distinguishes all patients with graft survival from all patients with graft failure, a scoring system employing those variables may well identify risk factors for graft loss but nevertheless have limited usefulness as a prognostic tool.5 Using these risk factors for individual or small-subgroup risk stratification (e.g., drawing a bright line between the “top 20%” of donors and recipients and the remaining 80%) will often generate mistakes, either by failing to identify many patients who will experience longer graft survival (low sensitivity) or by falsely identifying patients who will develop early graft failure as likely to have longer graft survival (false positive). The cost of these errors is that the intended benefits have a high chance of being offset by reduced opportunities for transplantation in patients who are incorrectly judged to be at high risk for graft loss or by incorrect allocation of kidneys to candidates who will not benefit for as long as predicted. Even if the problem of incorrectly identifying individual donor kidneys and candidates who will do well is “smoothed out” by enough iterations of correct assessments of kidneys and candidates, the model relies on past observations to correctly predict future trends and assumes that incorrect predictions will be randomly distributed over all groups of kidneys and recipients. Confidence in the ability to make such predictions reproducibly and reliably, which neither the KDPI nor the EPTS score provides, is critical for any substantive ethical discussion about the proposal for preferential allocation.
Any proposal to revise the allocation of organs from deceased donors also risks causing unintended consequences for patterns and trends in donation by living donors. Past revisions to the allocation system warrant careful attention, since about 40% of donated kidneys come from living donors. Living kidney donations, which are not (and cannot be) accounted for in the proposal, confer a greater survival benefit than even optimal deceased-donor kidneys. A new rule (“Share 35”) was adopted in 2003 and implemented in 2005 to prioritize the allocation of organs from deceased donors younger than 35 to pediatric candidates, because of the adverse effect of end-stage renal disease on growth and cognitive development. The policy did result in more rapid transplantation of kidneys into children. However, the increased availability of deceased-donor kidneys was accompanied by a decrease in the number of living-donor kidneys donated to pediatric candidates (see graph
Trends in Donation of Kidneys from Living and Deceased Donors to Pediatric Candidates.
). This observation raises the obvious concern that if the top quintile of adult candidates for kidney transplants is prioritized for the top quintile of organs from deceased donors, the rates of living donation to these candidates will fall in similar fashion. Data from the OPTN Web site show that in 2010, recipients 18 to 34 years of age and recipients 35 to 49 years of age received 53% and 41% of their organs from living donors, respectively. In contrast, recipients 50 to 64 years of age and recipients 65 or older received only 33% and 28% of their organs from living donors, respectively. A drop in the rates of living donation for younger candidates may result in the gain of fewer life-years than the current system, either because living donation would shift from younger to older recipients or because the reduced rate of living donation would disproportionately occur among healthier recipients, which would probably exacerbate the recent drop-off in total rates of living donation.
To ensure a fair allocation system, the goal of maximizing benefit must be balanced by concerns about equity (fair opportunity for everyone with end-stage renal disease). In the absence of reliable and reproducible prognostic tools for estimating graft survival, and without a clear understanding of the unintended consequences of a substantial change in allocation policy on trends in living donation, discussing what a fair allocation policy should look like is putting the cart before the horse.
The views expressed in this article are those of the authors and do not represent those of the United Network for Organ Sharing (UNOS) or the Organ Procurement and Transplantation Network.
Dr. Ross is a member of the ethics committee of UNOS, Richmond, VA.
Disclosure forms provided by the authors are available with the full text of this article at NEJM.org.
This article (10.1056/NEJMp1102583) was published on March 16, 2011, at NEJM.org.
SOURCE INFORMATION
From the Carolinas Medical Center, Charlotte, NC (B.E.H.); the Department of Surgery, Section of Transplantation (J.R.T., L.F.R.) and the Departments of Pediatrics and Medicine and the MacLean Center for Clinical Medical Ethics (L.F.R.), University of Chicago, Chicago.
http://www.nejm.org/doi/full/10.1056/NEJMp1102583?query=TOC