SEARCH

SEARCH BY CITATION

Keywords:

  • Deceased donor kidney transplantation;
  • LYFT;
  • organ allocation;
  • survival benefit

Abstract

  1. Top of page
  2. Abstract
  3. The Current State of Donor and Recipient Matching
  4. Utility-Based Allocation
  5. Options for Utility-Based Allocation
  6. A Proposed Framework for Evaluating Utility-Based Options
  7. Summary
  8. References

Over the last 5 years, a number of utility-based allocation systems have been proposed in an effort to increase the life-prolonging potential of deceased donor kidneys in the United States. These have included various adaptations of age-matching and net benefit, including the Eurotransplant Senior Program, Life Years From Transplant, and several systems for avoiding extreme donor/recipient mismatch. However, utility-based allocation is complex and raises issues regarding choice of metric, appropriateness of certain factors for use in allocation, accuracy of prediction models, transparency and perception, and possible effects on donation rates. Changing the role of utility in kidney allocation will likely cause changes to efficiency, equity, predictability, autonomy, controversy, trust and live donation. In this manuscript, various allocation systems are discussed, and a framework is proposed for quantifying the goals of the transplant community and evaluating options for utility-based kidney allocation in this context.

In 2004, the UNOS/OPTN Kidney Transplantation Committee was charged with reviewing the current kidney allocation system and making recommendations for improvement, so as to provide ‘equitable access for kidney transplant candidates to deceased donor kidneys for transplantation while improving the outcomes of recipients of such kidneys’ (1). From this complex, iterative, ongoing process have emerged the Kidney Allocation and Review Subcommittee (KARS), the concept of Life Years From Transplant (LYFT) as a measure of transplant utility (2), Donor Profile Index (DPI) as a measure of donor quality, Dialysis Time (DT) as a measure of equity and the Kidney Allocation Score (KAS) as an attempt to balance equity and utility. More debates than acronyms have ensued, both ethical and statistical, as the transplant community has struggled to decide how best to allocate a scarce resource which differs from other organs by the availability of alternative modalities for renal replacement.

Although much can be, has been and should be said about the roles of DPI and DT, this discussion will focus on the potential role of various utility-based options for kidney allocation, including LYFT, age matching and potential alternative systems that have been proposed.

The Current State of Donor and Recipient Matching

  1. Top of page
  2. Abstract
  3. The Current State of Donor and Recipient Matching
  4. Utility-Based Allocation
  5. Options for Utility-Based Allocation
  6. A Proposed Framework for Evaluating Utility-Based Options
  7. Summary
  8. References

The most recent UNOS document describing the allocation goals encourages the introduction of a better utility measure because ‘the current system does not match donors and recipients well’. The intuitive argument frequently cited is that a 75-year-old recipient probably should not receive a kidney from a 20-year-old so long as there are 20-year-olds on the waiting list who stand to better utilize the many years of potential life from that allograft. As an extreme example, this is understandable and convincing; however, as the example becomes less extreme, the answer becomes less clear.

To explore the status quo and set the stage for this discussion, it is worthwhile to explore data obtained from the OPTN website as well as a brief descriptive analysis of approximately 20 000 deceased donor transplants to adult recipients between January 1, 2006, and December 31, 2007, based on data from the UNOS/OPTN Standard Transplant Analysis and Research (STAR) files. During this time period, approximately 4100 kidney transplants were performed per year from donors aged 35 or younger (41% of all adult transplants), approximately 4100 new candidates aged 18–35 registered per year for the waiting list (13% of new adult candidates), approximately 2600 patients aged 18–35 received deceased donor kidneys and approximately 1350 patients aged 18–35 received live donor kidneys per year. The dichotomization at age 35 is chosen purely for simplicity, and similar descriptions could be performed using other forms.

So for those adults at age 35 or younger, it seems then that the overwhelming majority receive either a live donor or deceased donor kidney from a young donor. Of recipients aged 18–35 who received a deceased donor kidney, the mean donor age was 31, and 96% received an SCD kidney; 57% received a kidney from a donor aged 18–35, 77% received one from a donor aged 18–45 and only 1% received a kidney from a donor over 60 (Figure 1). Of donors of age 35 or younger, the mean recipient age was 49; 76% went to recipients under 60, and only 3.5% went to recipients over 70. So, despite the fear of the extreme example above, only 80 out of 10 000 transplants per year fell into the category of donor age ≤ 20 allocated to recipient age ≥ 70, demonstrating that, fortunately, this extreme scenario is cited as an example more often than it ever actually occurs.

image

Figure 1. Current state of donor and recipient age matching. Based on deceased donor allocation between 2006 and 2007, this figure shows the breakdown of donor age, by recipient age stratum.

Download figure to PowerPoint

As for general donor and recipient matching, the median (absolute value) difference between donor and recipient age was 14 years; 25% of transplants involved a donor and recipient who were less than 6 years apart, and 75% of transplants involved a matching less than 26 years apart. Although these differences in donor and recipient age were not egregious, some might argue that they could have been improved. Adding a utility measure to an organ allocation system could improve this matching, although it is possible that a similar improvement could result from addressing geographic disparities. In other words, it is possible that an allocation strategy that seeks to improve justice (by sharing organs from regions with many young donors to regions with far fewer) might also improve utility without the need to deliberately do so (with the inherent risks and controversies of utility-based allocation). For example, 26% of transplant centers drew from a deceased donor pool where half were under 35, while 5% drew from a deceased donor pool where half were over 50 years old. Similarly, 10% of transplant centers perform more than half of their transplants between donors and recipients who are less than 8 years apart, while a different 10% of transplant centers perform more than half of their transplants between donors and recipients who are more than 20 years apart.

Utility-Based Allocation

  1. Top of page
  2. Abstract
  3. The Current State of Donor and Recipient Matching
  4. Utility-Based Allocation
  5. Options for Utility-Based Allocation
  6. A Proposed Framework for Evaluating Utility-Based Options
  7. Summary
  8. References

Given that the kidney committee was only charged with evaluating the current allocation system, and not necessarily changing it, one option is to keep the status quo and focus on justice-based allocation. The current system is transparent: it is basically one long first-in first-out queue. As Americans, we stand in line at the grocery store, at the movies and even at the doctor's office; the queue is intuitive for us and, to many, feels fair. Additionally, living donation is encouraged for all, as long waiting times inevitably remind patients and their families of the advantages of live donation (3). The disadvantages are obvious and are stated above in the exploration of the status quo. The potential of certain organs might be wasted, and young people, fearing long waiting times, are more likely to accept organs with shorter half-lives and eventually require re-transplantation.

Adding a utility measure to allocation offers the intuitive possibility of decreasing wasted potential graft life, allowing organs which are expected to function for many years to be transplanted into patients who are expected to live for many years. On the other end of the spectrum, patients with few expected future years of life might be appropriate candidates for kidneys with fewer years of expected graft life. Put another way, the ideal situation might be that every allograft is transplanted into a recipient that the allograft would outlive, but only by a few days.

However, utility-based allocation raises a number of issues:

  • 1
    Metric: measures of ‘utility’ are for the most part relegated to measures of graft or patient survival, since quality of life is extremely difficult to objectively measure and has tremendous between-person variability. However, many patients pursue kidney transplantation for the drastic improvement in quality of life, rather than to necessarily extend survival.
  • 2
    Appropriateness: some factors which determine utility might not be appropriate for allocation. For example, African Americans are at higher risk of graft loss than Caucasians (4,5). Similarly, private insurance is associated with better outcomes (6). But few would support a system that favors transplantation into privately insured Caucasians so as to maximize graft life (7,8).
  • 3
    Accuracy: although good prediction models would clearly improve clinical decision-making, data availability and reliability in current national registries significantly limit the ability to accurately predict posttransplant outcomes (9), making it difficult to decide when prediction accuracy is good enough to drive allocation policy (10).
  • 4
    Transparency: utility is not simple to predict, and better predictive ability is inevitably associated with more complex methods. Complex methods reduce transparency, and patients may not feel comfortable letting a ‘black box’ decide their fate.
  • 5
    Donation: perceptions of the organ allocation system are likely to affect deceased donation rates, and a more complex allocation system might lead to further mistrust in the system and refusal to allow donation (11). Additionally, shorter waiting times have been associated with lower live donation rates (3,12), so a shift in organ recipients might cause a shift in live donation. It remains unclear whether to consider effects on donation rates and whether to consider live donation at all when attempting to maximize utility.

Options for Utility-Based Allocation

  1. Top of page
  2. Abstract
  3. The Current State of Donor and Recipient Matching
  4. Utility-Based Allocation
  5. Options for Utility-Based Allocation
  6. A Proposed Framework for Evaluating Utility-Based Options
  7. Summary
  8. References

Two major surrogates for ‘utility’ have been explored: age and regression models predicting survival benefit. Each surrogate can be used as a continuous measure, trying to match each individual organ with the recipient best suited for that organ, or a categorical (usually dichotomous) measure, trying to keep organs that are more suitable for one subgroup within that particular subgroup. A number of options are evaluated below, noting that many more options have been considered and discussed by the KARS subcommittee, the UNOS Kidney Committee and the transplant community at large. The potential early outcomes of many of these allocation options have been studied through KP-SAM (Kidney–Pancreas Simulated Allocation Model) probabilistic simulations by the SRTR. The KP-SAM model is described online (13) and the results of the immense simulation efforts by SRTR are detailed on their website (14).

Age matching

Various options exist for matching the age of donor with the age of recipient. The advantages of age matching combine the possibility of reducing wasted potential of allograft life with a simple, transparent heuristic. The most glaring disadvantage is that chronologic age is a poor measure of physiologic age. The largest scale example of age matching is ESP, the Eurotransplant Senior Program, which matches donors over the age of 65 with recipients over the age of 65 (15–17). Initial outcomes from ESP support the concept that the adverse effects of an older donor kidney might be attenuated when placing it into an older recipient compared with placing it into a younger recipient; this effect modification is likely driven by the increased competing risks of death in older adults. Possibly most important is the lesson from ESP that an efficient system for allocating marginal organs can significantly reduce cold ischemia times to a mean of 11.9 h for transplants that are likely quite sensitive to this (15), as opposed to the US system where the mean cold times for ECD kidneys are over 19 h and where recent changes to allocation logistics have actually worsened the situation (18). However, most objections to age matching rightfully argue that some 60-year-olds are healthier and more able to benefit from a kidney transplant than some 40-year-olds, and that age alone is not a perfect surrogate for utility.

Life years from transplantation

Regression modeling efforts of the SRTR, reported on their website (19) but not yet published, have resulted in a score aimed at estimating the difference between the number of life years predicted for a given patient after a kidney transplant versus the number of life years predicted for the same patient were dialysis the treatment provided. As such, LYFT hopes to improve on the accuracy of age as a surrogate for utility by attempting to account for other comorbidities, such as diabetes, that might explain why some 60-year-olds are healthier than some 40-year-olds.

Similar modeling efforts occurred prior to the implementation of the Model for End-Stage Liver Disease (MELD) for liver allograft allocation (20,21). However, an important difference is that MELD attempts to predict events within a 90-day window, while LYFT hopes to predict events throughout life. This limitation is seen in a lower concordance (c)-statistic for LYFT (0.60–0.68) when compared with MELD (0.78–0.87); in the setting of clinical prediction, a c-statistic of 0.6–0.7 has been classified as having limited value, while a c-statistic of 0.8 and above is considered adequate for general clinical utility (22,23). In addition, while LYFT compromises transparency with the requirement for many variables and the integration of three complex survival models, MELD uses only three variables and a simple equation.

Several of the general issues in utility-based allocation listed above apply particularly to LYFT, including appropriateness (should race be included in LYFT), accuracy (is LYFT good enough for allocation) and transparency (are patients and providers comfortable using a complex population-based regression model to set allocation priorities for individuals). Numerous allocation models based on LYFT have been proposed and simulated, ranging from pure LYFT-based allocation (where the increase in overall life years gained is highest but resulting disparities are of concern) to KAS-based allocation which integrates LYFT, DPI and DT (where the increase in overall life years gained is much lower but disparities are better addressed). Proponents of KAS have argued that the addition of DT makes the inaccuracy of LYFT less relevant, and that DT would function as the primary factor to discriminate between patients with similar LYFT scores, but this argument also rests on the yet unproven and unlikely assumption that patients with similar potential benefit will have similar LYFT scores.

Avoiding extremes

If chronologic age and even more complex regression models fail to accurately differentiate predicted outcomes between patients with similar comorbidities as they are currently captured, one consideration might be to implement a system that avoids the unwanted extremes but otherwise maintains the current queue. Examples of such proposals have included (a) prioritizing all patients under 35 (instead of only pediatric patients) when donors under 35 become available; (b) allocating the youngest 20% of donors to the youngest 20% of candidates on the waiting list; (c) allocating the organs with the best 20% DPI to candidates with the best 20% LYFT. Although these have yet to be simulated, it is likely that the ‘utility’ gained by these will be less than that gained by a pure LYFT-based system; however, the utility might be similar to that gained by the modified LYFT-based system using KAS. The advantages of such heuristics include the avoidance of clear instances of wasted potential of allograft life without loss of transparency. One controversial disadvantage is the introduction of perceived cut-point disparities; for example, those entering the system 1 day after the age-35 cut-point would feel disadvantaged by the thought that yesterday they were a priority but today they no longer are. However, a current cut-point disparity already exists at age 18. Furthermore, with heuristics such as youngest 20% or 20% with best LYFT, to be selected among current candidates, the cut-point would vary over time (as does the ‘top MELD’ candidate on the liver list) and thereby reduce the perceived disparities.

A Proposed Framework for Evaluating Utility-Based Options

  1. Top of page
  2. Abstract
  3. The Current State of Donor and Recipient Matching
  4. Utility-Based Allocation
  5. Options for Utility-Based Allocation
  6. A Proposed Framework for Evaluating Utility-Based Options
  7. Summary
  8. References

It is quite likely that a change to the current kidney allocation system is warranted, and that some measure of utility is a reasonable addition to the modified queue which currently represents kidney allocation. However, the extent of this change is not clear, and although the debate depends significantly on the statistics and the resulting added benefit (the resulting utility), it should extend well beyond them. The choice, as in clinical medicine, ultimately requires a balance between risk and benefit, and the concepts of risk and benefit need to be explored in the context of the goals for kidney allocation.

A potential framework for the measurable risks and benefits of a utility-based allocation system is illustrated in Figure 2. Implementation of this framework would involve two important steps. First, the number of points available for each aspect of kidney allocation (the ‘value’ of each aspect) are assigned by the transplant community as a quantified expression of the goals of allocation. For example, if the maximum possible utility could be achieved, how many points would that be worth? If it is worth 100 points, then how many points are available to deduct for complexity causing mistrust, in the most extreme case? Similarly, how many points are available to add or deduct for changes in efficiency or equity? Through these point assignments (the left hand column of Figure 2), a quantified statement about priorities of kidney allocation can be made. For example, if the effect on live donation is a nonissue, it gets assigned 0 available points; if Utility is assigned 100 points and Controversy is assigned 10 points, then an increase in utility equivalent to 1/10 the maximum attainable utility outweighs even the maximum possible controversy that could arise.

image

Figure 2. A proposed framework for evaluating the balance between risk and benefit in the context of a utility-based allocation system. Priority is first assigned to each aspect of the allocation system (utility, efficiency, etc.) by setting the maximum number of points (negative or positive) available for that aspect. Then, for each allocation system of interest, points for risk or benefit are assigned (up to the total points available), and total points indicate the benefit-to-risk balance for a particular allocation system.

Download figure to PowerPoint

Once the goals are decided and points are introduced, then every potential allocation system under evaluation can be explored through this framework, using the current system as a baseline. In this way, the balance between predicted impact and ‘side effects’ such as complexity and uncertainty can be evaluated within predefined objective boundaries. A hypothetical example follows, where Utility is assigned a maximum of 100 points and each remaining aspect is assigned a maximum of 10 points. Since Utility is assigned 100 points, then the perfect utility measure (for example, if LYFT were a perfect predictor) would receive 100 points, and a simpler measure that derives less utility (for example, a simple age-matching heuristic) might be assigned only 90 points. But if the perfect utility measure engenders mistrust because of its complexity, reduces equity and causes controversy because of the inclusion of age and race, for example, the 100 points of utility might be attenuated by −10 points for Trust, −10 points for Equity and −10 points for Controversy, while the simpler system might lose no points for Trust and only 5 points for each of Equity and Controversy. In this case, the simpler system, although achieving lower utility, is better suited to the predefined goals, and this is shown quantitatively as 80 points versus 70 points.

The better a predictor, the more utility we can expect and the less mistrust that will likely arise from worries that a poor predictor will underpin life and death decisions. For example, if LYFT were a perfect predictor, i.e. with 100% ability to differentiate who will benefit more from a kidney transplant, then the most important decision is the relative value given by the transplant community to utility (versus, for example, the controversy of including age in the calculation and the potential inequities or decreases in live donation that might ensue) in the grand scheme of kidney allocation. However, as LYFT is a relatively poor predictor, a LYFT-based allocation system does not earn the full points for utility and also loses points for trust, as it will be difficult to explain to patients (and providers) that the transplant community decided to use a poor predictor to choose their priority on the list. How this will compare to proposed alternative systems listed above will not only depend on simulation results but also depend significantly on the a priori selected values of the various aspects of kidney allocation.

Summary

  1. Top of page
  2. Abstract
  3. The Current State of Donor and Recipient Matching
  4. Utility-Based Allocation
  5. Options for Utility-Based Allocation
  6. A Proposed Framework for Evaluating Utility-Based Options
  7. Summary
  8. References

The current kidney allocation system places little value on utility and, as such, significant differences exist at times between the predicted lifetimes of allografts and those of the recipients to whom they are allocated. The addition of a utility-based measure to kidney allocation can minimize these differences, but not without potential side effects. In other words, it is not possible to simultaneously maximize utility, efficiency, equity and predictability, and it is unlikely that any of these can be maximized without the potential cost to autonomy, controversy, trust or live donation. Options for balancing these tradeoffs have been explored, including various adaptations of age matching, net benefit and strategies for avoiding extremes of mismatch. A framework for quantifying the goals of kidney allocation may be helpful in choosing among the many alternative strategies that have been proposed or have yet to be proposed as the debate continues.

References

  1. Top of page
  2. Abstract
  3. The Current State of Donor and Recipient Matching
  4. Utility-Based Allocation
  5. Options for Utility-Based Allocation
  6. A Proposed Framework for Evaluating Utility-Based Options
  7. Summary
  8. References