[1] “kidney exchange” by alvin roth, tayfun sonmez …...this waiting list is highly structured,...
TRANSCRIPT
[1]
“Kidney Exchange”
By Alvin Roth, Tayfun Sonmez and M. Utku Unver
Presented by Kevin Kurtz and Beisenbay Mukhanov
February 5th
, 2015
I. Introduction
Background:
Kidney transplantation is the treatment of choice for most kidney diseases, but
there are many more people in need of kidneys than there are kidneys available. Table 1
shows the extent of this demand, but also a staggering undersupply of kidneys. The result
is long waitlists of patients waiting, often for years, for a kidney to become available.
Due to this wait, thousands of patients die every year without receiving a transplant.
Kidneys for transplantation can come from two sources: donors who are willing to
give a kidney – usually to a loved one or relative; and cadavers. A problem with the
former is that not everyone who is healthy enough to donate a kidney and wishes to do so
can donate a kidney to his or her intended recipient. A successful transplant requires the
donor and recipient to be compatible in blood and tissue types. A similar problem exists
in the latter option as well, for the characteristics of cadavers cannot be guaranteed in
advance. This gives rise to the possibilities of a kidney exchange: where incompatible
patient-donor pairs can swap kidneys.
Source: (extracted from Roth 2004a)
[2]
II.A Kidney Transplantation
While there is a distinct shortage of live donors for all the number of transplants
needed, the sale of organs is strictly prohibited by the National Organ Transplant Act
(NOTA) of 1984. This has prevented a market oriented solution to the chronic shortage
of kidneys available for transplant in the United States. Live donor kidneys are preferable
to cadaver kidneys due to a higher survival rate from surgery. Thus, there is a distinct
need for an exchange mechanism of kidneys from live donors.
Types of Exchanges:
Direct Exchange – Also called paired exchange, this involves two patient-donor
pairs in which a transplant from the donor to the intended patient is infeasible, but
successful transplants are possible using the kidney from the other patient-donor
pair.
o Example: Suppose there are two patients, A and B. Each of them have
donors, X and Y, respectively, who are willing to give them a kidney.
Furthermore, suppose patient A is compatible with kidney Y, but not
kidney X, and patient B is compatible with kidney X, but not Y. It is
possible for patients A and B to exchange their donor kidneys with each
other rather than be put on a waitlist, which is a Pareto Efficient outcome.
Indirect Exchange - Also called list exchange, this involves an exchange
between one incompatible patient-donor pair and the cadaver queue. In return for
donating a kidney to the cadaver queue, the patient in the pair receives a high
priority listing on the cadaver queue. This is welfare improving for both the
general public and for the patient in the patient-donor pair.
o Note: this could have a negative impact on O-bloodtype patients, as they
have the fewest kidneys available to them on the cadaver queue. Since
anyone can receive a transplant from an O-bloodtype donor, it is less
likely that O-bloodtype patients without a donor will be able to receive a
kidney in the cadaver queue.
o Example: Suppose there is one patient, A, who has someone willing to
donate their kidney, X. If patient A is not compatible with kidney X, (s)he
has the option of giving kidney X to the pool of cadaver kidneys in
exchange for a high placement on the cadaver queue.
Cadaver Queue:
Those who do not have a donor – or whose donor’s kidney is not compatible with
them – are referred to a waiting list for cadaver kidneys. This waiting list is highly
structured, with scores being assigned to each candidate based on several factors. Each of
these factors (blood type, tissue [or HLA], age, size) can significantly affect whether or
not a transplant is likely to succeed. Each of these preferences is given a numeric score.
This allows for strict preferences within any kidney exchange mechanism.
[3]
II.B Mechanism Design
The Housing Market Analogy:
Roth compares the market for kidney exchange to that of a model of the housing
market created by Shapley and Scarf [1974]. In this model, each agent is endowed with
an indivisible good (their house), and has strict preferences across all houses, but there is
no money in the market. They use David Gale’s Top Trading Cycle (TTC) mechanism
to produce an allocation of houses. The TTC mechanism works in situations where the
quantity of goods is fixed and known, and where individual choice is ordered by strict
preferences. It works as follows:
Each agent points to the house that they want most.
Agents that point to their own house are removed from the market.
There is at least one cycle as a result of this. The trades in this cycle are carried
out, and both agents and houses are removed from the market.
Continue from the beginning of the process until no agents remain.
The result of the TTC process is a unique, Pareto efficient outcome. As Roth found in
later studies [1984], it does not pay for agents to lie in this mechanism, as they will be
rewarded with a house that is not their most preferred choice.
The College Dorm Analogy:
Another analogy that Roth makes is to a later study by Abdulkadiroĝlu and
Sönmez [1999] of the housing allocations on college campuses. The difference between
this and the housing market model mentioned above is that there is the introduction of
unallocated goods (unoccupied rooms) and agents who are not endowed with goods (new
students). So, the authors made some changes to the TTC model, which is as follows:
Each student reports their strict preferences over all rooms.
Assign the first student (based on priority) their first choice, and so on, until a
student requests a unit that is already owned.
Modify the ordering by moving the tenant of the requested unit to the front of the
line ahead of the person who requested that unit. Then continue with the
procedure once more.
If at any point a cycle forms, assign all students in that cycle the units they desire.
The key innovation here is that people who own a unit already are guaranteed to keep
that unit if they enter the market. By being placed in the front of the line before their unit
is allocated, they have a risk-free opportunity to upgrade. Thus, it is a Pareto efficient
system where every endowed agent enters the market.
There are also strong parallels between this system and the kidney exchange system.
The housing queue is a parallel to the cadaver queue. If a donor gives a kidney to the
[4]
cadaver queue, his intended recipient jumps to the top of the queue, just as in the college
room scenario. The only difference, albeit a key one, is that the number of rooms in the
college scenario is fixed, whereas the number of kidneys is not. It is also not known how
long one will wait in the cadaver queue until a compatible kidney becomes available.
III.A Top Trading Cycles and Chains Mechanism (TTCC)
Model Variables:
n number of patient-donor pairs
ki ith kidney; the kidney intended for the ith patient
ti ith patient
K Set of n available kidneys
Ki Feasible set of kidneys in K (Ki ⊂ K) which are compatible with patient i
w Option of entering the waitlist for a cadaver kidney with a high priority
Pi The strict preferences of ti over Ki ∪{ ki, w},
Ci A cycle which constitutes a direct exchange between two or more patient-
donor pairs
Wi W-chain, assigned in instances of multiple W-chains occurring in a single
period
Model Terminology:
Head – Pair who donate a kidney to the cadaver queue
Tail – Pair who receive a high priority for a kidney from the cadaver queue
Cycle – A series of direct exchanges between patient-donor pairs which form a
closed loop. It is represented by the left diagram in Fig. 1.
W-chain – A series of direct exchanges between patient-donor pairs that do not
form a closed loop, but instead contain indirect exchanges at the ends (referred to
as Head and Tail). At the head of the W-chain, the donor gives their kidney to the
cadaver queue in exchange for another patient-donor pair’s kidney. At the tail of
the patient-donor queue, the patient receives high priority on the cadaver queue in
exchange for giving their donor’s kidney to another patient from the live donor
pool. It is shown by the right diagram in Fig. 1.
There are a few important things to note:
In a patient’s preference set, Pi, there can never be an option ranked lower
than ki. This ensures that the process is Pareto Efficient, as no one will end up
worse off than if they had not entered
Preferences are determined based on probability of survival
Lemma 1 - Consider a graph in which both the patient and the kidney of each pair are
distinct nodes as is the wait-list option w. Suppose that each patient points either toward
a kidney or w, and each kidney points to its paired recipient. Then either there exists a
cycle, or each pair is the tail of some w-chain.
[5]
Fig. 1 – Visual representation of two conditions of lemma 1
III.B The Exchange Mechanism
Step 1 – All kidneys are available, and all agents are active. Each patient ti points
either to his most preferred kidney or to the wait-list option, w. Each remaining
kidney ki points to its paired patient, ti.
Step 2 – At this point, by Lemma 1, there is either a cycle or a w-chain, or both.
o If there is not a cycle, proceed to step 3.
o If there is a cycle, carry out the corresponding exchanges and remove all
patients and kidneys involved from the mechanism.
Once this has been done, have all remaining patients point to their
top choice among the remaining kidneys. Locate any new cycles,
remove them, and repeat until there are no more cycles.
Step 3 – At this point, all remaining pairs are the tails of w-chains.
o Select one of the chains according to whichever chain selection rule you
choose. This assignment is final for those in the selected w-chain.
Depending upon which chain selection rule you use, the w-chain is
either removed from the mechanism, or patients and kidneys
remain in the mechanism, but are passive instead of active (cannot
accept new assignments)
Step 4 – After the w-chain is selected, new cycles may form, so you must repeat
steps 2 and 3 until there are no patients or kidneys are left.
By the end of this procedure, every patient with a living donor has either been
assigned a living kidney for transplant or a high priority position on the waiting list.
w
ti
ki
ti
ki
ki
w
ti
ki
ti
[6]
Example:
Suppose there is a list of patient-donor pairs with the following preferences.
t1: k6, k7, k5
t2: k1, k5, k7, w
t3: k1, k9,w
t4: k1, k5
t5: k1,w
t6: k3, k4, k8, k7, k1
t7: k9, w
t8: k4, k1, k8, k6
t9: k8, k4, k9
Round 1
Fig. 2 – Following Step 1, each patient points to the kidney they prefer most.
t2
t6 t3 t7
t4 t8
t9
t5
k6
:
k2
:
k3
:
k5
:
k1
:
k4
k8
:
k9
:
k7
:
w
t1
[7]
Round 1
Fig. 3 – As a part of Step 2, a cycle C1= (k1, t
1, k6, t6, k3, t3) is identified.
Round 1
Fig. 4 – As part of Step 2, the cycle is removed from the pool of patient-donor pairs. This allocation is now
fixed, and those kidneys in it are no longer available for consideration
t2
t6 t3 t7
t4 t8
t9
t5
k6
:
k2
:
k3
:
k5
:
k1
:
k4
k8
:
k9
:
k7
:
w
t1
t2
t7
t4 t8
t9
t5 k2
:
k5
:
k4
k8
:
k9
:
k7
:
w
[8]
Round 2
Fig. 5 – As per Step 2, those remaining pairs choose their most preferred kidney. Since there are no longer
any cycles, we proceed to Step 3. There are two W-chains remaining, W1 = (k7, t7, k9, t9, k8, t8, k4, t4, k5, t5, w)
and W2 = (k2, t2, k5, t5, w). Which you choose depends upon what chain selection rule you are using. For this
example, we will use the longest chain rule. Thus, we take W1 and switch its occupants from an active role
to a passive role. Thus, they are still in the pool of donor pairs, but cannot change their current assignments.
Round 3
t2
t7
t4 t8
t9
t5
k2
:
k5
:
k4
k8
:
k9
:
k7
:
w
t2
t7
t4 t8
t9
t5
k2
:
k5
:
k4
k8
:
k9
:
k7
:
w
Fig. 6 – As per Step 4, we
repeat all previous steps
once a W-chain is selected.
As a result, t2 switches to
their next preferred kidney,
k7. This creates a single W-
chain and is the end of the
TTCC mechanism. Kidney
k2 will be given to someone
from the cadaver queue,
and patient t5 will enter the
cadaver queue with a high
priority position
[9]
Alternative W-Chain Rules:
Minimal w-chains:
o In this rule, the smallest w-chains to be formed are removed from the
mechanism first. The result from this will be an increased number of live
donor kidneys being allocated to the cadaver queue.
Maximum w-chains:
o In this rule, the largest w-chains to be formed are removed from the
mechanism first. If there are multiple w-chains of the same length which
are not unique, then a tiebreaker is used (compatibility, age, etc) to
determine which chain is removed.
o There is also the option of keeping the longest w-chain in the mechanism,
but having it remain in a passive role, rather than an active role. This
means that the selected patients have their choices locked, and cannot
choose a new kidney. This is done in hopes that after the mechanism goes
through one more iteration, the w-chain can be lengthened.
The benefit of keeping the w-chain in the mechanism for the next
round is that it can be a Pareto-improving option. In the example
above, patient t2 would have to move to the cadaver queue if the w-
chain was removed at the end of round 2. By keeping the w-chain
there, but in a passive role, t2’s welfare improved, but not at the
expense of any other patient or those in the cadaver queue (the
effect on which we will consider net neutral).
o The benefit of this chain rule is increased welfare among those in the live
donor pool as opposed to using the minimum w-chains rule.
Prioritized pairs:
o Choose the patient-donor pair that has the highest priority and remove it
from the mechanism. This has the benefit of making sure that the patients
with the most need are able to receive the kidney that best suits them.
o As with the maximum w-chains rule above, this can be altered so that
instead of removing the w-chain from the mechanism immediately, it can
remain in a passive role in hopes of lengthening it.
Prioritize O-type donor pairs:
o This is a special case of the prioritized pairs rule above. Patients whose
donor has O-type blood are given priority in this situation. The w-chain
starts with the highest priority pair, and if the donor in that pair has O-type
blood it is immediately removed.
o The effect of this is a significant increase of type O kidneys to the cadaver
queue, while there may be significant efficiency losses within the
mechanism itself.
IV. Efficiency and Incentives
The paper considers the Pareto efficiency of the kidney allocation obtained by the
TTCC mechanism. If there is no other patient-kidney matching which is not worse than
the initial matching for all patient-donor pairs and strictly better for at least one pair, then
[10]
this initial matching is Pareto efficient. In the article a kidney exchange mechanism
regards as efficient if it always gives a Pareto efficient at any given time.
The authors claim the following two theorems about the TTCC mechanism’s
efficiency and when the TTCC is strategy- proof.
The Theorem of TTCC mechanism’s efficiency
Claim: The TTCC mechanism is efficient if it applies a chain selection rule in
which w-chain chosen at every intermediate round keeps in the procedure and its tail
stays available for the following round.
On the other hand the TTCC mechanism in which w-chain’s tail is not available
for the next round is not necessary Pareto-efficient matching.
Proof: Suppose that the TTCC mechanism is implemented with a chain rule such
that w-chain chosen at every intermediate round keeps in the procedure and its tail stays
available for the following round. Then, after Round 1 every patient that takes his final
assignment has his top-preferred kidney. In Round 2 every finalized patient takes his
top-preferred kidney among leftover grafts including the tail kidney from chosen in
Round 1 w-chain. Hence these patients can be made better off only making worse off the
patients finalized in Round 1. Continuing this procedure, there is no any patient that can
get better choice without making worse off another patient that was finalized in previous
rounds. Therefore, the TTCC mechanism implemented with a chain rule that keeps tail-
kidney for following round is the mechanism leading to Pareto efficient allocation at any
given time.
Example 1: Suppose there are the following preferences of five patient-donor
pairs.
t1: k5 k1
t2: k5 k3 k2
t3: k4 k5 w
t4: k5 w
t5: w
Suppose one uses the TTCC mechanism with the chain selection rule that chooses
the longest w-chain and extracts it.
Then, in Round 1 there is no any cycle and the longest w-chain is (k3, t3, k4, t4,
k5, t5, w). After removing it there will be two cycles (k1, t1) and (k2, t2). The final
outcome of the matching will be (t1-k1, t2-k2, t3-k4, t4-k5, t5-w). However, the Pareto
efficient matching will be (t1-k1, t2-k3, t3-k4, t4-k5, t5-w).
The Theorem of TTCC’s Strategy-Proofness
The second theorem lists the chain selection rules that guaranty the strategy-
proofness of the TTCC. In this aspect the paper limits the strategy space to the space of
declared preferences. At the same time the kidney transplant process might consist other
[11]
strategic issues that are not considered. For example, patient might register in multiple
regional transplant centers and as result be on multiple queue lists.
According to Roth (1982), truly stated preferences in the housing model are the
necessary strategy to prevent the profit gain by an agent that misrepresents his
preferences. Because the static kidney exchange and the housing model are similar, the
authors refer Roth (1982) results for strategy-proofness of the TTC mechanism for the
kidney exchange without indirect exchanges.
The TTCC’ strategy-proofness depends from an implemented chain selection rule.
Claim: The TTCC mechanism applying any of the mentioned alternative w-
chain rules except the maximum w-chain rule is strategy-proof.
The proof of this theorem is skipped and can be found in the paper in question.
At the same time the most attractive chain selection rules are the prioritized pairs
and prioritize O-type donor pairs rules with the condition that w-chain-tail kidney keeps
for the next round.
Both of them lead to an efficient and strategy proof TTCC mechanism. Applying
of rule f, in addition, increases the number of deficit type kidney O available for patients
waiting cadaveric kidneys.
On the other side, the TTCC with a chain selection rule choosing the longest w-
chain is not strategy proof because a patient can benefit by influencing the w-chain
lengths via preference falsification.
Consider the following illustration of the preference misrepresentation.
Example 2.
Suppose there are seven donor-patient pairs (k1,t1), ….(k7,t7) with the following
truthful preferences.
t1: k2 k5 w
t2: k5 k4 k1 w
t3: k2 k6 k3
t4: k6 k1 k6
t5: k1 k3 k5
t6: k5 w
t7: k3 k7
In this example patient t3 stays with his donor for the next time, while patient t4
gets kidney k6. However, if patient t3 misrepresents his preferences as k6, k2,k3, he
receives kidney k6 instead of patient t4.
V. Simulations
The paper provides the simulations’ results in order to compare the welfare gains
that might be obtained by applying the TTCC exchange mechanism and other
mechanisms, and by this way support the TTCC concept.
V.A. Data
[12]
The data for simulations was limited because of lack of wide detailed information
about patients and donors. The information that the authors could find is represented in
Table II (was taken without any changes form the paper). In addition, the authors used
the Zenios’s (1996) HLA characteristics’ distribution based on the UNOS registration
data for period of 1987-1991. Because the information about the willingness of donor-
patient pairs to exchange the donor’s kidney for priority on the cadaveric waiting list, the
authors tested the reliability of results by simulating a wide diapason of preferences. In
the current article they demonstrate the results of the simulations with two assumptions
that there is no the pair who is willing to exchange the donor’s kidney for priority on the
cadaveric graft queue and that 40% of the pairs are willing to make that exchange.
* The information about similar computations in constructions of other economic empirical designs might
be find in Roth [2002]
V.B. Assumptions
Assumption 1
[13]
“All HLA proteins and blood type are independently distributed following
Zenios”. To simplify the simulations, the authors consider a scenario with unrelated pairs
of donors and patients (spouses, friends and so on). In 2001 the rate of this group was
about 25.3% among all living-donor grafts.
Assumption 2
The authors assume that all patients and donors are adults (of age 18-79). Hence
they calculate the conditional age distributions of the patients and the unrelated
nonspousal donors given that the donors and patients are adults. For this purposes Table
II was used.
Assumption 3
HLA characteristic and blood type distribution of patients and donors are the
same, “the characteristics of a nonspousal unrelated donor are independently distributed
with the patient, and the characteristics of a spouse are independently distributed with the
patient except his or her age”. The ages of spouses are the same.
Assumption 4 – Preference Determination
The authors make assumption that preferences of donors and patients over
available kidneys depend from the probability that the implanted graft will be not
rejected. They use the results of survival analysis published in Mandal et al. (2003) and
based on data obtained in 1995-1998 from the United States Renal Data System
(USRDS). The authors suggest two types of preference construction.
“Rational”
for patients
from 18 to 59
for patients
from 60 to 80
U(x,y) = -0.514x-y/10 U(x,y) = -0.510x-y/10
a monotone decreasing function, where
x – the number of HLA mismatches, x 0, 6 y – the donor age, y 18, 80
“Coutious” The patient ti prefers the kidney of donor kj if and only if - kidney of its own donor ki is incompatible with him, or - kidney of its own donor is compatible, but has more HLA
mismatches than kidney kj has. In both methods the preference of a patient ti are determined only over kidneys kj
that are ABO compatible with the patient.
HLA mismatches determine through pre-transplant crossmatching test. The test
may be positive or negative. The positive crossmatch means that patient’s antibodies will
attack donor’s HLA that increase the graft failure risk.
The Marginal Rates of Substitution of one additional HLA mismatch by
decreasing in the donor age were determined by using Mandal’s et al. (2003) estimations,
and are
[14]
5.14 for patients from 18 to 59, and
5.10 for patients from 60 to 80
However, in Mandal et al. (2003) there are also such factors as patient race and
age, patient health history (especially the history of diabetes and the period of the
treatment with dialysis) that influence the failure risk.
Assumption 5
The authors used statistics from different papers that were based on data from
different periods. Therefore they assume that characteristics’ distributions of new patients
are independent from time period and the same for the same population, i.e. American
Caucasian ESRD patients represented in this paper.
V.C. Simulated Mechanism
Method of
simulation
Monte-Carlo
Size of simulation 100 trials
Size of population
(n) a) 30 donor-patient pairs
b) 100 donor-patient pairs
c) 300 donor-patient pairs
Exchange regimes 1) no-exchange
2) paired exchange
3) TTC mechanism
4) Paired and indirect exchange
5) TTCC mechanism implemented with efficient and strategy-
proof w-chain selection rule when patient-donor pairs prioritize
in a single list , and w-chain is chosen with the highest priority
pair and keeps for next Round
Steps of the
Simulations
1) Random simulation of a sample of n-size population using the
characteristics of donor-patient pairs
2) Determination of four preference sets for each patient
- two sets using “rational” utility function and assuming 0% and
40% of donor-patient accepting cadaveric wait-list option.
- two sets using “cautious” approach and also assuming 0% and 40%
of donor-patient accepting cadaveric wait-list option
3) Simulation of the five mechanisms using all four preference set
for every population size.
V.D. Results of the Simulations
The simulations’ results are represented in Tables III, IV and V taken from the
paper. The rows of the tables consist of different population sizes and exchange regimes
under the different preference constructions.
[15]
Table III Column 4 is the percentage of living grafts that were received by patients
under every exchange regime.
Column 4 = Column 5 + Column 6
Column 7 is the number of HLA mismatches for an average graft.
The numbers in parentheses are standard errors of the evaluations.
Table IV The last five columns consist information about the percentage of the
population size that didn’t receive transplant and wasn’t willing to trade
their donor’s kidney for the priority in the waiting list.
Table V The columns of the table represent the number, the average length and
the maximum length of cycles and w-chains. The last columns called
“Longest” show the length of the longest cycle/w-chain among all
simulated 100 trials.
[16]
[17]
The authors claim that the applying of the TTCC mechanism leads to significant
gains in the number of kidney exchanges and the quality of compatibility. There is some
interpretation of the obtained results in the paper.
1. The TTCC mechanism gives higher rate of adaptation of the kidneys from
unrelated living donors and decreases the number of HLA mismatches. This
positive effect improves as the donor-patient size rises.
2. The average and maximal lengths of cycles and w-chains increase with
increasing of the sample size.
3. The patients with the O type blood but without living donor advantage from
TTCC mechanism compared with indirect/paired-kidney exchange
mechanism. The TTCC decreases the number of the O type patients having
incompatible living donors and willing to change their kidneys for priority on
the cadaveric waitlist because there is no an available compatible kidney from
the other donor-patient pairs. This result might be explained by the fact that in
the TTCC mechanism A, B or AB patients that have donors with O type blood
but with some HLA mismatches can be matched with other pairs’ donors with
the same blood type and less HLA mismatches, and therefore it makes
available more O type donors for other O type patients including the O type
patients from the waitlist.
More detailed discussion consists in Roth, Sonmez and Unver (2003).
VI. Developments Since Publication
At the time of Roth’s paper, there was only the infrastructure in place for pairwise
exchanges to take place within the same hospital or treatment center, and this was the
extent of kidney exchange for a few years. But in subsequent years more papers came out
supporting the concept and by 2010 several infant kidney exchange programs had
cropped up, encompassing about 50 hospitals each. However, participation was not
uniform, as 20% of enrolled hospitals accounted for 50% of submitted kidney pairings.
In Ashlagi and Roth [2014], the authors studied some problems that have arisen
from the partial implementation of the kidney exchange system. As the program has
expanded, and hospitals began administering it rather than doctors, free riding began to
emerge. Hospitals would still do paired exchanges that arose within their system (“Easy
matches”) rather than submit them to the regional exchange network. Only “hard”
matches would be submitted to the exchange network. Hospitals would often join
multiple networks in an effort to free ride off of both of them. This led to efficiency
losses in the form of shortened w-chains and suboptimal kidney allocation, as well as
competing systems for kidney exchange.
More recently, a Johns Hopkins team has been able to nullify the effects of having
both positive crossmatch and different blood types on the probability of a successful
kidney transfer. The technique, called plasmapheresis, could potentially render a kidney
[18]
exchange system obsolete, as one of the largest barriers to transplant compatibility would
be removed. Currently this practice is still in early trials, so there will still be a need for
an exchange system for the near future, at least.
VII. Extensions and critique
Applications:
Due to the many requirements for a TTCC mechanism (no currency, indivisible
goods, and strict preferences) there are few other situations where the TTCC mechanism
could be useful. Universities competing for students may be able to use this system to
optimize admission of candidates. If universities were to share lists of applications and
their strict preferences for each student, they could collude to selectively admit students
that they wish to accept – knowing that other universities would not accept them. This
would replace the current system where they accept many students and expect to have
some students turn them down. However, there are significant legal ramifications to this
and it is possible that lying about preferences may be beneficial for colleges in these
scenarios. Also, there would have to be a system created to determine the tiebreakers for
universities. This would work better in other countries which a central authority allocates
students to schools, rather than the individual choice that presides in the United States.
The TTCC mechanism could also be applied to the distribution of public goods, in
special scenarios. After natural disasters, certain public goods could be scarce, and the
different affected regions could have different needs for aid. For example while after a
hurricane all residents on an island may need temporary shelter, some may have no
access to fresh water. An aid drop of water to the shelter-less community could be
diverted to the community in need of water, in exchange for a promise to give the first
available building materials to them. So a barter system could be put in place where the
government receives the needs of each community and allocates a scarce supply of aid to
each. There are also similar problems with this application, as misrepresenting
preferences could be beneficial to communities.
Focus on patient preferences.
The suggested exchange algorithm ranges compatible with a patient kidneys
relying only on their survival rate. This approach may miscount preferences of other
players participating in the exchange process. Some of these omissions might encourage
blocking the complex exchange. For example, hospitals might accept only the exchange
schemes that don’t decrease the number of transplant surges inside of their patients.
Therefore, as mentioned in Ashlagi and Roth [2014] they will try to provide for TTCC
exchange only donor-patients hard-matching with their other pairs. This will lead to
significant decrease of TTCC mechanism’s efficiency. As another example, a donor-
patient pair might have preferences not only over compatible kidneys but also over
patients who can get their kidney. And therefore there might be situation when the pair
will want to change their preferences over compatible kidneys in order to make the
[19]
allocation where their more preferable patient will get their kidney. That means that a
donor-patient pair might have preferences over kidneys depending from preferences over
possible patients.
The length of the cycle/w-chain.
The application of the TTCC mechanism implies long cycles and w-chains.
Moreover, the length of the cycles and w-chains increase as the population grows that is
the necessary condition for the increase of the TTCC efficiency. For example, when the
population size is 30 pairs, the longest cycle of “cautious” TTCC mechanism involves 10
pairs. As population grows to 300 pairs, this index increases to 22 pairs. This fact
requires two times more operation rooms and surgeon teams because all operations
should be made at the same time to avoid the risk of donor’s rejection. The authors
mention that kind of problem that should be overcome. However, in order to address the
current economical capabilities this paper might be extend by considering the ways to
decrease potential cost from TTCC’s implementation.
VII. Conclusion
The authors suggest adopting the centralized TTCC mechanism that based on the
idea of getting higher potential gains from multi-side trades rather than from simple two-
side trades. Consequently this mechanism will increase the number and quality of kidney
exchanges in comparison with existing pair and indirect exchange practices. The new
mechanism is designed by the extension of Gale’s top trading cycle (TTC) mechanism
involving donor-patient pairs participating in indirect exchanges.
References
Ashlagi Itai, Roth, Alvin E. (2014), “Free riding and participation in large scale, multi-
hospital kidney exchange.” Theoretical Economics, 9, 817-863.
Roth, Alvin E., Tayfon Sonmez and M. Utku Unver (2004),“Kidney exchange.”
Quarterly Journal of Economics, 119, 457-488.
Roth, Alvin E., Tayfon Sonmez, and M. Utku Unver, “Kidney Exchange.” NBER
Working Paper No. 10002, September, 2003.
http://www.hopkinsmedicine.org/transplant/programs/kidney/incompatible/