Remember me
Less than 35% of waitlisted patients in the United States receive a kidney transplant within 5 years of listing. Only 36% of patients on dialysis survive for 5 or more years.1 Despite the shortage of kidneys for transplantation, approximately 20% of deceased donor kidneys are discarded each year.2
Most of the kidneys discarded in the United States are of lower than average quality. More than 50% of the kidneys in the lowest 15% quality range (those with a Kidney Donor Profile Index [KDPI] of over 85%) are discarded.3 However, studies suggest that many of these kidneys are transplantable, and using more of these kidneys could benefit patients waiting for a transplant.1,3–7
Allocation policies for KDPI >85 kidneys and regulatory oversight likely contribute to unnecessary discards in the United States. A 2019 study that applied French kidney acceptance practices to US deceased donor kidneys found that 62% of the kidneys discarded in the United States between 2004 and 2014 would have been transplanted under the French system.3 Studies also show significant differences in acceptance patterns between transplant centers in the United States and find that differences in patients' probability of transplantation are related to their transplant centers' willingness to accept hard-to-place kidneys.8,9
Although evidence from real-world transplantation decisions provides insights into the acceptability of kidneys on the basis of various organ and donor characteristics, information on recipient characteristics and how these affect suitability for a lower-quality kidney from a deceased donor is lacking. This information can be critical in the assessment of policies that seek to improve the allocation of lower-quality kidneys. Understanding of how clinicians make decisions around accepting lower-quality kidney offers for specific waitlisted patients can improve our ability to predict the acceptability of lower-quality kidneys and reduce discard.
This study assesses the trade-offs that physicians are willing to make when selecting recipients for lower-quality or high-KDPI kidneys. The study uses a discrete choice experiment (DCE) to evaluate the relationship between the characteristics of various kidneys and the clinical attributes of potential recipients on the acceptability of organs for transplantation. DCE is a rigorous method for quantifying stated preferences in the health care field.10–12 DCEs have been used to assess the relative importance of different transplant outcomes and to measure patients' treatment preferences with regard to willingness to accept lower-quality kidneys and kidneys with greater infectious disease risk.13–15 DCEs have also been used to measure transplant physicians' preferences with regard to candidate selection and kidney allocation policies; however, these studies have not focused specifically on patient selection for high-KDPI kidneys.16,17
MethodsWe developed and implemented a DCE survey to be administered online to transplant surgeons and nephrologists across the United States. In a DCE, respondents are asked to evaluate a series of experimentally controlled scenarios and to make choices that involve trade-offs between the alternatives in the scenarios.18 The frequency with which respondents accept specific trade-offs indicates the relative importance of the scenario attributes traded.19
DCE questions presented physicians with one of several lower-quality kidneys and a choice between two potential recipients, one of whom was ranked higher on the waiting list. If respondents chose the candidate with a higher rank, they were asked whether they would consider offering the kidney to the other candidate, should the first candidate decline the kidney. Physicians could also choose to not offer the kidney to any candidate. Thus, respondents were asked to judge the expected value of the current kidney for a potential recipient against the expectation that a future kidney could yield better net outcomes for the patient. Other background and practice-specific questions were included in the survey instrument.
Table 1 summarizes the specific kidney and recipient attributes and attribute levels in the DCE questions. Potential recipients had varying levels of time on dialysis, recipient age, and diabetes history. In addition, respondents were provided with information on the recipient degree of sensitization (calculated panel reactive antibody [CPRA]), ejection fraction, HLA mismatch after physical cross match, and overall functional status.
Table 1 - Attribute table Attribute Attribute Levels Kidney characteristics Kidney quality • Kidney 1: 75% (KDPI), 55 (age), R: 0.25 F: 100 (initial pump parameters), R: 0.25 F: 105 (pump parameters after 6 h), 0.8 (nadir creatinine), 1.2 (terminal creatinine), 15% (glomerulosclerosis), no DCD, donor diabetesCreatinine levels shown in milligrams per deciliter. KDPI, Kidney Donor Profile Index; R, resistance levels (in millimeters of mercury [pressure] per milliliters per minute [cardiac output]); F, flow levels (in milliliters per minute); DCD, donation after circulatory death; EPTS, estimated post-transplant survival; CPRA, calculated panel reactive antibody.
The kidneys presented were all of lower-than-average quality (≥75 KDPI) and varied for KDPI, pump parameters, serum creatinine levels, glomerulosclerosis, donor age, and whether the donor had diabetes or donation was made after circulatory death. In total, six kidney profiles were constructed by combining these attributes in a clinically relevant way. One of these kidney profiles excluded information on pump parameters and glomerulosclerosis. Kidney profiles also included information on expected cold ischemia time as a way to portray system-level factors that can affect graft survival. Throughout this manuscript, we refer to the kidney profiles presented as lower-quality kidneys given that all were of lower-than-average quality.
The survey was tested with eight physicians during individual 1-hour cognitive interviews to verify the clarity and completeness of the instrument and to evaluate length of the survey.20 On the basis of feedback obtained from the interview participants, we finalized the instrument (Supplemental Appendix A) and developed the final layout of the DCE questions (Figure 1).
Figure 1: Example choice question. Creatinine levels shown in milligrams per deciliter. CPRA, calculated panel reactive antibody; F, flow levels (in milliliters per minute); KDPI, Kidney Donor Profile Index; R, resistance levels (in millimeters of mercury [pressure] per milliliters per minute [cardiac output]). Figure 1 can be viewed in color online at www.cjasn.org.An experimental design was prepared to control the combinations of kidneys and recipients presented in the choice questions. Study participants were transplant surgeons and nephrologists in the United States who were involved in kidney acceptance decision making at their transplant centers. Participants were recruited through the American Society of Transplant Surgeons, a listing of transplant center medical directors and the American Society of Transplantation. All participants provided informed consent. Recruitment and survey implementation protocols were reviewed and approved by the Institutional Review Board at Northwestern University (STU00208614). Additional information on the development and testing of the survey and recruitment is presented in Supplemental Appendix B.
AnalysisWe summarized the answers to each of the questions probing respondents' demographic and practice characteristics. For categorical questions, we computed the number and percentage of the sample providing each response. Continuous response questions were summarized by the mean and SD values.
We modeled physician choices using an exploded random-parameters logit model. The statistical analysis of choices provides a measure of the effect of changes in the kidney and recipient attributes on the likelihood that a kidney is accepted for a specific patient. For these data, we jointly modeled occasions in which respondents only provided the most preferred option among alternatives and the occasions in which a full ranking of the alternatives was provided. The latter was modeled assuming respondents provided the best and worst option among the alternatives following the assumptions outlined in Flynn et al.21
First, we evaluated whether the kidneys in the experiment could be grouped on the basis of systematic differences in their acceptability for the average recipient. On defining a final set of kidney groups, we added recipient information to the model to help explain acceptability on the basis of both kidney and recipient characteristics.
We then calculated a set of mean kidney acceptability curves (and the 95% confidence intervals [CIs]) by kidney group with longer cold ischemia time using the preference results. We used the preference estimates from the four levels of cold ischemia time in the DCE questions to derive a nonlinear function characterizing how cold ischemia time was associated with changes in the likelihood of acceptance. This function was used to derive choice probabilities for every level between 1 and 40 hours of cold ischemia time.
We used latent-class analysis to maximize our ability to detect groups with similar preferences within our sample using a data-driven process. In latent-class analysis, the model groups choices in a predefined number of classes on the basis of similarities in choice patterns across respondents. Each respondent is probabilistically assigned to the classes. The probability of class membership can be then be correlated with respondents' characteristics.
ResultsA total of 128 physicians accessed the survey between May and October 2021, and 108 answered 4–10 choice questions. Table 2 summarizes the characteristics of these respondents.
Table 2 - Respondent characteristics (N=108) Variable N=108 No. % Age, yr 31–40 11 10 41–50 23 21 51–60 13 12 61–65 23 21 Older than 70 2 2 Missing 36 33 Specialty Nephrologist 20 19 Surgeon 68 62 Missing 20 19 Years of experience, mean (SD) 17.4 (11.6) Which type of patients do you care for? Adult 50 46 Pediatric 2 2 Both 36 33 Missing 20 19 Experience in transplant policy making Institution-level 78 72 UNOS committee participation 36 33 UNOS regional committees 33 31 AST/ASTS committee/workgroup participation 47 44 National other committees 7 7 None of the above 4 4 Missing 20 19 Which of the following best describes your type of program? Private hospital, affiliated with university 16 15 Private hospital, not affiliated with university 5 5 Academic hospital 66 61 Public hospital 1 1 Missing 20 19 What was your transplant center (kidney) volume in 2020? (No. of transplants) 0–99 34 32 100–200 29 27 More than 200 25 23 Missing 20 19 Does your program currently maintain a patient target list? Yes 47 44 No 38 35 Do not know 3 3 Missing 20 19 On average, how long do your blood group O patients wait for a kidney from the moment they start dialysis? Less than 5 yr 56 52 At least 5 yr 52 48UNOS, United Network for Organ Sharing; AST, American Society of Transplantation; ASTS, American Society of Transplant Surgeons.
Estimates from the analysis of respondents' choices are included in Supplemental Appendix C. These results can be interpreted as log odds indicating the change in the relative likelihood that a kidney is accepted. On evaluation of the model specification, we grouped the presented kidneys into three groups (i.e., a group that included kidneys 1–3, a group that included kidneys 4–5, and kidney 6). Before controlling for recipient characteristics, kidneys 1–3 were the most likely to be accepted, followed by kidneys 4–5. All differences between these groups were statistically significant at the 95% confidence level. Expected cold ischemia time was also associated with lower acceptability of kidneys.
The results from the model that included recipient characteristics are shown in Table 3 and are also summarized in Figures 2 and 3.
Table 3 - Model results with recipient covariates Attributes Coef SE(Coefs) SD SE(SD) Main kidney effects Kidneys 1–3 2.62 0.81 2.65 0.67 Kidneys 4–5 −1.04 0.50 2.09 0.34 Kidney 6 −1.58 1.22 4.74 0.84 Cold ischemia time 1-h longer −0.14 0.02 0.12 0.02 Label Ranked above 0.73 0.12 0.74 0.15 Ranked below −0.73 0.12 0.74 0.15 Ejection fraction 38% −0.08 0.04 0.01 0.09 48% 0.08 0.04 0.01 0.09 HLA mismatch 1 of 6 0.04 0.04 −0.03 0.10 5 of 6 −0.04 0.04 −0.03 0.10 Degree of sensitization 0% 0.11 0.07 0.00 0.08 70% −0.15 0.07 −0.01 0.18 95% 0.05 0.07 0.01 0.19 Performance status Normal activity with effort 0.06 0.07 −0.07 0.17 Care for self, unable to carry on normal activity or to do active work 0.13 0.07 0.09 0.16 Requires considerable assistance and frequent medical care −0.19 0.07 0.02 0.23 Recipient age, yr 45 2.65 0.59 0.02 0.24 65 2.72 0.57 0.01 0.12 75 −5.38 −1.14 0.03 0.28 Time on dialysis, yr 2 5.38 1.14 0.00 0.12 5 −2.63 0.57 0.06 0.20 10 −2.75 −0.60 0.06 0.22 Recipient diabetes No diabetes 2.77 0.62 0.03 0.15 Yes, but not insulin dependent 2.65 0.56 0.00 0.09 Insulin dependent, but <10 yr −5.41 1.15 0.03 0.19SE(Coefs), standard error of coefficient; SE(SD), standard error of standard deviation.
Figure 2: Preference weights (log-odds of acceptance given the recipient characteristics) by recipient characteristics. (A) Characteristics considered in EPTS scores. (B) Other recipient characteristics. *Indicates significance at the 90% confidence level, **indicates significance at the 95% confidence level, and ***indicates significance at the 99% confidence level. EPTS, estimated post-transplant survival. Figure 2 can be viewed in color online at www.cjasn.org.
Figure 3: Relative importance of recipient characteristics. Note that bars add up to 100%. Figure 3 can be viewed in color online at www.cjasn.org.Figure 2A presents how the preference weights for each of the levels in the recipient characteristic included in estimated post-transplant survival (EPTS) scores (i.e., recipient age, diabetes history, and time on dialysis). Figure 2B shows the preference weights for the rest of the recipient characteristics. Higher preference weights indicate greater willingness to give a lower-quality kidney to a recipient with each level under the studied characteristics. Note that the weights are plotted relative to the average preference weight for the characteristic. Thus, positive and negative values must be interpreted as indicating preferences above or below the average preference weight of the recipient characteristic. On average, respondents were more willing to accept lower-quality kidneys for younger recipients and those who have spent less time on dialysis and have no history of diabetes. Physicians also were more willing to accept lower-quality kidneys for patients who are listed higher in the waiting list, those with a lower ejection fraction, more extreme (either 0% or 90%) CPRA than those with a CPRA of 70%, and recipients with better functional status. Finally, differences in HLA were not significantly affecting physicians' acceptance decisions.
The two sets of characteristics were plotted separately because the scale of the weights in these two groups varied significantly. This can be seen in Figure 3 where we include the relative importance weights of all recipient characteristics. These importance measures represent the maximum difference in preference weights within a characteristic, indicating the greatest effect each characteristic could have on kidney acceptability.
The kidney acceptability curves (Figure 4) indicate a strong but varying association between expected cold ischemia time and kidney acceptability. For kidneys 1–3, the average acceptance ranged between 15.6% (95% CI, 0.0% to 39.9%) at 30 hours of cold ischemia time and 92.2% (95% CI, 80.9% to 100.0%) at 1 hour of cold ischemia time for about an 83% reduction in the chance of acceptance, and for kidneys 4–5, between 0.4% (95% CI, 0.0% to 1.2%) at 30 hours of cold ischemia time and 23.4% (95% CI, 5.8% to 41.0%) at 1 hour of cold ischemia time for about a 98% reduction in the chance of acceptance. Finally, the acceptance of kidney 6 was between 0.3% (95% CI, 0.0% to 1.1%) at 30 hours of cold ischemia time and 15.2% (95% CI, 0.0% to 46.0%) at 1 hour of cold ischemia for about a 98% reduction in the chance of acceptance. Percentages in the CIs were truncated at 0% and 100%.
Figure 4: Probability of acceptance of kidneys by kidney-quality groups and expected cold ischemia time. Figure 4 can be viewed in color online at www.cjasn.org.We identified two respondent classes on the basis of model fit and parsimony (model estimates in Supplemental Appendix C). The main difference in preferences between the two classes identified was that respondents in class 2 cared almost exclusively about cold ischemia time (i.e., only the parameter for cold ischemia time was significant at the 95% confidence level), suggesting they were more sensitive to increases in the expected time to transplantation. Figure 5 presents a comparison of the average effect of cold ischemia time on acceptability of kidneys across the two classes.
Figure 5: Probability of acceptance of kidneys by kidney-quality groups, expected cold ischemia time, and preference class. Figure 5 can be viewed in color online at www.cjasn.org.Eight covariates were used to explain class membership. Four of the covariates were found to be predictive of class membership at the 95% confidence level (i.e., transplant center volume, type of patient generally treated by the respondent, whether the respondent had experience in transplant policy making, and whether the respondent transplant center has a patient target list), whereas years of experience and the respondent's race were statistically significant at the 90% confidence level. Respondents in class 2 were more likely to (1) be in centers that transplant <100 patients per year, (2) treat primarily pediatric patients, (3) have no experience in transplant-related policy making, and (4) work in a public hospital. Respondents in class 2 were also less likely to work in academic hospitals.
DiscussionThis is the first study that uses a choice experiment to collect evidence from nephrologists and transplant surgeons on decisions to accept lower-quality kidneys for specific recipients. We were able to estimate how various kidney attributes were associated with kidney acceptability and how recipient characteristics moderated such changes. This information is important to inform organ allocation policies that can improve recipient matching for lower-quality kidneys and reduce kidney discard.
As expected, higher KDPI was associated with a lower likelihood of acceptance. Missing information on the donated organ pump parameters and glomerulosclerosis was associated with a lower likelihood of accepting a kidney. Our results suggest that missing information on the kidney pump parameters and glomerulosclerosis meant that physicians likely assumed a worse-case scenario for the quality of the kidneys and reduced their acceptance of these kidneys to levels comparable with very low-quality organs. Cold ischemia time was a significant factor, potentially reducing acceptance by over 80% within a 30-hour window.
Recipient EPTS score appeared to play a key role in determining the acceptability of low-quality kidneys. The recipient attributes associated with EPTS scores carried a substantial and nearly identical weight in respondents' decision making. That said, acceptability changed nonlinearly with changes in recipients' age, dialysis time, and diabetes history. Acceptability appeared to hit a threshold when recipients were between the ages of 65 and 75 years and had between 2 and 5 years of dialysis. Finally, we find that physicians largely avoided the acceptance of low-quality kidneys for recipients who were insulin dependent, even if they had only used insulin for <10 years. The thresholds for age and time on dialysis point to physicians' willingness to offer low-quality kidneys to patients who are expected to receive offers relatively soon. Contrarily, the avoidance of recipients with a history of insulin use points to physicians' potential concerns about patient survival after surgery. Taken together, our results suggest that accepting a kidney for transplantation is a multifactorial problem and may require trade-offs between patient's expected waiting time to next offer, patient EPTS, KDPI of the kidney offer, histological findings, and cold ischemia time.
Cold ischemia time is particularly relevant given that previous research suggesting that an expedited placement of kidneys would be acceptable to a majority of physicians and patients.22 Making a kidney available to all transplant centers after it has accrued a certain amount of cold ischemia time has been shown to offer significant potential for reducing kidney discards from a system optimization perspective.23 Our results also support such a policy and suggest that reductions in cold ischemia time could have an effect on physicians' acceptance of lower-quality kidneys, further reducing the possibility of discard. Our results could help inform the specifics for such a policy, for example, help determine after how many hours lower-quality kidneys should be expedited to minimize the chance of discard.
The results from the latent-class analysis further support the emphasis on curbing cold ischemia time as we identified respondents for whom this was a key factor in acceptance decisions. Center size and working in a public hospital presumably are system-level characteristics that affect physicians' ability to quickly process and transplant lower-quality kidneys, thus making them more sensitive to cold ischemia time.
Avoidance of lower-quality kidneys for younger recipients suggests that delayed graft function (DGF) is a factor in physicians' decisions. Younger recipients would be expected to benefit more from greater longevity of higher-quality kidneys because they would naturally be expected to live longer. However, our result suggests that concerns about surviving DGF overrides concerns over kidney longevity and adds an additional challenge to national allocation policy making.
Several limitations are worth noting in our study. First, although we developed the survey instrument in a way that encouraged preference-revealing answers, the choices reported in the survey do not carry the same level of consequentiality as decisions made in the real world. The qualitative work done to evaluate the validity of the instrument suggests the measure was clear and that the trade-offs presented were meaningful. Both of these help reduce measurement error and can help minimize scenario rejection by respondents.24
We did not show the expected time to next offer for hypothetical patients in the choice questions. Although the team considered adding this information, including it would have required tailoring the choice scenarios in ways that would make their implementation infeasible. The time to next offer would have had to be presented in a way that did not conflict with respondents' expectation for this time given their center policies and the condition of the hypothetical patients. For this reason, we decided to leave out this detail and opted instead for a label signaling the relative position of the patients in the waitlist. Admittedly, the label only provides some sense on the relative time to next offer expected for the two patients. Future work should focus on understanding the effect of expectations about time to next offer on the acceptability of low-quality kidneys.
Majority of respondents were transplant surgeons, so their views were likely more influential in the average preference results. Our analysis of preference heterogeneity did not find that physician specialty was associated with systematic variations in preferences. However, this could be the result of having too small of a sample for this evaluation. Nevertheless, we expected to have more surgeons than nephrologists in the sample because surgeons are typically the ones making final kidney acceptance decisions at their transplant centers.
In addition, our survey did not require respondents to report their transplant center. It is likely that multiple clinicians from some of the same centers responded to the survey. This can potentially bias the results toward physicians from larger centers. A concerted effort was made to ensure adequate representation of physicians from centers with a yearly transplant volume of <100 transplants per year. However, this required using nonrandomized approaches to recruitment (i.e., snowball sampling). Representativeness is a key issue in studies like this. It is very difficult to claim representativeness in preference research because preferences likely correspond to a variety of factors that go into individual experiences and cultural factors. Thus, our work primarily serves as a way to uncover perspectives that exist in the population, without necessarily claiming that those perspectives represent some average for the overall population. Even with this limitation, our results provide important information for discussions around organ acceptance.
In addition, while certain terms were not described using detailed clinical information—for example, ejection fraction was not described as left ventricular ejection fraction with the imaging modality by which this was determined—pretest interview participants did not have concerns with the terms provided. Nevertheless, it is possible that the use of general terms to characterize kidneys or organ recipients increased measurement error in our data. This would have been the case if respondents' choices varied with unobserved variations in their interpretation of the characteristics.
More broadly, kidney or recipient characteristics in the questions may not have been enough to make a decision that properly captured respondents' views on the trade-offs we needed them to consider. In such a situation, respondents could have imputed information on excluded attributes to fill any gaps in the scenario.25 However, during the cognitive testing of the survey instrument, physicians were able to make decisions on the basis of the information presented, making the issue unlikely.
Finally, we recognize that organ acceptance decisions involve multiple stakeholders. However, physicians are a key decision maker in this process. The DCE allowed us the opportunity to isolate how physicians think about recipient characteristics in these decisions. This would not be possible with real-world acceptance data where the perspectives of multiple stakeholders can influence the recorded decisions.
Our results provide information on the views of physicians who are part of the kidney acceptance decision-making process. They suggest that physicians prefer to avoid giving lower-quality organs to the highest-risk patients who may not be able to handle issues such as DGF. They also seemed to favor patients who would likely have to wait longer for another organ offer. Our study also highlights the critical role that factors, such as cold ischemia time and biopsy and pump information, play in physicians' decision to accept lower-quality kidneys. Thus, a policy to reduce cold ischemia time for lower-quality kidneys should be considered to potentially reduce kidney discard in the United States. Our study also finds that EPTS may be an important consideration in the design of such a policy. EPTS <20 is already included in the current kidney allocation system, with KDPI <20 kidneys only offered to the patients with highest EPTS; however, more research is needed to determine how EPTS could be used in the allocation of lower-quality kidneys.
DisclosuresY. Becker's spouse reports employment
Comments (0)