Jump to
Abstract
Background—Despite a national organ-donor shortage and a growing population of patients with end-stage heart disease, the acceptance rate of donor hearts for transplantation is low. We sought to identify donor predictors of allograft nonuse, and to determine whether these predictors are in fact associated with adverse recipient post-transplant outcomes.
Methods and Results—We studied a cohort of 1872 potential organ donors managed by the California Transplant Donor Network from 2001 to 2008. Forty-five percent of available allografts were accepted for heart transplantation. Donor predictors of allograft nonuse included age>50 years, female sex, death attributable to cerebrovascular accident, hypertension, diabetes mellitus, a positive troponin assay, left-ventricular dysfunction and regional wall motion abnormalities, and left-ventricular hypertrophy. For hearts that were transplanted, only donor cause of death was associated with prolonged recipient hospitalization post-transplant, and only donor diabetes mellitus was predictive of increased recipient mortality.
Conclusions—Whereas there are many donor predictors of allograft discard in the current era, these characteristics seem to have little effect on recipient outcomes when the hearts are transplanted. Our results suggest that more liberal use of cardiac allografts with relative contraindications may be warranted.
Introduction
Despite the availability of successful medical therapies for end-stage heart failure, and now of mechanical circulatory support,1 heart transplantation remains the best option for appropriate candidates with end-stage heart disease.2 The severe and persistent shortage of donor organs, however, considerably limits the availability of heart transplantation. Although it is estimated that >20 000 patients could benefit from this life-saving procedure each year,3 only 1853 heart transplants were performed in the United States in 2009, with a concurrent waiting list mortality of 13.7% (OPTN/SRTR Annual Data Report 2010). These alarming statistics have motivated the search for ways to increase the size of the donor pool and the use of available organs.4
Clinical Perspective on p 309
In 2003, Sheehy et al5 estimated the annual number of brain-dead potential organ donors in the United States to be between 10 500 and 13 800; of these potential donors, only 42% donated ≥1 organ for transplantation. Although public health efforts to increase donor identification and consent rates address a major limiting factor in donor availability, the transplant community must also focus on ways to increase the use of suitable grafts from available donors. At this time, ≈60% of currently available cardiac allografts are discarded because of stringent acceptance criteria that have not been rigorously tested in clinical and research settings. In fact, single-center experience using marginal- or high-risk donor hearts for transplantation has demonstrated excellent clinical results.6–9
Given the discrepancy between the pressing need for donor organs and the low-use rate of available grafts, we sought to identify current predictors of cardiac allograft nonuse, and to determine whether these predictors are in fact associated with adverse recipient outcomes.
Methods
Subjects
We studied a contemporary cohort of 1872 organ donors between the ages of 14 and 69 years who were managed by the California Transplant Donor Network (CTDN, Oakland, CA) from 2001 to 2008. This age range was chosen for study based on the observation that allografts from donors <14 and >69 years of age were not accepted for transplantation into adult recipients during this time period. CTDN is the largest organ procurement agency in Northern California and supplies donor organs mainly to transplant centers in northern and central California, and occasionally to neighboring regions. Potential brain-dead organ donors were identified by treating physicians at hospitals throughout the region, and consent for organ donation was obtained from family members or next-of-kin. Management of the organ donor was subsequently assumed by CTDN staff, and consent was obtained from the donor’s family to collect patient data and biological samples. This study was approved by CTDN and by the Stanford University Institutional Review Board.
Donor Management
During the 8-year time period studied, all brain-dead organ donors at CTDN were managed according to a standardized protocol that included: methylprednisolone administered at the onset of donor management and until organ procurement (15 mg/kg every 12 hours); dopamine as the first-line inotropic agent (maximum 20 μg/kg per minute); phenylephrine as the second-line vasoactive agent (maximum 300 μg/min); intravenous fluid or loop diuretic administration to obtain a goal central venous pressure of 5 to 8 mm Hg and a urine output of >30 mL/h; electrolyte repletion to achieve normalization of potassium, phosphorous, magnesium, and calcium levels; empirical antimicrobial therapy with vancomycin and levofloxacin; and inhaled, nebulized albuterol (2.5 mg every 4 hours). Vasoactive and inotropic medications were titrated according to pulmonary artery catheter readings to achieve a target systemic vascular resistance of 800 to 1200 dynes-seconds/cm5 and cardiac index >2 L · min−1 · m−2. Esmolol infusions were initiated for tachycardia that was deemed unrelated to β-agonist infusion and were discontinued on initiation of organ procurement. Thyroid hormone (levothyroxine) was administered when requested by the accepting transplant centers.
Clinical Data
On assumption of donor management, comprehensive data on donor-level variables were recorded by CTDN staff in a standardized fashion, including demographic variables, cause of death, health-related behaviors, and past medical history. Data on comprehensive laboratory testing were also recorded, including serologies; hematologic, hepatic, and renal function indices; and cardiac enzyme assays. Standard testing for potential donors who were not immediately ruled out for cardiac graft donation (because of known coronary artery disease treated with percutaneous stents or bypass surgery, previous cardiac valve surgery, lack of consent for donation, or coroner exclusion) included an ECG; ≥1 echocardiogram; and a coronary angiogram for male donors older than 40 years and female donors older than 45 years. All cardiac testing was performed and interpreted at the donor hospital, and results were recorded by CTDN personnel. Data on vital signs, invasive hemodynamics, and medications were also recorded.
Donor data were extracted from the medical records and were entered into the research study database by study personnel. A subsequent quality-assessment review of 5% of the medical records was performed, reviewing 177 fields per donor chart, and demonstrated >95% accuracy of data collection.
Allograft Use
All donor hearts transplanted in the United States were considered used and followed for recipient outcomes. Data on heart transplant recipient characteristics and post-transplant outcomes were obtained from the United Network for Organ Sharing provided by way of Standard Transplant Analysis and Research files. Any hearts that were part of a multiorgan transplant were excluded from further follow-up.
Donor Predictors of Cardiac Allograft Nonuse
Eleven donor risk factors for allograft nonuse were selected a priori, based on previous literature. These included (1) donor age >50 years,2,10–12 (2) female sex,13–17 (3) cerebrovascular accident (CVA)/stroke as the cause of death,12,18–20 (4) hypertension,11,12 (5) diabetes mellitus,10,11 (6) history of cocaine or methamphetamine use,21–23 (7) high-inotrope requirement (dopamine >10 μg/kg per minute) during donor management,10–12,24 (8) cardiac troponin I >1.0 μg/L,25–27 (9) left-ventricular dysfunction (defined as left-ventricular ejection fraction <50%),12,28 (10) left-ventricular regional wall motion abnormalities,10 and (11) left-ventricular hypertrophy (defined as septal or posterior wall thickness >1.1 cm).6,12,29,30 A cutoff of 1.0 μg/L was used to define an elevated troponin level based on the knowledge that donor hospitals used a variety of assays from multiple manufacturers to perform this test, and reference values for a positive troponin level varied from 0.04 to 1.0 μg/L, depending on the specific assay used. We therefore selected the upper boundary to ensure specificity in capturing abnormal troponin values, albeit with a recognized loss of sensitivity.
Recipient Outcomes
Based on the assumption that donor characteristics would preferentially influence short-term post-transplant outcomes (rather than long-term outcomes such as overall survival), we examined the following 3 recipient outcomes: delayed hospital discharge (hospital discharge after 21 days post-transplant), 30-day graft survival, and 1-year graft survival. A cutoff of 21 days for recipient post-transplant hospitalization was chosen before data analysis based on the clinical experience of the authors. Recipients were classified as having a length of stay less than 21 days if they were discharged within 21 days after the heart transplant procedure and did not die (or were not retransplanted) before discharge. Graft failure was defined as death or retransplantation within the specified time period.
Recipient covariates included age, sex, pathogenesis of heart disease, serum creatinine and total bilirubin at transplant, presence of diabetes mellitus, most recent panel of reactive antibodies, pulmonary artery systolic pressure, use of inotropic support, medical condition at time of transplantation, waiting list priority status (1A, 1B, or 2), requirement for mechanical ventilation, and allograft cold ischemic time.
Statistical Analysis
Allograft Use
Time trends of rates of allograft use and prevalence of donor risk factors were calculated. Odds ratios were calculated to study associations between these 11 donor risk factors and cardiac allograft nonuse.
An allograft use score was then derived using all available donor predictors in our research database. These included demographic factors (race, weight, and height), clinical factors (use of cardiopulmonary resuscitation, history of coronary artery disease, diabetes mellitus, and hypertension), laboratory values (hemoglobin and creatinine), hemodynamics (heart rate, central venous pressure, and systolic and diastolic blood pressure), use of vasoactive medications (phenylephrine, norepinephrine, and esmolol), and hormonal therapy (corticosteroids and levothyroxine). To combine these donor characteristics into a single-use score, the predicted probability of allograft use was calculated. To predict allograft use, we used the Random Forest algorithm31—a robust extension of decision trees that has been used extensively in other biomedical studies. Two notable features of the Random Forest model are that it can handle missing values and it validates internally, alleviating the need for cross-validation. To compare the utility of these scores in predicting actual allograft use, receiver operating characteristic curves and the associated c-statistics were calculated.
Recipient Outcomes
Logistic regression models were used to examine associations between donor predictors and recipient outcomes. We first examined the 11 donor risk factors identified a priori, and then assessed the relationship between the total number of donor risk factors present and recipient outcomes. To examine the utilization choice, hearts were dichotomized based on the predicted probability of use (<50% versus >50%) from the Random Forest scores and were then compared. Finally, Random Forest models were then generated for the 3 recipient outcomes of interest, using all available donor predictors. All reported P values are 2-tailed. Analyses were performed using R version 2.15 with the Random Forest package.
Results
From 2001 to 2008, 1872 potential organ donors were managed by CTDN. Allografts from 808 (43%) of these donors were accepted for heart transplantation. Demographic characteristics of the donors whose hearts were and were not transplanted are presented in Table 1. Donors whose hearts were not accepted for transplantation were more likely to be older, female, and had CVA/stroke as a cause of death. These donors had a higher incidence of smoking, hypertension, diabetes mellitus, and coronary artery disease, and were more likely to have a positive cardiac troponin assay. These donors had a slightly higher dopamine requirement during the donor management period and were less likely to have received corticosteroids and thyroxine supplementation. Finally, nonused grafts were less likely to have had an echocardiogram, had a lower mean left-ventricular ejection fraction and a higher incidence of left-ventricular regional wall motion abnormalities and left-ventricular hypertrophy. These results confirmed our initial assumption that the donor risk factors selected a priori for analytic purposes were significant predictors of cardiac allograft nonuse, except for high-donor dopamine requirement (>10 μg/kg per minute) during the donor management period and donor history of cocaine or methamphetamine use.
Donor Characteristics Stratified by Cardiac Allograft Acceptance for Transplantation
During the study period, cardiac allograft use decreased by an average of 4.2% per year (95% confidence interval, 1.9%–6.4%), from a high of 56% use in 2002 to a low of 37% use in 2007. This decrease in use was independent of changes in donor risk factors with a 4.3% (2.3%–6.3%) annual decrease in allograft acceptance for transplantation after adjusting for other covariates. Of the 11 preidentified donor risk factors, only the incidence of diabetes mellitus increased during this time period (P<0.0054). In fact, the following donor risk factors decreased in incidence: donor age >50 years (P<0.03), death attributable to CVA (P<0.0004), and high-dopamine requirement (P<1×10−8).
We then studied associations between the 11 donor risk factors selected a priori and cardiac allograft acceptance for transplantation. Figure 1 demonstrates the distribution of these donor risk factors in grafts that were and were not accepted. All of the risk factors were highly associated with allograft use (P<0.001), except for the amount of dopamine administered during donor management and donor history of cocaine or methamphetamine use. The total number of donor risk factors was also strongly associated with allograft use (P<1×10−8) with accepted grafts having a median of 2 (interquartile range, 1–3) risk factors and discarded grafts having a median of 3 (interquartile range, 2–5) risk factors.
Odds ratios for cardiac allograft acceptance for transplantation, by donor risk factors.
Finally, all available donor variables were used to predict allograft use using the Random Forest algorithm. A total of 59 covariates were used and 2000 trees were grown. The most important predictors in the Random Forest models were donor age, cause of death, left-ventricular ejection fraction, and history of hypertension. The Random Forest model demonstrated excellent predictive ability with an overall c-statistic of 0.86.
Donor Predictors of Recipient Outcomes
Recipient outcomes data were available for 806 of the 808 cardiac allografts accepted for transplantation. Of the recipients, 29 were excluded from analysis as they had received multiorgan transplants. Table 2 demonstrates recipient characteristics at the time of transplantation, stratified by those who received an allograft with >50% probability of acceptance, based on the Random Forest predictions. Not surprisingly, 64% of the allografts were those that had a >50% probability of use. The recipients who received the less desirable grafts were more likely to be female, had better clinical status, and were less likely to be on life support; no other significant clinical differences were identified.
Heart Transplant Recipient Characteristics Stratified by Probability of Allograft Use
The primary outcomes examined were time to hospital discharge and recipient 30-day and 1-year survival. Overall, the outcomes were generally positive and consistent with national heart transplant statistics2 with 75% of recipients discharged within 21 days, and only 3.9% and 11.4% dying within 30 days and 1 year, respectively. Kaplan–Meier curves stratified by the probability of graft acceptance (based on the Random Forest predictions) are presented in Figure 2. Overall, recipients of allografts that were more likely to be accepted were discharged earlier from the hospital (log rank P<0.035). However, there was minimal difference in overall survival (P=0.067).
Effect of allograft predicted probability of use (< or >50%) on (A) time to hospital discharge after heart transplantation, and (B) 1-year graft survival.
Associations between the 11 donor risk factors were identified a priori, and the recipient outcomes of interest were then examined, and the odds ratios are presented in Table 3. Whereas most donor risk factors seemed associated with prolonged hospital stay post-transplant, only CVA/stroke as the donor cause of death was marginally significantly associated (odds ratio, 1.41 [95% confidence interval, 1.00–2.00]) with prolonged recipient hospitalization. Similarly, whereas many donor risk factors seemed to increase recipient risk of death, left-ventricular hypertrophy and history of diabetes mellitus were the only two donor characteristics significantly associated with recipient 30-day and 1-year mortality. After adjustment for recipient characteristics, diabetes mellitus remained the only donor predictor of recipient mortality (odds ratio, 3.58 [95% confidence interval, 1.18–10.84] for 1-year mortality). A notable finding was that the presence of multiple donor risk factors did not increase the occurrence of adverse recipient outcomes (Figure 3).
Unadjusted Odds Ratios for Associations Between Donor Risk Factors and Recipient Post-Transplant Outcomes
Associations between number of donor risk factors and heart transplant recipient outcomes (discharge within 21 days post-transplant, 30-day graft survival, and 1-year graft survival).
Finally, we used all of the available donor characteristics to predict the 3 outcomes of interest using Random Forest models. The receiver operating characteristic curves, along with the receiver operating characteristic curve for allograft use, are shown in Figure 4. Whereas the donor characteristics were highly predictive of graft use, those same characteristics were not predictive of recipient outcomes.
Receiver operating characteristic curves, based on Random Forest models, for donor prediction of (1) cardiac allograft use for transplantation, (2) discharge within 21 days post-transplant, (3) 30-day graft survival, (4) 1-year graft survival.
Discussion
National transplant data collected by the Association of Organ Procurement Organizations (www.aopo.org, accessed June 21, 2012) reveal a cardiac allograft use rate (number of hearts accepted for transplantation/total number of donors) of 28.2% to 30.1% from 2009 to 2011. Many reasons exist for discarding donor hearts, including older donor age, small size, comorbidities, logistical issues, left-ventricular hypertrophy, and donor left-ventricular dysfunction. Unfortunately, the current criteria for acceptance of donor hearts are poorly standardized and are often based on retrospective single-center studies and anecdotal experience. Indeed, large registry analyses have shown that very few donor characteristics have significant impact on recipient outcomes,32,33 suggesting that recipient factors figure more prominently toward the risk of death after heart transplantation.
Our aim in conducting this study was to closely examine current practices with respect to cardiac allograft acceptance in a contemporary cohort of potential organ donors. During the 8-year time period examined, we found that use rates decreased by an average of 4.2% per year, from a high of 56% in 2002 to a low of 37% in 2007. This decline in allograft acceptance occurred despite a decreased incidence of donor risk factors for nonuse, such as death attributable to CVA and older age. Although use rates in our donation service area are higher than the national average, the significant decline over time suggests increasing avoidance of risk—a practice that could potentially lead to longer waiting list times and increasing waiting list mortality. One explanation for the trend toward avoidance of high-risk donors could be related to advances in mechanical circulatory support (particularly ventricular assist devices) as a bridge to transplantation. Improvements in ventricular assist device design and a reduction in size of the devices has allowed for predictable clinical stabilization of a growing population of end-stage heart failure patients,1,34 even those who are critically ill. The concern with adopting this approach indiscriminately is that bridging devices are expensive and can double the cost of an already expensive procedure. Ventricular assist device implantation also exposes patients to the risk of additional surgical procedures, infections, and development of human leukocyte antigen antibodies,35 which may contribute to poorer outcomes after heart transplantation.36,37
We first examined the associations between the 11 donor risk factors identified a priori, based on our review of the literature, and allograft use. Almost all of these risk factors, except for high-dopamine requirement and cocaine/methamphetamine use, significantly predicted nonuse. The total number of donor risk factors was also strongly associated with nonuse.
Recognizing the fact that other donor characteristics may influence allograft acceptance decisions, we used all available donor covariate data to construct a Random Forest algorithm. Random Forest models confer several advantages: they can handle a large number of input variables, they can give an estimate of which variables are important in the classification, and they can produce a highly accurate classifier. In addition, they internally validate, thereby obviating the need for cross-validation. This model had excellent discrimination, with a c-statistic of 0.86.
After identifying donor predictors of allograft nonuse, we examined the associations between these predictors and recipient outcomes. Only CVA as the donor cause of death marginally predicted prolonged recipient post-transplant hospitalization, and diabetes mellitus was the only donor predictor of increased recipient mortality. These findings concur with previous studies demonstrating the relatively small contribution of donor characteristics to post-transplant adverse events.4,38,39
One criticism of studies examining the influence of donor characteristics on recipient outcomes is that of selection bias. Allografts with many undesirable features are rarely accepted for transplantation, making it difficult to determine whether the recipient outcomes would have been acceptable. On the other hand, grafts with 1 or 2 unfavorable characteristics (such as reduced left-ventricular systolic function) may be accepted if the graft is otherwise favorable (eg, from a young and healthy male donor). Thus, it is difficult, if not impossible, to determine the relative contributions of unfavorable characteristics to recipient events. In other words, the grafts with undesirable characteristics are very carefully chosen to minimize the risk of adverse events. This phenomenon, combined with the relatively low mortality within 30-days to 1-year post-transplant, may paradoxically make it seem as though unfavorable characteristics are associated with improved recipient outcomes.
We used the excellent discriminative ability of our Random Forest model to predict which allografts in our cohort would and would not have been accepted for transplantation, based on a threshold of >50% predicted probability of use. Based on this model, 36% of the allografts accepted for transplantation had a <50% probability of use, based on donor characteristics. We once again examined recipient post-transplant outcomes, this time based on the predicted probability of allograft acceptance, and were unable to identify any significant differences in recipient survival. Thus, receipt of an allograft that was unlikely to be used for transplantation, based on combinations of unfavorable donor characteristics, did not result in adverse recipient events. Finally, donor characteristics in aggregate did not prove to be a reliable predictor of recipient outcomes.
This study has significant strengths and limitations. First and foremost, this represents the largest existing research database of detailed, rigorously adjudicated clinical data on >1800 potential organ donors managed in a standardized fashion in the current era. The recipient outcomes selected for analysis are robust, because all heart transplant centers in the United States are required to report metrics such as length of post-transplant hospitalization and recipient survival to United Network for Organ Sharing. The primary limitation is the observational nature of the data. We are reluctant to make any causal statements about the relationship between donor characteristics and recipient outcomes. There are likely selection effects that may explain the relatively positive outcomes among recipients of the less desirable allografts. Also, reasons for allograft discard were not documented in the donor medical record. Thus, it is possible that reasons other than donor characteristics (lack of a suitable recipient, time considerations, donor family preference) could have accounted for some cases of nonuse. Nonetheless, we were still able to identify a number of strong donor predictors of allograft nonuse. Several important predictors were based on echocardiography: reduced left-ventricular systolic function, left-ventricular regional wall motion abnormalities, and left-ventricular hypertrophy. However, donor echocardiograms were interpreted at local hospitals and were not centrally reviewed; we therefore cannot verify the accuracy of echocardiogram interpretation and measurements. Another limitation lies in the fact that relatively few recipients died within the first year post-transplant; this study may therefore have been underpowered to detect subtle influences of donor characteristics on post-transplant outcomes. Finally, this study was limited to donors managed by CTDN and may not be generalizable to heart transplant procedures performed throughout the United States or worldwide.
In conclusion, we have demonstrated the persistent low use of available cardiac allografts for transplantation in the face of a national organ-donor shortage and a growing number of patients with end-stage heart disease. We identified predictors of allograft nonuse and demonstrated that the anticipated relationship between these donor predictors and adverse recipient outcomes was not seen in our heart transplant cohort. These findings support the following statement recently put forth by the National Heart, Lung, and Blood Institute regarding the next decade of heart transplantation research: “Without clear evidence about the outcomes associated with different donor characteristics informing the donor selection process, it is probable that many potentially useful organs are currently being discarded. Because an important rate limiting factor in [heart transplantation] is the number of available donor organs, studies that define how to optimize donor use and develop biomarkers to define organ utility might increase the donor pool by providing evidence that would support the use of those organs deemed to be less than perfect.”40 The field of heart transplantation would therefore benefit greatly from prospective, multicenter trials studying the effects of liberalizing allograft acceptance criteria.
Acknowledgments
We thank the California Transplant Donor Network staff for access to the donor data required for this study.
Sources of Funding
This work was supported in part by the Health Resources and Services Administration contract 234-2005-370011C, by the American Heart Association (0865249F, Dr Khush), and by the National Heart, Lung, and Blood Institute (K23HL091143, Dr Khush).
Disclosures
The content of this study is the responsibility of the authors alone and does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the US Government.
- Received August 29, 2012.
- Accepted January 28, 2013.
- © 2013 American Heart Association, Inc.
References
Clinical Perspective
Currently, only 30% to 40% of available donor hearts are accepted for transplantation in the United States, despite a national organ-donor shortage and a growing population of patients with end-stage heart disease. Using a well-characterized cohort of 1872 potential organ donors managed by the California Transplant Donor Network from 2001 to 2008, we sought to identify donor characteristics that predict cardiac allograft nonuse for transplantation. We then determined whether these donor predictors are in fact associated with adverse recipient outcomes such as prolonged hospitalization or reduced survival after transplantation when such hearts are accepted for use. In this cohort, predictors of allograft nonuse included donor age >50 years, female sex, death attributable to cerebrovascular accident, hypertension, diabetes mellitus, a positive troponin assay, left-ventricular dysfunction, and left-ventricular hypertrophy. Of these predictors, only use of a heart from a donor with diabetes mellitus was predictive of increased recipient mortality within the first year post-transplant. We conclude that whereas there are many donor predictors of allograft nonuse in the current era, hearts from donors with these predictors appear to have little effect on recipient clinical outcomes after transplant. Our results suggest that careful liberalization of current cardiac allograft acceptance practices may be warranted.
This Issue
Jump to
Article Tools
Share this Article
- Donor Predictors of Allograft Use and Recipient Outcomes After Heart TransplantationClinical Perspective
