Geographic variation occurs in a variety of health outcomes. Regional influences on outcomes before and after listing for pediatric heart transplantation have not been assessed. Review of the UNOS dataset identified 5398 pediatric (≤18 years) patients listed for heart transplantation 2000–2011. Patients were stratified based on the region of listing. Regional-level variables were correlated with individual risk-adjusted outcomes. Mean time spent on the waitlist varied from 91.0 ± 163 days (Region 6 [R6]) to 248.1 ± 493 days (R4, p < 0.0001). Regions with more transplant centers (p < 0.0001) and fewer transplants (p = 0.0015) had higher waitlist mortality. Risk-adjusted individual waitlist mortality varied from 6.9% (R1, CI 6.2–7.8) to 19.2% (R5, CI 18.0–20.6). Waitlist mortality was higher for individuals awaiting transplant in regions with more listings per center (OR 1.04, CI 1.01–1.08) and lower in regions with more donors per center (OR 0.95, CI 0.90–0.99 per donor). Posttransplant risk-adjusted survival varied across regions (R4: 5.4%, CI 4.2–7.4; R7: 18.0%, CI 12.4–32.5), but regional variables were not correlated with outcomes. Outcomes among children undergoing heart transplantation vary by region. Factors leading to increased competition for donor allografts are associated with poorer waitlist survival. Equitable allocation of cardiac allografts requires further investigation of these findings.
Donation Service Area
extracorporeal membrane oxygenation
intra-aortic balloon pump
organ procurement organization
pulmonary capillary wedge pressure
pulmonary vascular resistance
Rural-Urban Commuting Area
United Network for Organ Sharing
zip code tabulation area
Geographic variation has been documented in a variety of health treatments and outcomes [1-4]. In noncardiac solid organ transplantation, donor availability and quality have varied by region [5, 6]. In heart transplantation, the association between longer donor organ ischemic times and poorer posttransplant outcomes suggests that the distance between donor and recipient (which is highly correlated with ischemic time)  should impact allocation [8, 9]. However, nonclinical factors may also play a role as the result of administrative differences between regions, the allocation of organs based on the arbitrary boundaries of local Donation Service Areas (DSAs) and the density of transplant centers and DSAs within each region, among others.
While the 2009 changes to pediatric heart allocation attempted to minimize the direct impact of geography on allocation, they did not eliminate it. Further, these other regional factors, including DSA size and number, have been neither investigated nor addressed. An understanding of how region affects transplant outcomes is necessary in order to mitigate the geographic determinants of transplant outcomes.
UNOS provided de-identified patient-level, center-specific data from the Thoracic Registry (data source #9/7/2011, data through 6/30/2011) of the UNOS Standard Transplant and Research Dataset. The center and region of each transplant were provided within the dataset. Geographic data including land area, population and measures of socioeconomic status were obtained from the 2000 U.S. Census American FactFinder 1. Rural/urban categorization was based on zip code of residence and the Rural-Urban Commuting Area (RUCA) Codes available online .2 DSA-specific and regional donor rates and other statistics were obtained from the OPTN with data as of March 2011 (http://optn.transplant.gov/data). Use of these data is consistent with the regulations of our Institutional Review Board.
Between 2000 and 2011 (end harvest date: June 30, 2011), 5398 pediatric patients (≤18 years old) were listed for primary cardiac transplantation; of these patients 3519 underwent transplantation. Socioeconomic status was calculated based on zip code of residence and socioeconomic data from the 2000 U.S. Census based on zip code tabulation area (ZCTA). The is based on previously described methodology [12-15], and full details are available in Supplementary Methods.
Statistical analysis was performed using SAS 9.2 for AIX (SAS Institute, Cary, NC). Regional level and individual level outcomes were evaluated.
Regional level outcomes
Unadjusted outcomes were calculated for each region. The association between regional variables (see Supplementary Table 1) and regional outcomes (regional mean time spent on the waitlist, regional waitlist mortality [mortality prior to removal or transplant], regional likelihood of transplant, and regional posttransplant mortality [mortality within 30-days or prior to hospital discharge]) was assessed by Pearson correlation (univariable) and linear regression (multivariable, stepwise, p < 0.2).
Definition of high risk donors and recipients
High risk donors and recipients were defined based on previously published criteria [16, 17], and had at least one of these criteria: age greater than 50 years, cause of donor death = stroke, presence of diabetes mellitus, presence of hepatitis C seropositivity, donor:recipient weight ration <0.8. High risk recipients had any one of the following: elevated pulmonary vascular resistance index >6 woods units m2, creatinine clearance <40 mL/min, hepatitis C seropositivity, panel reactive antibodies >40%, or less than 1 year of age.
Primary waitlist outcomes were: likelihood of receiving a transplant, and mortality while waiting.3 Primary posttransplant outcomes were: postoperative mortality (30 days or prior to hospital discharge whichever occurs last), 1-year posttransplant mortality, and long-term posttransplant survival.
Continuous variables were compared using the Student's t-test (two-tailed) and one-way analysis of variance. Categorical variables were compared by χ2-test. Some continuous variables, including body surface area (BSA), weight, age, hemodynamic variables (cardiac index, pulmonary capillary wedge pressure [PCWP], pulmonary vascular resistance [PVR]), laboratory variables (creatinine, albumin) and transplant procedure variables (ischemic time, donor transplant volume) were used as both continuous variables and categorized into subgroups, subgroups were defined based on previously published stratification schemes where available [18, 19]; the most predictive method was used.
Kaplan–Meier analysis (log-rank test with Sidak correction for multiple comparisons, p < 0.05) and Cox regression (backward selection, p < 0.2) were used for time-to-event analysis; the assumption of hazards proportionality was tested by introducing terms of interaction with log(time) in the Cox model. Center-clustered data were fitted by marginal model using the robust sandwich estimate (COVSANDWICH option in PROC PHREG, SAS 9.2 for AIX). Competing risks analysis of outcomes on the waitlist was used in a manner previously reported ; outcomes considered were: alive and waiting, transplanted, removed, died.
The authors had full access to the data and take full responsibility for its integrity. All authors have read and agree to the manuscript as written.
Handling of missing values
For regression modeling, missing variables were imputed using the technique of multiple imputation (10 imputations, Markov Chain Monte Carlo procedure). Multiple imputation is a technique in which missing values are replaced by m > 1 simulated versions; the simulated datasets are then analyzed and finally, the results of the m analyses combined to produce a single inferential result taking into account the uncertainty in the simulated values. This technique is more accurate than the use of complete-case analysis or mean value imputation . The percentage of missing variables and a detailed explanation of the handling of missing data within the UNOS dataset has been previously reported .
Outcomes were modeled using hierarchical logistic regression (PROC GLIMMIX, stepwise, p < 0.2) performed on the multiple imputation datasets and including a random transplant center effect. Predicted mortality for each patient was calculated as the mean prediction across all (n = 10) imputations. Predicted mortality for the each region was calculated using linear models (PROC GLM) including 95% confidence limits. Risk-adjusted outcomes for the region were calculated:
The 11 transplant regions (Figure 1) differ widely in geography (area, population, population density and the relative contribution of urban and rural areas), density of transplant services (number of transplant centers, number of DSAs) and the number of patients listed and transplanted (Table 1).
|Region||Population||Area (mi2)||Population density (per mi2)||Rural population||% Rural||Centers||Population per center||DSAs||Population per DSA||Listings||Transplants||Pediatric donor recoveries|
|1||7 897 236||54 969||138||2 156 781||28||6||1 316 206||2||3 948 618||185||127||354|
|2||30 620 476||88 100||331||5 165 027||18||15||2 041 365||4||7 655 119||564||326||1139|
|3||47 411 075||308 539||137||10 172 735||24||13||3 647 006||9||5 267 897||857||565||1951|
|4||28 896 912||330 464||74||4 843 835||20||13||2 222 839||4||7 224 228||407||261||1431|
|5||52 529 889||589 342||79||3 471 151||8||16||3 283 118||7||7 504 270||884||597||1898|
|6||13 822 842||962 792||13||2 854 981||24||5||2 764 568||4||3 455 711||127||87||454|
|7||25 308 314||334 364||72||5 284 373||22||12||2 109 026||4||6 327 079||415||267||1048|
|8||19 307 563||484 261||37||4 978 837||28||9||2 145 285||6||3 217 927||618||431||1063|
|9||25 925 731||55 054||460||2 920 977||12||6||4 320 955||4||6 481 433||371||257||535|
|10||27 903 946||133 619||205||6 869 128||25||11||2 536 722||6||4 650 658||500||322||1138|
|11||32 847 343||199 360||145||10 559 434||37||14||2 346 239||8||4 105 918||470||279||1401|
|Overall||312 471 327||3 540 863||88||59 277 259||19||120||2 603 928||58||5 387 437||5398||3519||12 412|
|Mean||28 406 484||321 897||154||5 388 842||22||11||2 612 121||5||5 439 896||491||320||1128|
Regional differences in outcomes
There were significant regional differences in baseline demographic and clinical characteristics at listing (Table 2). Unadjusted waitlist mortality was highest in regions 2 (16.8%) and 11 (16.8%) and lowest in regions 1 (7.0%) and 6 (7.1%; p = 0.009). The likelihood of being transplanted varied from 57.8% (region 2) to 69.7% (region 8; Table 3, p = 0.0002). Regional variation in time to transplant (based on competing risks analysis) is shown in Figure 2 (see Supplementary Figures 1 and 2 for illustration of other risks: death and removal). Survival following listing differed by region (Figure 3, p = 0.0049).
|Status at listing|
|Region||Days on waitlist (mean ± SD)||Unadj waitlist Mortality||Risk-adj waitlist mortality (95% CI)||Unadj % transplanted||Risk adj transplant likelihood (95% CI)||Unadj postTxpl mortality||Risk adj postTxpl mortality (95% CI)||Unadj 1 year postTxpl mortality|
|1||92 ± 164||13/185 (7.0%)||6.9% (6.2–7.8)1||127/185 (68.6%)||76.7% (72.0–82.1)||10/127 (7.8%)||10.5% (8.3–14.3)1||12/127 (9.4%)|
|2||206 ± 526||95/564 (16.8%)||18.7% (17.3–20.1)2||326/564 (57.8%)||62.1% (60.0–64.4)||30/320 (9.3%)||12.7% (9.9–18.0)1||36/320 (11.2%)|
|3||152 ± 387||116/857 (13.5%)||14.4% (13.6–15.4)||565/857 (65.9%)||72.0% (69.9–74.2)||36/553 (6.5%)||9.5% (7.9–11.9)1||66/553 (11.9%)|
|4||248 ± 493||53/407 (13.0%)||13.8% (12.5–15.3)||261/407 (64.1%)||67.3% (64.6–70.2)||9/254 (3.5%)||5.4% (4.2–7.4)1||26/254 (10.2%)|
|5||159 ± 405||138/884 (16.4%)||19.2% (18.0–20.6)2||597/884 (67.5%)||69.1% (67.2–71.0)||35/588 (5.9%)||9.3% (7.5–12.1)1||48/588 (8.1%)|
|6||91 ± 109||9/127 (7.1%)||8.7% (7.3–10.4)1||87/127 (68.5%)||77.4% (71.7–84.1)||4/85 (4.7%)||7.4% (4.8–16.4)1||7/85 (8.2%)|
|7||198 ± 400||51/415 (12.3%)||14.1% (12.9–15.5)||267/415 (64.3%)||68.3% (65.6–71.2)||29/259 (11.%)||18.0% (12.4–32.5)||35/259 (13.5%)|
|8||182 ± 363||83/618 (13.3%)||15.9% (14.8–17.3)2||431/618 (69.7%)||74.4% (72.0–77.0)||28/423 (6.6%)||11.9% (9.0–17.7)1||56/423 (13.2%)|
|9||86 ± 283||50/371 (13.5%)||16.8% (15.1–18.6)2||257/371 (69.2%)||69.8% (67.1–72.8)||19/256 (7.4%)||14.1% (10.0–23.9)||28/256 (10.9%)|
|10||128 ± 265||66/500 (9.3%)||15.7% (14.4–17.1)2||322/500 (64.4%)||69.1% (66.9–71.9)||25/319 (7.8%)||15.4% (12.3–20.6)||39/319 (12.2%)|
|11||144 ± 313||79/470 (16.8%)||18.8% (17.4–20.7)2||279/470 (59.3%)||66.4% (63.8–69.3)||19/276 (6.8%)||15.7% (10.2–34.7)||28/276 (10.1%)|
|p < 0.0001||p = 0.009||p = 0.0002||p = 0.08||p = 0.4|
|Overall||752/5398 (13.9%)||3519/5398 (65.2%)||244/3460 (7.1%)||381/3460 (11.0%)|
Regional differences in recipient risk among listed patients
There was a lower proportion of high risk listings within regions with an increased population per DSA (r = −0.71, p = −0.01), more listings per DSA (r = −0.64, p = 0.03), or more transplants per DSA (r = −0.64, p = 0.03), and among regions with smaller geographic area (r = 0.60, p = 0.049). In multivariable analysis the following variables were associated with a lower proportion of high risk recipients (model r2: 079, p = 0.008): decreased regional land area (β: 10.74/million, r2: 0.20, p = 0.02), increased population per DSA (β: −1.23/million, r2: 0.50, p = 0.12), decreased rural population percentage (β: 0.42, r2: 0.10, p = 0.10).
Regional factors influencing wait list times and likelihood of transplantation
In univariate analysis (Supplementary Table 2), the demography and geography of each region (total population, land area, population density, percent rural population) was not correlated with time spent on the wait list or the likelihood of transplantation. Regions with more transplant centers had longer waitlist times (r = 0.69, p = −0.02), as did those with an increased ratio of donors to DSAs (r = 0.87, p = 0.0005). Regions with more status 1A listings had shorter wait times (r = −0.56, p = 0.07). No other regional level variables were correlated with mean regional waitlist times in univariate analysis.
In a multivariate model (Supplementary Table 2), the following regional characteristics were correlated with longer mean regional wait times (model r2: 0.96, p = 0.002): increased number of donors per DSA (β: 3.17, r2: 0.76, p = 0.0006), increased number of listings per center (β: 0.59, r2: 0.03, p = 0.13), decreased population per DSA (β: −13.6/million, r2: 0.05, p = 0.03), increased percentage of regional listings at status 1A (β: 8.16, r2: 0.10, p = 0.006) and an increased percentage of listings at status 1B (β: 2.44, r2: 0.02, p = 0.19).
Regional wait list mortality
In univariate analysis, the following variables were correlated with a higher regional waitlist mortality: longer mean regional waiting time (r = 0.75, p < 0.008), the number of centers (r = 0.81, p = 0.003), higher population (r = 0.71, p = 0.01), total number of listings (r = 0.70, p = 0.02) and the total number of pediatric donors available within the region (r = 0.66, p = 0.03).
A multivariate model of higher regional mortality on the waitlist (model r2: 0.96, p = 0.0002) included the following variables: an increasing number of transplant centers within the region (β: 1.67, r2: 0.65, p < 0.0001), a decreasing number of donors per DSA within the region (β: −0.03, r2: 0.03, p = 0.05), an increasing number of listings per center (β: 0.28, r2: 0.09, p = 0.0007) and a decreasing number of transplants performed within the region (β: −0.03, r2: 0.18, p = 0.0015).
Risk-adjusted individual outcomes on the wait list
Risk-adjusted mortality on the waitlist varied widely by region; from 6.9% (region 1) to 19.2% (region 5; Table 3). In a multivariate model of individual mortality on the waitlist, regional variables associated with higher likelihood of individual death on the waiting list included: increased regional land area, fewer donors per transplant center, more listings per transplant center, a smaller difference between the number of donors and the number of transplant recipients (Table 4 and Supplementary Table 3).
|Odds ratio||95% confidence interval||p-Value|
|Age at listing|
|2 to 5 years||1.00||ref|
|13 to 18 years||0.77||0.60–0.98||0.03|
|BMI > 35||1.73||1.07–2.82||0.03|
|IABP at listing||2.08||0.79–5.47||0.14|
|VAD at listing||0.65||0.42–1.01||0.05|
|ECMO at listing||2.12||1.67–2.69||0.00|
|Albumin < 3.5||1.19||0.92–1.53||0.18|
|Creatinine clearance < 40 cc/min||0.77||0.61–0.98||0.03|
|Year of listing (per year)||0.9216||0.8967–0.9472||<.0001|
|Number of listings at center (per listing)||0.9783||0.9644–0.9923||0.0025|
|Region area (per million sq mi)||1.003||0.9940–1.0116||NS|
|Donors per center (per donor)||0.947||0.900–0.997||0.0382|
|Listings per center (per listing)||1.043||1.01–1.081||0.02|
|Number of organs imported (per organ)||0.996||0.992–0.999||0.02|
Cox proportional hazards regression of mortality following listing included risk factors related to individual patient characteristics (neonatal age, mechanical ventilation, IABP, ECMO, hospitalization status, year of listing, etiology, ABO type, socioeconomic status) as well center-specific measures: pediatric heart listings in the most recent year (hazard ratio [per listed patient]: 0.98, 95% CI 0.98–0.99), and regional variables: area (per 1000 mi2 1.02, 1.006–1.046), population density (per person/mi2 1.003, 1.000–1.005), regional ratio of donor recoveries to listings (0.51, 0.19–1.35), percentage of transplants with local donors (1.05, 1.02–1.10), percentage of transplants from Zone A donors (1.05, 1.01–1.09; see Supplementary Table 4 for full model).
Regional differences in posttransplant outcomes
Clinical condition at transplantation by transplant region is given in Table 5. Unadjusted postoperative mortality (p = 0.08) and mortality at 1 year following transplant (p = 0.35) were not significantly affected by the region of transplant (Table 3). Unadjusted long-term survival after transplant differed by region (Figure 4, p = 0.0068). Risk-adjusted posttransplant mortality varied widely by region; from 5.4% (region 4) to 18.0% (region 7; Table 3).
|Age at transplant||<0.0001|
|Etiology of heart failure||<0.0001|
|Status at transplant||<0.0001|
|Creat clearance <40 mL/min||5.9%||7.72||9.19||9.44||7.04||2.60||8.70||11.08||5.22||9.97||9.34||NS|
|PVR >3 WU||25.0%||28.97||21.50||30.49||28.78||25.81||25.42||38.54||33.04||37.04||33.04||0.05|
|Donor age <13 years||55.9%||65.42||66.55||67.32||59.52||67.06||68.08||71.66||62.50||68.03||63.54|
|Donor age ≥50 years||0.00||0.00||0.36||0.00||0.34||1.18||0.38||0.00||0.00||0.31||0.72|
|ABO blood type||NS|
|ABO type: A||30.7%||31.46||28.78||27.24||28.57||41.18||33.85||33.26||29.69||32.92||28.52|
|ABO type: O||55.9%||54.21||61.33||66.93||62.93||47.06||55.00||59.02||57.81||58.31||66.43|
|ABO type: AB||3.2%||2.49||1.62||0.78||0.51||3.53||2.69||0.47||2.34||1.25||0.72|
|ABO type: B||10.2%||11.84||8.27||5.06||7.99||8.24||8.46||7.26||10.16||7.52||4.33|
|Cause of death: CVA||10.2%||8.49||9.71||10.89||8.84||8.24||7.31||6.56||12.60||9.75||6.86||NS|
|Infusions at cross-clamp|
|Donor:recipient wgt ratio <0.7||0.8%||3.16||2.54||0.80||0.68||0.00||1.59||0.95||1.67||1.27||2.22||NS|
|Mean (h, ±SD)||3.5 ± 1.1||3.4 ± 1.0||3.2 ± 1.1||3.5 ± 1.2||4.1 ± 1.6||4.4 ± 1.6||3.5 ± 1.1||3.7 ± 1.2||3.4 ± 1.2||3.4 ± 0.9||3.6 ± 1.1||<0.0001|
Association between regional characteristics and posttransplant survival
In both univariate and multivariate analysis there was no association between most regional factors and regional posttransplant mortality, including geographic factors, regional administrative details (number of DSAs, number of transplant centers), regional donation and transplant volume. The mortality rate was lower with.in regions with a higher percentage of local donor/recipient matches (r = −0.61, p = 0.04). Regional mortality at 1 year was similarly unaffected by most regional variables. However, regions with a higher percentage of allografts received from donors in Zone A had higher 1-year mortality (r = 0.68, p = 0.02), while those with a higher percent of local donors had lower 1-year mortality (r = −0.64, p = 0.03).
Variation in recipient and donor selection across regions
There was increased use of high risk donors within regions with more DSAs per person (r = 0.62, p = 0.04), and those with a lower proportion of rural population (r = −0.70, p = 0.016) and higher population density (r = −0.63, p = 0.036). Regions with a higher percentage of high risk recipients received more donor organs from zone A (r = 0.84, p = 0.001), and fewer from zone B (r = −0.56, p = 0.08), zone C (r = −0.71, p = 0.02) and zone D (r = −0.70, p = 0.02).
In a multivariate model the following characteristics were correlated with an increased use of high risk donors within a region (model r2: 0.95, p = 0.003): smaller percent of population living in rural area (β: −0.34, r2: 0.49, p = 0.003), lower land area (β: −4.27/million, r2: 0.20, p < 0.005), increased donors per center (β: 0.13, r2: 0.17, p < 0.006), decreased number of transplant (β: −0.71/hundred, r2: 0.05, p < 0.04), increased population per DSA (β: 0.34/million, r2: 0.03, 0.16).
Transplantation of high risk recipients was more common among rural populations (r = 0.67, p = 0.02), and regions with increased population per DSA (r = −0.66, p = 0.03). In a multivariable model (r2: 0.77, p = 0.013), the following characteristics were associated with increased proportion of high-risk transplant recipients: increased proportion of rural population (β: 0.53, r2: 0.45, p < 0.013), decreased population per transplant center (β: −1.69/million, r2: 0.10, p = 0.12), and increased regional land area (β: 7.73/million mi2, r2: 0.21, p = 0.03),
Risk-adjusted Individual posttransplant outcomes
Individual postoperative mortality and mortality at 1 year following transplantation were unaffected by regional variables in multivariable analysis.
Initial impact of 2009 allocation changes on waitlist mortality
Among patients listed after 2010, unadjusted survival on the waitlist continues to vary significantly by region (Figure 5). Regional factors continued to be important predictors of waitlist mortality in multivariable analysis: number of pediatric donor hearts per transplant center (OR: 0.95, 95% CI 0.91–0.99 per donor, p = 0.047), number of listings per center (1.04 per listing, 1.00–1.07, p = 0.029) and the number of donor organs brought into the region from other donors in other regions (0.996 per organ, 0.992–0.999, p = 0.02; see Supplementary Table 5 for full hierarchical model).
Regional variation in a variety of health outcomes—including nonthoracic organ transplantation—has been previously described [1-4, 23]. Our data confirm that geographic disparities exist in outcomes before and after heart transplantation in children. Given the extraordinary variation in the geographic and demographic characteristics of the UNOS regions, this is not surprising. However, the degree to which the region of listing is correlated with the risk-adjusted likelihood of surviving the waitlist and subsequent transplant is striking. Baseline characteristics of the listed and transplanted patients vary widely across regions, but even after risk adjustment, wide regional disparities exist in survival both before and after pediatric heart transplantation.
Underlying geography may play a role with variation in topography, culture, and the urban/rural composition of each region affecting multiple factors likely to result in differential survival. Although population density affects donation rates, regional health-related factors resulting in regional variation in age-specific death rates, health status at the time of death, and clinical practice surrounding trauma care and declaration of brain death may lead to differences in the frequency of suitable donors [5, 6]. Variable and somewhat arbitrary DSA boundaries may exacerbate these differences .
Regional factors with the potential to influence outcome can be broadly divided into several groups: geographic factors, UNOS regional practice variation, DSA/Organ Procurement Organization (OPO) factors, center-specific factors and patient-specific factors (Table 6). Importantly, administrative decisions such as allocation schemes may either mitigate or exacerbate these underlying factors.
|Geography||Administration||Regional practice variation||DSA practice variation||Center variation||Patient variation|
|Area||Number of DSAs||Listing exceptions||Donor rates||Transplant volume||Reoperations|
|Population||Number of transplant centers||Donor quality||Patient selection||Etiology of heart failure|
|Population density||Geographic allocation rules||Donor management||Waitlist management||Clinical condition|
|Travel times||Multiple organ listings||Donor selection|
|Technical aspects of recovery|
This work represents an initial attempt to understand which geographic factors are associated with patient outcomes. As such, it focuses predominantly on identifying correlations. We can only speculate as to the underlying causal mechanisms. This is especially difficult in the setting of transplantation where interactions among the multiple factors—including, for example, donor quality and availability, recipient status, allocation rules and center experience—influence OPO behavior, individual center decisions to list and transplant, and outcomes on the waitlist.
The potential for interaction and multidirectional causality is illustrated by our analysis of factors associated with longer regional wait times. While several factors correlated with longer waittimes may be indicative of increased competition for available allografts among waiting recipients (fewer potential recipients per DSA, more patients listed at higher status and more listings at each center), the correlation between more donors per DSA and longer wait times cannot be explained in a similar fashion. However, it is possible that regions with longer wait times lead to more aggressive identification of donors by the OPO or a lower threshold for donor utilization by the transplant centers.
Unfortunately, our ability to identify high risk donors is severely limited. Characteristics making any particular donor high risk are subjective and determined in part by recipient characteristics. Despite these limitations, our data suggest that regions with more donors per transplant center also have a higher percentage of high-risk donors. In addition, regions with fewer transplants being performed use a higher proportion of high-risk donors. Better delineation of high-risk cardiac donors might enable a better understanding of how regional factors may influence transplant center decisions regarding allograft acceptance.
Variability in transplant center behavior may explain some portion of regional variation. We found that regions with fewer people per transplant center had a higher propensity to perform transplants for high-risk recipients. While multiple factors may contribute to this correlation, one possibility is that competition among transplant centers for a limited number of potential recipients results in the listing of more marginal transplant candidates at each center. On an individual basis, risk-adjusted outcomes on the waitlist were poorer for listed patients in regions with factors which might result in increased competition for available allografts, including fewer donors per center, more listings per center, fewer pediatric donors for each recipient.
Competition for donor allografts is a plausible explanation for many of the correlations we identified. In regional level analysis, regions with factors suggesting the possibility of increased competition had higher waitlist mortality. These included: more regional transplant centers, fewer donors per DSA, more listings per center and fewer transplants performed within a region. While regions with more transplant candidates cannot help but have increased competition for the limited allograft supply, increasing that supply may lessen the competition between centers.
As noted above, donor availability may be influenced by regional variation in health and healthcare delivery [5, 6], but individual OPO practice may also influence both donor availability and donor quality. Management of donor patients, as with many areas of intensive care medicine, may be influenced by volume and experience [24-26]. Both individual and institutional experience determine the relationship between volume and outcome in other areas, and both may be higher in OPOs serving larger DSAs. Even in the era of DonorNet©, expedited donation continues to be used to prevent organ discard ; in these cases individual OPO practice and organ allocation may vary significantly from the normal allocation scheme, contributing to geographic variation. Quality improvement processes, including cross-OPO sharing of best practices, may enable improvement in cardiac allograft recovery rates even in the absence of changes to the legal framework around organ donation .
The allocation process itself may influence waitlist mortality by directing organs toward patients likely to survive longer at the expense of sicker patients . Early allocation schemes relied heavily on the arbitrary boundaries of DSAs, but in order to minimize the impact of DSA boundaries on waitlist survival, heart allocation was changed in 2006 to allow broader geographic sharing. Singh et al.  have demonstrated improved outcomes following the changes (although other changes—including the increased use of VADs) may have also influenced these results. Additional changes in the allocation of pediatric donors in May 2009 further decreased the relevance of DSA geography. Currently, all patients in status 1A/1B within 500 miles or in the local DSA have the same priority (rather than prioritizing the local DSA). However, local DSA status 2 recipients still have priority over Zone B (500–1000 miles) status 1A recipients. While status 2 recipients have declined over time , in the current dataset fully 22% of transplants occurred to status 2 recipients, suggesting the possibility that DSA boundaries may continue to influence allocations. Further research utilizing data identifying all lower-ranked potential matches at the time of each allocation is required to assess the continued impact of these DSA-based allocation rules, as has been done with lung transplantation . While heart allocation does not occur at the level of UNOS regions, policy and clinical practice vary between regions. Listing status criteria are standardized, but listing exceptions are handled by individual regional committees and policies. In liver transplantation, the percentage of pediatric recipients transplanted with listing exceptions varies across regions . The same may be true of heart allocation. In the absence of objective criteria for exceptions, it is plausible that the threshold for an exception may be higher in regions with more centers, more listed patients and more “competition” for each donor organ. Consistent with this possibility, we found an increased number of transplant centers within a region to be predictive of longer wait list times and higher waitlist mortality.
There was significant regional variation in posttransplant outcomes, but regional variables were not correlated with outcomes following transplantation (other than the percentage of local donors which likely reflects the influence of shorter ischemic times [7, 9]). This is consistent with the likelihood that—in cardiac transplantation—regional variables are associated predominantly with variability in organ allocation rather than organ quality. Outcomes in other solid organ transplants support this hypothesis: while outcomes after renal transplant have been decreasingly influenced by regional variation, that is not the case with liver transplantation [23, 33, 34], where there is significant variation in donor quality among transplanted hepatic allografts . In contrast, donor cardiac allografts are recovered from a more limited number of donors, and—within the subset of recovered and transplanted allografts—donor quality may play less of a role [19, 35].
If allograft allocation and regional variables do not influence posttransplant outcomes, what is the source of the large regional variation in risk-adjusted mortality rates (from 5.4% to 18.0%)? We can speculate that this is the result of geographic variation in access to follow-up, transplant center practice variation and other factors not included in our analysis. However, further research should be directed at identifying these factors and mitigating the extent to which the location of a patient's transplant alters their likelihood of long-term survival.
As with all large dataset retrospective studies, there are several limitations to this analysis. First, although the risk adjustment models are relatively robust in comparison to other published models , they still only account for approximately 80% of the variation in risk, a more accurate model might result in more accurate assessment of the impact of regional variation on outcomes. Second, individual center variation may account for some of the regional variation, although hierarchical modeling should identify significant underlying center variation as the cause of regional differences. Third, and most importantly, correlation is not causation and the relationships identified here need to be further investigated to determine whether changes to allocation or regional administration are necessary in order to minimize geographic variation.
Outcomes on the waitlist and following transplantation vary widely by region. Waitlist mortality is higher in regions with increased competition for available donor allografts—those with higher population density, more DSAs, more transplant centers and more transplants. The effect on posttransplant mortality is less clear and warrants further investigation. The interaction between geographic variability in topography and health, regional and DSA policies and the decisions of individual transplant centers regarding acceptable donors are complex and multidirectional. Minimizing the impact of arbitrary geographical boundaries on allocation is a start, but further work is needed to identify areas where policy changes might mitigate the impact of geography on outcomes in pediatric heart transplantation.
This work was supported in part by Health Resources and Services Administration contract 234-2005-300711C. The content 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 U.S. Government.
The authors of this manuscript have no conflicts of interest to disclose as described by the American Journal of Transplantation.
Correction made after online publication May 28, 2013: Supporting Information has been updated.
* Census 2000 was used because complete data have not yet been published for Census 2010.
† RUCA 2.0, available online at http://depts.washington.edu/uwruca/.
‡ Secondary analyses were performed using the composite end-point of death prior to removal from the list or within 90 days of removal. Results were substantially the same and are not reported.