SEARCH

SEARCH BY CITATION

Keywords:

  • Children;
  • heart failure;
  • heart transplantation;
  • modeling;
  • pediatric;
  • risk factors;
  • survival

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. Disclosure
  9. References

We sought to develop and validate a quantitative risk-prediction model for predicting the risk of posttransplant in-hospital mortality in pediatric heart transplantation (HT). Children <18 years of age who underwent primary HT in the United States during 1999–2008 (n = 2707) were identified using Organ Procurement and Transplant Network data. A risk-prediction model was developed using two-thirds of the cohort (random sample), internally validated in the remaining one-third, and independently validated in a cohort of 338 children transplanted during 2009–2010. The best predictive model had four categorical variables: hemodynamic support (ECMO, ventilator support, VAD support vs. medical therapy), cardiac diagnosis (repaired congenital heart disease [CHD], unrepaired CHD vs. cardiomyopathy), renal dysfunction (severe, mild-moderate vs. normal) and total bilirubin (≥ 2.0, 0.6 to <2.0 vs. <0.6 mg/dL). The C-statistic (0.78) and the Hosmer–Lemeshow goodness-of-fit (p = 0.89) in the model-development cohort were replicated in the internal validation and independent validation cohorts (C-statistic 0.75, 0.81 and the Hosmer–Lemeshow goodness-of-fit p = 0.49, 0.53, respectively) suggesting acceptable prediction for posttransplant in-hospital mortality. We conclude that this risk-prediction model using four factors at the time of transplant has good prediction characteristics for posttransplant in-hospital mortality in children and may be useful to guide decision-making around patient listing for transplant and timing of mechanical support.


Abbreviations: 
CHD

congenital heart disease

CHD-R

congenital heart disease repaired

CrCl

creatinine clearance

ECMO

extracorporeal membrane oxygenation

HT

heart transplantation

IQR

interquartile range

OPTN

Organ Procurement and Transplant Network

OR

odds ratio

ROC

receiver–operator characteristic

UNOS

United Network for Organ Sharing US, United States

VAD

ventricular assist device

Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. Disclosure
  9. References

Heart transplantation (HT) can be a life-saving therapy for children with end-stage heart failure (1–4). Although long-term survival of children undergoing HT is quite favorable compared to older candidates (3,5), short-term mortality continues to be a major problem for children—surpassing that of all other age groups (3,5). Most of these early deaths occur before the patient is discharged home from the hospital after HT (3,6,27,12), often in the intensive care unit after weeks or months of aggressive medical support has proven to be futile. Because of the critical shortage of donor organs in the United States (7), questions have emerged as to whether it is possible to more accurately predict before listing which patients are unlikely to survive transplant, as well as to determine whether a child's candidacy for transplant could be improved, for example, through the use of a ventricular assist device (VAD), as has been demonstrated in adult heart transplant candidates.

Although risk factors for early mortality after pediatric HT have been well described (3,6,27,12), no validated quantitative risk-prediction model for early mortality has been described. The availability of an accurate risk-prediction tool for posttransplant in-hospital mortality could be a useful aid to guide decision-making regarding a child's overall candidacy for HT, how transplant candidacy might be changed by successful transition to a VAD, as well as how we might work toward reducing the number of futile transplants, a major goal of contemporary organ allocation policy. Therefore, the specific objective of this study was to develop and validate a risk-prediction model for in-hospital mortality after HT in children using information readily available at the time of HT.

Materials and Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. Disclosure
  9. References

Study population and data source

The Organ Procurement and Transplant Network (OPTN) is an internally-audited, mandatory, multicenter government-supported solid-organ transplant registry that collects information on all solid-organ transplants in the United States (7). Demographic and clinical information is reported by centers to OPTN prospectively at the time of initial listing, eventual transplant and annually thereafter. Subjects eligible for the primary analysis (for development and internal validation of model) included all children <18 years of age who underwent primary orthotopic HT in the United States between January 20, 1999 and January 14, 2009 (n = 2707). Patients listed for heart retransplantation or multiviscera transplants were excluded. All patients were followed from the time of HT until death or hospital discharge.

Study definitions and endpoints

All clinical and demographic variables were defined at the time of HT. Transplant outcomes were modeled based on patient characteristics at the time of transplant rather than at listing because they were felt to be more directly relevant to transplant outcome and may have changed appreciably from conditions at the time of listing which may be weeks or months earlier. The level of hemodynamic support was categorized into four mutually exclusive categories: extracorporeal membrane oxygenation (ECMO), ventilator/no ECMO, VAD/no ventilator or medical therapy (none of the above). Children on VAD and a ventilator were too few in number to model separately (N = 35) so were categorized in the ventilator group because their baseline characteristics and outcomes seemed most similar to this group. Cardiac diagnosis was divided into three mutually exclusive groups: nonstructural heart disease (cardiomyopathy and myocarditis), repaired congenital heart disease (CHD-R) and unrepaired congenital heart disease. Creatinine clearance (CrCl) was estimated using the Schwartz Formula (21) and analyzed as a continuous variable, and as three mutually exclusive age-dependent CrCl categories: normal, severe renal dysfunction defined by a CrCl ≤50% of the lower limit of normal for age or dialysis, and mild-moderate renal dysfunction for all CrCl values in-between (8). Total serum bilirubin, a marker of hepatic function in heart failure, was analyzed as a continuous variable and in three mutually exclusive categories: <0.6 mg/dL (normal), 0.6 to <2.0 mg/dL (mild-moderate hepatic dysfunction) and ≥2.0 mg/dL (severe hepatic dysfunction). Race/ethnicity data (categories included: black, white, Hispanic, other) were analyzed as reported by the transplant center.

The primary endpoint was death before hospital discharge (i.e. in-hospital mortality) after pediatric HT. Subjects who died before discharge were considered to have reached the primary endpoint (i.e. had an event). Statistical analyses were performed using SAS version 9.1 (SAS Institute Inc, Cary, NC, USA) and STATA version 10.0 (StataCorp LP, College Station, TX, USA).

Model development and validation

Two-thirds of study participants were randomly assigned to the model-development cohort. Summary statistics are presented as median (interquartile range [IQR]) or number (percent). Donor data such as ejection fraction, geographic distance or crossmatch results were not considered to simulate the circumstances whereby initial patient evaluation and listing decisions are made. Missing values for CrCl (3.4%) and total bilirubin (12.6%) were imputed using linear regression analysis. Variables used for imputation were age, gender, diagnosis, type of support, total bilirubin (to predict CrCl) and CrCl (to predict total bilirubin).

Logistic regression was used to develop a risk-prediction model for in-hospital mortality using forward stepwise selection. Variables significant at the 0.05 level based on the likelihood ratio test and which minimized the Akaike information criterion were retained in the final model. Model discrimination was quantified using the area under the receiver–operator characteristic curve (C-statistic). Model calibration was assessed using the Hosmer–Lemeshow goodness-of-fit test. The final model was used to estimate the probabilities of in-hospital death for patients with specified baseline characteristics.

The best-fitting model was tested in the internal validation cohort, which was comprised of the remaining one-third of study participants. The C-statistic was used to quantify model discrimination and the Hosmer–Lemeshow goodness-of-fit test was used to assess model calibration. Because model prediction may perform less well as clinical practice changes over time, the model was also validated externally in a second, independent, prospective cohort of children transplanted during the period between January 15, 2009 and February 28, 2010.

The authors had full access to and take full responsibility for the integrity of the data. All authors have read and agreed to the manuscript as written.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. Disclosure
  9. References

Patients

Of 2965 patients who underwent HT during the study period (1999–2009), 2707 met the study inclusion criteria. Two hundred five were excluded because of retransplantation, 14 were excluded because of multiorgan transplantation and 39 were excluded because of missing data on posttransplant survival. Of these, 1805 subjects (two-thirds) were randomly selected for the model development cohort, and 902 subjects (the remaining one-third) were used for model validation.

Model development cohort

The baseline characteristics of the model-development and model-validation cohorts are presented in Table 1. Overall, the median age of subjects was 4 years (IQR 0, 13 years), 47% had cardiomyopathy, 7% were on ECMO, 8% were on VAD support, 14% were on ventilator support (no ECMO) and 71% were on medical therapy. Thirteen percent met criteria for mild-moderate renal dysfunction whereas 5% met criteria for severe renal dysfunction. By contrast, 48% met criteria for mild-moderate hepatic dysfunction and 20% met criteria for severe hepatic dysfunction.

Table 1.  Baseline characteristics of the study participants (N = 2707)
VariableDevelopment cohort (N = 1805)Validation cohort (N = 902)
  1. Values are summarized as number (percent) or median (25th–75th percentile), mean ± standard deviation.

  2. ECMO = extra-corporeal membrane oxygenation; LVAD = left ventricular assist device; BIVAD = biventricular assist device; UNOS = United Network for Organ Sharing; CrCl = creatinine clearance; PRA = panel of reactive antibodies.

Demographic information  
 Age at transplant (year)5 (0–13)4 (0–12)
 6.5 ± 6.26.3 ± 6.0
 Weight (kg)16 (7–41)16 (7–41)
 26.6 ± 24.925.9 ± 23.3
 Female gender799 (44)401 (44)
 Nonwhite race781 (43)381 (42)
Cardiac diagnosis  
 Cardiomyopathy853 (47)413 (46)
 Congenital heart disease  
   Repaired522 (29)274 (30)
   Unrepaired236 (13)105 (12)
 Other194 (11)110 (12)
Hemodynamic support  
 ECMO130 (7)67 (7)
 Ventilator261 (14)126 (14)
 VAD  
   LVAD49 (3)27 (3)
   BIVAD78 (4)42 (5)
   Other3 (<1)1 (<1)
 Medical therapy  
   Inotropic support577 (32)276 (31)
   Oral therapy707 (39)363 (40)
Medical condition  
 Home573 (32)305 (34)
 Hospitalized258 (14)116 (13)
 Intensive care974 (54)481 (53)
UNOS listing status (n = 1771, 881)  
 1A1013 (57)522 (59)
 1B263 (15)135 (15)
 2495 (28)224 (25)
End-organ function  
 Renal  
   Creatinine clearance (CrCl)110 (74–142)109 (74–147)
 115 ± 66117 ± 66
   Dialysis39 (2)24 (3)
   CrCl categories for age (n = 1740, 868)  
   Normal1436 (83)703 (81)
   Mild-moderate234 (13)117 (13)
   Severe70 (4)48 (6)
 Hepatic  
   Serum total bilirubin0.7 (0.4–1.5)0.8 (0.4–1.4)
 1.7 ± 3.41.6 ± 2.9
   Bilirubin categories  
   <0.6 mg/dL573 (32)281 (31)
   0.6–2.0 mg/dL874 (48)451 (50)
   >2.0 mg/dL358 (20)170 (19)
PRA peak > 10%245 (14)171 (19)

For the model-development cohort, there were a total of 132 deaths (7.3%) before hospital discharge after a median posttransplant duration of 16 days (range 0–428 days). Univariable factors associated with in-hospital mortality included younger age, lower weight, nonwhite race, CHD (both repaired and un-repaired), inotropic support, United Network for Organ Sharing (UNOS) status 1A, pretransplant hospitalization in intensive care, ECMO support, ventilator support, VAD support, renal dysfunction, higher total bilirubin and panel of reactive antibodies >10%.

Table 2 summarizes the best-fitting model for in-hospital mortality. Four covariates, each with mutually exclusive categories, were selected for the best-fitting model: (1) level of hemodynamic support using four categories (ECMO, ventilator, VAD, medial therapy), (2) cardiac diagnosis using three categories (nonstructural heart disease, CHD-R, unrepaired CHD), (3) renal function using three categories (normal, mild-moderate dysfunction, severe dysfunction) and (4) hepatic function using three total bilirubin categories (normal, mild-moderate dysfunction and severe dysfunction). Overall, the model performed well with a C-statistic of 0.78 (95% confidence interval, 0.74, 0.83) and Hosmer–Lemeshow goodness-of-fit p value of 0.89. The following model equation describes the relationship between the covariates and outcome (solved for individual risk categories in Table 3):

  • image

Figure 1 depicts the predicted risk of in-hospital mortality for children undergoing HT based on cardiac diagnosis and the hemodynamic support, the two most important predictors in the model assuming mild-moderate end-organ dysfunction (Panel A). The predicted risk of death varies by 10-fold (4% to more than 40%). Overall, patients transplanted from a ventilator had a 50% higher predicted risk of death compared to patients transplanted from VAD support.

Table 2.  Best-fitting model2 for in-hospital mortality after pediatric heart transplantation
 Odds ratio (95% Confidence interval)p-Value
  1. 1Model equation: ln(p/[1-p]) =–4.3449 + 1.7262 (ECMO) + 1.1848 (Ventilator) + 0.6786 (VAD) + 1.1374 (Repaired CHD) + 0.6071 (Unrepaired CHD) + 0.5886 (CrCl Mild-Moderate) + 1.1838 (CrCl Severe) + 0.5575 (Bilirubin ≥0.6, <2.0) + 1.0348 (Bilirubin ≥2.0).

  2. 2Reference category.

  3. ECMO = extracorporeal membrane oxygenation; VAD = ventricular assist device; CHD = congenital heart disease.

Hemodynamic support  
 ECMO5.6 (3.3, 9.6)<0.001
 Ventilator3.3 (2.0, 5.4)<0.001
 VAD2.0 (1.0, 4.0)0.06
 Medical therapy2-
Cardiac diagnosis  
 Repaired CHD3.1 (2.0, 4.8)<0.001
 Unrepaired CHD1.8 (1.0, 3.3)0.04
 Cardiomyopathy2
Creatinine clearance for age  
 Mild-moderate decreased1.8 (1.1, 2.9)0.02
 Severe decreased3.3 (1.7, 6.1)<0.001
 Normal2
Total bilirubin (mg/dL)  
≥0.6, <2.01.7 (1.0, 3.0)0.04
≥2.02.8 (1.6, 4.8)<0.001
<0.62
Table 3.  Predicted percent risk of posttransplant in-hospital mortality in risk categories
Creatinine clearance
 Normal for age (normal range for age) Total bilirubin (mg/dL)Mild-moderately decreased (50th to 99th percentile for age) Total bilirubin (mg/dL)Severely decreased (<50th percentile for age) Total bilirubin (mg/dL)
<0.60.6–1.2>1.2<0.60.6–1.2>1.2<0.60.6–1.2>1.2
Cardiomyopathy         
 Medical therapy1.3%2.2%3.5%2.3%3.9%6.2%4.1%6.9%10.7%
 Ventricular assist device2.5%4.3%6.7%4.4%7.5%11.5%7.7%12.7%19.0%
 Ventilator4.1%6.9%10.7%7.1%11.8%17.7%12.2%19.5%28.1%
 ECMO6.8%11.3%17.0%11.6%18.7%27.0%19.2%29.4%40.1%
Congenital heart disease–unrepaired         
 Medical therapy2.3%4.0%6.3%4.1%7.0%10.8%7.2%12.0%18.0%
 Ventricular assist device4.5%7.6%11.7%7.8%12.9%19.2%13.3%21.1%30.1%
 Ventilator7.2%12.0%18.0%12.3%19.7%28.3%20.3%20.8%41.7%
 ECMO11.8%18.9%27.4%19.4%29.6%40.4%30.4%43.3%55.2%
Congenital heart disease–repaired         
 Medical therapy3.9%6.6%10.2%6.8%11.3%17.0%11.7%18.8%27.1%
 Ventricular assist device7.4%12.2%18.3%12.6%20.1%28.8%20.7%31.3%42.3%
 Ventilator11.7%18.8%27.1%19.2%29.4%40.2%30.2%43.0%54.9%
 ECMO18.5%28.4%30.0%20.1%41.7%53.6%42.6%56.5%67.6%
image

Figure 1. Effect of level of hemodynamic support at transplant on predicted posttransplant in-hospital mortality according to cardiac diagnosis. All predicted values assume presence of mild-moderate end-organ dysfunction (i.e. CrCl 50th–99th percentile of lower limit of normal for age, total bilirubin 0.6–2.0 mg/dL). CMP = cardiomyopathy; CHD = congenital heart disease; ECMO =extracorporeal membrane oxygenation; VAD = ventricular assist device; VENT = mechanical ventilator.

Download figure to PowerPoint

Figure 2 depicts the predicted risk of in-hospital mortality in children with cardiomyopathy (Panel A) and repaired congenital heart disease (Panel B) with increasing end-organ dysfunction; the predicted risk of death varies by more than 50-fold (1–68%). The model predicts relatively low mortality in children with preserved end-organ function, even among children supported with ECMO where the predicted in-hospital mortality is just 7%. A stepwise decrease in end-organ function is associated with a sharp rise in risk of death independent of level of hemodynamic support or cardiac diagnosis.

image

Figure 2. Effect of worsening end-organ function on predicted posttransplant in-hospital mortality in children with cardiomyopathy (A) and in children with repaired congenital heart disease (B). ECMO = extracorporeal membrane oxygenation; VAD = ventricular assist device; Ventilator = mechanical ventilator.

Download figure to PowerPoint

Model validation

Baseline characteristics and in-hospital mortality (6.5%) were similar between the model development and internal validation cohorts (Table 1). When the model was applied to the internal validation cohort, the performance characteristics for discrimination (C-statistic 0.75, 95% confidence interval 0.68, 0.82) and calibration (p = 0.49) were similar to that obtained in the development cohort, although calibration seemed somewhat better at the lower end of the risk spectrum relative to the upper end. Figure 3(A) shows the predicted versus observed in-hospital mortality risk based on deciles of risk for the internal validation cohort.

image

Figure 3. Comparison of predicted versus observed posttransplant in-hospital mortality for the internal validation cohort (A; N = 902), and the independent validation cohort (B; N = 388). The dotted line is the line of identity in each graph.

Download figure to PowerPoint

Because model performance may have changed in the most recent era where the use of VADs has become significantly more common, we validated the performance of the model in an independent, prospective cohort of 338 children transplanted in the most recent year (January 15, 2009–February 28, 2010. As anticipated, more children were transplanted from VAD support (16% vs. 8%) and fewer were transplanted from ECMO (5% vs. 7%, p < 0.001) during 2009–2010. Nevertheless, the C-statistic (0.81, 95% confidence interval 0.73, 0.89) and the Hosmer–Lemeshow goodness-of-fit (p = 0.53) suggested excellent prediction of in-hospital mortality despite greater utilization of VAD therapy in children (Figure 3B).

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. Disclosure
  9. References

In this study, we developed and validated a simple risk-prediction tool that can be used to predict a child's chances of surviving a HT to hospital discharge. Overall, the risk of in-hospital mortality is 7%, however, the predicted risk for individual children varies by more than 50-fold (i.e. 1% to greater than 65%) based on four patient covariates. These covariates include cardiac diagnosis, level of hemodynamic support, CrCl for age and serum total bilirubin. These factors are readily available to clinicians at the bedside and can be used to estimate the risk of dying before discharge if transplanted in their present condition. The model has good prediction characteristics based on validation studies in children transplanted contemporaneously with the model-development cohort and more recently where pediatric VAD use has become significantly more common (9–11).

The risk factors we identified for early mortality are consistent with published data (6,27,12) from the Pediatric Heart Transplant Study (PHTS) Group (9,10), the UNOS (12) and the International Society of Heart and Lung Transplantation Registry Reports (3). However, to date no studies have incorporated these covariates into a quantitative risk-prediction tool and subsequently validated it using an independent cohort of patients. Davies et al. (12) developed a risk-prediction model for both 30-day and 1-year mortality after pediatric HT; however, the model included covariates believed, a priori, to be associated with clinical outcome rather than covariates selected solely on the basis of improving model performance, had a lower C-statistic (0.66–0.68) and was not validated (12). To our knowledge, this is the first report to develop and validate a simple risk-prediction model that can be used to predict the likelihood that a child with certain characteristics will survive to hospital discharge if transplanted in their present condition.

We elected to focus on early mortality after HT because mortality during convalescence from surgery accounts for a disproportionate number of the deaths observed after HT (3,5,9,12). This is especially true for infants and children where the alpha portion of transplant survival curve dives steeply immediately after transplant before settling into the slower beta portion or decay phase (3). One of the most compelling ways to improve overall survival in pediatric HT would be to improve early posttransplant survival through refinements in patient selection, strategic deployment of mechanical support devices and broader community consensus on when the risks of transplant death are sufficiently high that transplant may be either futile or unethical given the critical shortage of organ donors and high waiting list mortality (13,14). In contrast to prior studies of early mortality in the field (3,5,12), we chose to analyze the endpoint of death before hospital discharge rather than mortality to a fixed time interval such as 30-days or 1-year post-HT (3,5,12,15). Use of hospital discharge as the primary endpoint allows some natural variability for recovery time and permits clinical information (recovery from surgery sufficient to be discharged home) that represents an important milestone for the clinicians and the patient. This approach has been recently favored by clinical researchers in adult cardiovascular surgery for similar reasons both for the evaluation of risk factors (16,17) and for risk prediction (18).

We were somewhat surprised to discover how strongly renal function predicted early mortality given that other national risk-adjustment models—most notably the online Scientific Registry for Transplant Recipients (SRTR) model (19) used for public reporting of transplant center performance (20)—has found no association between renal function and early posttransplant mortality (19). We suspect this difference may be explained by how renal function was analyzed as a covariate in the two models: In contrast to the SRTR model where serum creatinine is analyzed without adjustment for pediatric body-surface area or age (19), we created age-specific CrCl categories (corresponding to categories used by the National Kidney Foundation in children; Ref. 8) to reflect the normal age-related changes in clearance over time. (N.B., for adolescents, a CrCl of 35 mL/min/1.73 m2 is severely decreased, but only moderately decreased for a toddler, and entirely normal for an infant <3 months of age; Ref. 21). Not only does renal function become highly significant in our model similar to adult SRTR models for 1-month and 1-year mortality (19), but overall model performance also seems to be superior based on the performance characteristics of the two models (15,19).

Our findings have several important implications. First, our model may be useful as a guide to clinicians to assess more objectively a child's overall candidacy for HT and in counseling families regarding their child's probability of surviving HT to hospital discharge across changing conditions. Indeed, one of the greatest challenges facing HT clinicians is determining the optimal timing to offer transplant especially when alternative or competing treatment options exist (e.g. additional palliative surgery, biventricular pacing or ongoing medical therapy). Fundamentally, the decision to offer transplant depends on assessment that heart transplant offers a clinically important survival advantage over alternative treatment options (2). Such decision can be especially difficult in complex patients where multiple comorbidities may coexist in varying degrees of severity. Having access to a validated risk-prediction tool capable of providing a summary risk assessment may offer some advantages over the current, qualitative approach to risk assessment, which relies heavily on individual clinician judgment and center experiences. Likewise, the model may be useful to clarify which patients are highly unlikely to benefit from transplant such that transplant may be considered futile or unethical (22,23) given the critical shortage of donor organs and high wait-list mortality (13,14).

Second, the present model may also be useful for evaluating how specific interventions, such as VAD implantation, might be used in selected clinical situations to improve a child's posttransplant outcome (24,25). To date much of the attention surrounding pediatric VADs has focused on the opportunity to improve wait-list survival. Our analysis suggests VAD support may also be justifiable on the basis of improving posttransplant survival (i.e. by improving the medical condition of a child before transplant surgery). We identify at least three patient groups/clinical situations where posttransplant survival may be improved by transitioning to VAD support. These include (1) children experiencing a decline in end-organ function while on inotropic support that may be reversed with VAD before transplant (2), children requiring mechanical ventilation even in the absence of end-organ dysfunction and (3) children on ECMO support even in the absence of overt end-organ dysfunction (Figures 2A and B). It is important to note that the present model is limited to predicting early posttransplant mortality assuming a specific set of clinical conditions at the time of transplant and does not seek to evaluate the impact of VAD placement on the risk of wait-list mortality, an important issue that is the subject of a separate analysis.

Third, our findings may provide useful clues into how policy makers might revise the current pediatric heart allocation to maximize overall transplant survival for children (26). A well-recognized limitation of the present heart allocation system is that it attempts to prioritize donor organs solely on the basis of medical urgency without taking into account the potential benefit of transplant to the recipient. As such, there is no formal mechanism to avoid allocating donor organs to children who are medically urgent but unlikely to benefit from transplant (e.g. children on ECMO with poor end-organ function; Refs. 3,10,12,27). To the extent the present model can provide a validated means for predicting which patients are likely to benefit from transplant, policy makers may be able to adapt the strategy to incorporate both medical urgency and transplant benefit into the allocation of pediatric donor hears (similar to the Lung Allocation Score; Ref. 28). This would better align the pediatric heart allocation system with contemporary allocation standards, as outlined in the US Department of Health and Human Service's Final Rule on Organ Allocation (29).

Finally, although our model was developed for risk-prediction purposes, our model may also be useful for risk-adjusting patient outcomes. The availability of accurate risk-adjustment models has become increasingly important in the current era where greater attention is being paid to the standardized evaluation of center performance for quality improvement efforts. Because different risk-adjustment methodologies may lead to divergent assessments of center performance and risk (30,31), every effort should be made to select the most accurate risk-adjustment tools for reporting center performance.

This study has several potential limitations. First, we identified patients retrospectively through a national registry of transplant recipients, which creates the opportunity for selection bias if a skewed population of patients was examined. However, because (1) OPTN captures all patients rather than a subset of patients undergoing HT in the United States, and (2) data collection occurs prospectively as new information becomes available, it is unlikely that selection bias played any role in the findings. Second, missing data were present for some covariates such as CrCl (3.4%), and bilirubin (12.6%). Because these variables were associated with outcome, we imputed the missing values for bilirubin and CrCl using other potentially associated variables that yielded similar final models with comparable performance characteristics. Third, OPTN data lacks detailed cardiac diagnosis and procedure information, which if present, could improve the prediction of early posttransplant outcomes. Nevertheless, the performance of the model by all conventional standards is good-to-excellent. Opportunities to improve risk prediction by combining diagnostic information from other national data registries like the PHTS Group would be a worthwhile endeavor. Finally, the sample sizes of the validation cohorts were small which could account for the observation that model calibration for the internal validation cohort was better at the lower end of the risk spectrum than upper. However, the Hosmer–Lemeshow calibration statistics were highly nonsignificant in both validation cohorts and the model calibration seemed better in the more recent, prospective dataset (Figure 3B).

In summary, we have developed and validated a simple tool to predict a child's likelihood of surviving HT to hospital discharge. This prediction tool, based on four covariates readily available at the bedside may be helpful to guide medical decision-making around patient eligibility for HT and utility of VAD therapy for improving a child's chances of surviving transplant. Our findings may also be useful to policy makers who are seeking to revise the current pediatric heart allocation system to minimize overall transplant-related mortality, which remains unacceptably high in the current era.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. Disclosure
  9. References

K.G. had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of data analysis. The work was supported in part by Health Resources and Services Administration contract 234–2005-370011C. The data reported here were supplied by the UNOS as the contractor for the OPTN. The data reported here are based on OPTN data as of February 28, 2010. The interpretation and reporting of these data are the responsibility of the authors and in no way should be seen as an official policy of or interpretation by the OPTN or the US Government. This study was supported by Heart Transplant Education and Research Fund, Department of Cardiology, Children's Hospital, Boston.

Disclosure

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. Disclosure
  9. References

Dr. Almond is a nonpaid consultant to Berlin Heart, Inc. Dr. Almond serves as Co-PI of the Berlin Heart EXCOR IDE Clinical Trial and as PI on an FDA grant supporting the trial (R01 FD 03557).

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. Disclosure
  9. References
  • 1
    Kantrowitz A, Haller JD, Joos H, Cerruti MM, Carstensen HE. Transplantation of the heart in an infant and an adult. Am J Cardiol 1968; 22: 782790.
  • 2
    Canter CE, Shaddy RE, Bernstein D, et al. Indications for heart transplantation in pediatric heart disease: A scientific statement from the American Heart Association Council on Cardiovascular Disease in the Young; the Councils on Clinical Cardiology, Cardiovascular Nursing, and Cardiovascular Surgery and Anesthesia; and the Quality of Care and Outcomes Research Interdisciplinary Working Group. Circulation 2007; 115: 658676.
  • 3
    Kirk R, Edwards LB, Aurora P, et al. Registry of the International Society for Heart and Lung Transplantation: Twelfth Official Pediatric Heart Transplantation Report-2009. J Heart Lung Transplant 2009; 28: 9931006.
  • 4
    Rosenthal D, Chrisant MR, Edens E, et al. International Society for Heart and Lung Transplantation: Practice guidelines for management of heart failure in children. J Heart Lung Transplant 2004; 23: 13131333.
  • 5
    Taylor DO, Stehlik J, Edwards LB, et al. Registry of the International Society for Heart and Lung Transplantation: Twenty-sixth Official Adult Heart Transplant Report-2009. J Heart Lung Transplant 2009; 28: 10071022.
  • 6
    Morrow WR, Naftel D, Chinnock R, et al. Outcome of listing for heart transplantation in infants younger than six months: Predictors of death and interval to transplantation. The Pediatric Heart Transplantation Study Group. J Heart Lung Transplant 1997; 16: 12551266.
  • 7
    Organ Procurement and Transplant Network (OPTN). Available at; http://www.unos.org. Accessed November 21, 2010.
  • 8
    Hogg RJ, Furth S, Lemley KV, et al. National Kidney Foundation's Kidney Disease Outcomes Quality Initiative clinical practice guidelines for chronic kidney disease in children and adolescents: evaluation, classification, and stratification. Pediatrics 2003; 111: 14161421.
  • 9
    Chrisant MR, Naftel DC, Drummond-Webb J, et al. Fate of infants with hypoplastic left heart syndrome listed for cardiac transplantation: A multicenter study. J Heart Lung Transplant 2005; 24: 576582.
  • 10
    Dipchand AI, Naftel DC, Feingold B, et al. Outcomes of children with cardiomyopathy listed for transplant: a multi-institutional study. J Heart Lung Transplant 2009; 28: 13121321.
  • 11
    Morales DL, Almond CS, Jacquiss RDB, et al. Bridging children of all sizes to cardiac transplantation: The initial multi-center North American experience with the Berlin Heart EXCOR Pediatric Ventricular Assist Device. J Heart Lung Transplant 2011; 30: 18.
  • 12
    Davies RR, Russo MJ, Mital S, et al. Predicting survival among high-risk pediatric cardiac transplant recipients: An analysis of the United Network for Organ Sharing database. J Thorac Cardiovasc Surg 2008; 135: 147155, 155.e141–142.
  • 13
    Almond CS, Thiagarajan RR, Piercey GE, et al. Waiting list mortality among children listed for heart transplantation in the United States. Circulation 2009; 119: 717727.
  • 14
    McDiarmid SV, Cherikh WS, Sweet SC. Preventable death: Children on the transplant waiting list. Am J Transplant 2008; 8: 24912495.
  • 15
    Dickinson DM, Shearon TH, O’Keefe J, et al. SRTR center-specific reporting tools: Posttransplant outcomes. Am J Transplant 2006; 6: 11981211.
  • 16
    Birkmeyer JD, Siewers AE, Finlayson EV, et al. Hospital volume and surgical mortality in the United States. N Engl J Med 2002; 346: 11281137.
  • 17
    Karamlou T, Diggs BS, Person T, Ungerleider RM, Welke KF. National practice patterns for management of adult congenital heart disease: Operation by pediatric heart surgeons decreases in-hospital death. Circulation 2008; 118: 23452352.
  • 18
    Nashef SA, Roques F, Michel P, Gauducheau E, Lemeshow S, Salamon R. European system for cardiac operative risk evaluation (EuroSCORE). Eur J Cardiothorac Surg 1999; 16: 913.
  • 19
    Risk-adjustment models, Scientific Registry for Transplant Recipients. Available at; http//www.ustransplant.org/csr/current/modtabs.aspx. Accessed November 21, 2010.
  • 20
    Heart Transplant Survival: Center Reports, Scientific Registry of Transplant Recipients [database on the Internet] 2009. Available at: http://www.srtr.org/csr/current/Centers/Default.aspx. Accessed November 21, 2010.
  • 21
    Schwartz GJ, Haycock GB, Edelmann CM Jr., Spitzer A. A simple estimate of glomerular filtration rate in children derived from body length and plasma creatinine. Pediatrics 1976; 58: 259263.
  • 22
    Ethical considerations in the allocation of organs and other scarce medical resources among patients. Council on Ethical and Judicial Affairs, American Medical Association. Arch Intern Med 1995; 155: 2940.
  • 23
    Kilner JF. Chapter 1: Limited Resources. In: Who lives? Who dies? Ethical criteria in pateint selection. New Haven, CT : Yale University Press, 1990.
  • 24
    Lietz K, Miller LW. Patient selection for left-ventricular assist devices. Curr Opin Cardiol 2009; 24: 246251.
  • 25
    Shams OF, Ventura HO. Device therapy for heart failure: When and for whom? Am J Cardiovasc Drugs 2008; 8: 147153.
  • 26
    Heart Subcommittee of the Thoracic Committee. Currency of Criteria in Pediatric Heart Allocation Policy, edited by UNfOS. February 26, 2010.
  • 27
    Mital S, Addonizio LJ, Lamour JM, Hsu DT. Outcome of children with end-stage congenital heart disease waiting for cardiac transplantation. J Heart Lung Transplant 2003; 22: 147153.
  • 28
    Egan TM, Murray S, Bustami RT, et al. Development of the new lung allocation system in the United States. Am J Transplant 2006; 6: 12121227.
  • 29
    Federal Register, Vol. 63, pp. 1629516332, April 2, 1998, [codified at Title 42 CFR Part 121].
  • 30
    Atherly A, Fink AS, Campbell DC, et al. Evaluating alternative risk-adjustment strategies for surgery. Am J Surg 2004; 188: 566570.
  • 31
    Steinberg SM, Popa MR, Michalek JA, Bethel MJ, Ellison EC. Comparison of risk adjustment methodologies in surgical quality improvement. Surgery 2008; 144: 662667.