Characteristics Associated With High-Performing Pediatric Heart Transplant Centers in the United States From 2006 to 2015

Key Points Question What are the characteristics associated with high-performing pediatric heart transplant (HT) centers? Findings In this cohort study of 3211 recipients of HT younger than 18 years at 44 pediatric HT centers in the US during 2006 to 2015, the top one-third of centers by performance had significantly lower 90-day mortality across recipient risk spectrum and significantly lower 90-day mortality in recipients who developed rejection or posttransplant kidney failure during transplant hospitalization. Meaning These findings suggest the presence of superior processes and systems of care at high-performing pediatric HT centers and may provide insights for designing quality improvement initiatives at low-performing centers.


Introduction
Heart transplant (HT) in children younger than 18 years is currently offered at more than 60 centers in the United States, and these centers perform 400 to 450 transplants per year. 1 Although some of these are adult centers performing occasional HT in older children or teenagers, most are low-volume pediatric HT centers performing fewer than 5 transplants annually. With such a large number of low-volume centers and overall 1-year survival that exceeds 90%, discerning outcome differences among centers adjusted for case-mix is difficult. We have previously described risk modeling for early mortality after HT in children using baseline characteristics at HT. 2,3 Such models, when generated using the entire cohort, can be used to assess center case-mix, assessed as expected mortality, and center performance, assessed as standardized mortality ratio (SMR; the ratio of observed to expected mortality), with lower SMR in higher-performing centers. 4 Using this approach, our previous work has demonstrated that most of the outcome differences between centers materialize within the first 90 days after transplant. 4 High-performing centers have lower risk-adjusted 90-day mortality and maintain their performance advantage over other centers on longer-term follow-up.
Previous studies have often focused on center volume as the key characteristic associated with pediatric HT outcomes. 5,6 While countries with national health care systems usually offer HT at a small number of large-volume centers, many pediatric centers in the US are low-volume centers.
Understanding signature characteristics of high-performing centers may guide quality improvement (QI) initiatives at low-performing centers that may help improve HT outcomes at these centers and overall. 7 We hypothesized that high-performing pediatric HT centers have significantly better 90-day outcomes in higher-risk recipients.
The specific aims of this study were to compare high-and low-performing US pediatric HT centers for 90-day posttransplant mortality across recipient risk spectrum, 90-day cause-specific mortality, and incidence of rejection and posttransplant kidney failure requiring dialysis during the transplant hospitalization and 90-day posttransplant mortality among recipients with these complications.

Study Setting and Participants
We identified all children younger than 18 years in the Organ Procurement and Transplant Network Children who received heart retransplantation or multiorgan transplantation were excluded. We also excluded centers if they had not performed an HT in a child younger than 10 years (to exclude adult centers), were inactive for 5 years consecutively, or had performed a total of fewer than 10 HT procedures during the 10-year study period. The OPTN database includes baseline demographic and clinical information at HT in all recipients in the US submitted by transplant centers. These data are supplemented with death data from the Social Security master death file.

Study Design and Variables
This was a retrospective cohort study. The data were analyzed during January to March 2020. The primary outcome was 90-day posttransplant mortality across recipient risk spectrum. The exposure was high-, medium-, and low-performing centers, stratified using 90-day SMR for each center (SMR tertiles).
Demographic and clinical variables were defined at transplant. Race/ethnicity was recorded as reported by the center and analyzed as White (ie, non-Hispanic White), Black (ie, non-Hispanic Black), Hispanic, or other. Kidney function was analyzed as estimated glomerular filtration rate (eGFR, in mL/min/1.73 m 2 ) using serum creatinine and the modified Schwartz equation. 8 For children aged 1 year or older, normal kidney function was defined as eGFR greater than 60; moderate dysfunction, eGFR 30 to 60; and severe dysfunction, eGFR less than 30 or dialysis support. For infants younger than 1 year, normal kidney function was defined as eGFR greater than 40; moderate dysfunction, eGFR 20 to 40; and severe dysfunction, eGFR less than 20 or dialysis support. 9 There were no missing data for age, sex, race/ethnicity, cardiac diagnosis, blood type, hemodynamic support (ie, inotrope support, ventilator, and type of mechanical support), health insurance (ie, Medicaid), dialysis, and the dates of transplant, death, or retransplant. For children with missing values of serum creatinine or bilirubin, we used multiple imputation to impute their eGFR and serum bilirubin using clinical variables at transplant; 10 imputations were used for each missing value. 10

Statistical Analysis
Baseline recipient characteristics are presented as median (interquartile range [IQR]) or number (percentage). Expected 90-day mortality (ie, risk of 90-day mortality) for each patient was assessed using a risk model. This model was developed using the entire study cohort and considered baseline recipient characteristics at HT and year of transplant. We used logistic regression with forward selection, retaining variables significant at the .10 level based on a likelihood ratio test. Model discrimination was assessed using the area under the receiver operating characteristic curve (ie, C statistic) and calibration. The model was internally validated using bootstrapping with 100 random samples and 3211 patients per sample with replacement. Model coefficients were used to estimate the probability of death within 90 days for each recipient, the expected 90-day mortality for each center (ie, case-mix) and 90-day SMR (ie, performance) for each center. Centers were stratified as high-, medium-, or low-performing based on SMR tertiles. Baseline characteristics of recipients at these centers were compared using the χ 2 test for categorical variables and the Kruskal-Wallis test for continuous variables. Observed (with 95% CIs) vs model-estimated expected mortality was assessed in the 3 SMR groups across the recipient risk spectrum. The groups were also compared for causespecific 90-day mortality, incidence of treated rejection and posttransplant kidney failure, and mortality within 90 days after HT following these complications using tests for trend performed using logistic regression, adjusting the SEs of model coefficients to account for the clustering of patients within centers.
Data were analyzed using SAS statistical software version 9.4 (SAS Institute) and Stata version 15 (StataCorp). All statistical tests were 2-sided, and P < .05 was considered statistically significant.

Results
During the 10-year study period, 3495 children younger than 18 years underwent their first HT at 86 US transplant centers. Of these, 31 centers were inactive during 2014 and 2015, 6 centers performed fewer than 10 HT procedures during the study period, and 5 centers were inactive during 5

Risk-Standardization for 90-Day Mortality
The risk model for 90-day posttransplant mortality ( Table 1) had an excellent ability to discriminate children who died from those who did not (C statistic, 0.80) and was well calibrated. Figure 1 illustrates estimated vs observed mortality in study children with expected mortality of less than 5% (ie, low risk), 5% to 9.9%, 10% to 14.9%, and greater than 15% (ie, high risk). On internal validation by bootstrapping, the C statistic ranged from 0.76 to 0.84 in repeated samples (optimism-corrected   The line indicates the line of identity; dots, observed vs the expected mortality within the expected mortality ranges of less than 5%, 5% to 9.9%, 10% to 14.9%, and 15% or greater. medium-performing centers across the recipient risk spectrum ( Figure 2B).  Figure 2C). Comparison of cause-specific mortality, as defined in the database, showed lower mortality associated with graft failure, stroke, multiorgan failure, infection, and hemorrhage deaths in high-performing centers ( Table 3).

Early Posttransplant Complications and Associated Mortality
Acute rejection was treated prior to posttransplant discharge in 100 recipients (10.3%) at highperforming centers, 131 recipients (10.3%) at medium-performing centers, and 94 recipients (9.7%) The line indicates the line of identity; dots, observed vs the expected mortality within the expected mortality ranges of less than 5%, 5% to 9.9%, 10% to 14.9%, and 15% or greater; and error bars, 95% CIs for observed mortality.

Discussion
We have previously reported that center differences in long-term survival of pediatric recipients of HT can be explained mostly by differences in center performance (risk-adjusted mortality) within the first 90 days. 4 In this study, we stratified US centers based on their 90-day SMR over a 10-year period into equal number of high-, medium-, and low-performing centers to further understand what highperforming centers did better. We found that in general, there were no significant differences in transplant volume or case-mix among the 3 groups. High-performing centers had statistically significantly lower-than-expected 90-day mortality across the recipient risk spectrum and had lower cause-specific mortality associated with most causes. Their recipients were less likely to develop kidney failure requiring dialysis prior to hospital discharge and were more likely to survive if they did.
Although the incidence of treated rejection was similar among the 3 groups, 90-day mortality among those who developed rejection was lower at high-performing centers. Thus, patients were less likely to die at high-performing centers after developing kidney failure or rejection. These findings suggest presence of superior patient care processes and systems for managing HT at highperforming centers.
Several studies have reported worse posttransplant outcomes at low-volume HT centers. 5,6,11,12 There was no statistically significant difference in distribution of transplant volume among the 3 groups in this current study, and half of high-performing centers were low-volume centers. The low expected mortality at these low-volume centers suggests that these centers limited their recipients to patients with low risk. The larger-volume high-performing centers, in contrast, accepted and transplanted many recipients with high risk with better-than-expected outcomes in such patients.
Although these findings of volume vs case-mix within the stratum of high-performers may be somewhat intuitive, they illustrate that high-performing centers, irrespective of volume, were better at selecting and transplanting patients appropriate for their center's expertise, systems, and resources at the time. The fact that institutional factors other than volume play an important role has also been reported for adult recipients, where center volume explained only 16.7% of center variability in 1-year post-HT mortality among US centers. 7 A similar incidence of rejection was seen among the 3 groups and is not surprising, considering the availability and adoption of modern immune suppression at most centers. 13 Early post-HT kidney failure requiring dialysis has been previously reported in 6% of pediatric recipients and is associated with pretransplant risk factors of patient complexity and severity of illness. 14 Despite a similar case-mix and frequency of pretransplant kidney failure among groups, the incidence of posttransplant kidney failure was significantly lower at high-performing centers. However, 90-day mortality among recipients who developed rejection or kidney failure during transplant hospitalization was significantly lower at high-performing centers compared with medium-and low-performing centers. These comparative results have parallels in surgical literature, in which the term failure to rescue has been used to describe in-hospital death after adverse events, such as postsurgical complications. 15 There is extensive discussion regarding this phenomenon in critical care and postsurgical literature, with focus on institutional quality of care. 16,17 Whether or not one considers posttransplant deaths that follow rejection or acute kidney failure as failure to rescue, the systems of care at high-performing centers that allowed them to have much better outcomes in such patients could act as models for improving care at lower-performing centers.

JAMA Network Open | Cardiology
We used the study cohort for developing the risk model rather than using a previously published model because our primary goal was risk adjustment among centers for severity of case-mix. Risk models perform best within the cohort in which they are developed, and the performance metric of the model was as good as in any previously reported model. We also chose the entire cohort to develop the model (rather than two-thirds or three-fourths of the cohort) to be able to use the full information for developing model coefficients. 18

Study Implications
Such center comparisons are potentially useful for developing QI initiatives at low-performing centers. Most QI initiatives in medicine are targeted to improve processes, with the ultimate goal to improve outcomes. Because early survival after HT is excellent in the current era, it is difficult to design and illustrate the effectiveness of any QI initiative targeting mortality in a reasonable time period. This is particularly true at pediatric HT centers, where a single death may dramatically reduce a center's SMR with little chance of recovery until the event is no longer counted by the regulatory agencies. Our findings provide a potential roadmap for centers seeking to improve their early posttransplant outcomes through QI initiatives that focus on improving processes by benchmarking the observed-to-expected outcomes reported in this analysis, including the identification of the principle complications of acute rejection and kidney failure. An example may be for a transplant team to quantify the risk of early posttransplant mortality using a published risk model in candidates for HT instead of labeling such patients as low or high risk based on the team's perception. When considered in the context of posttransplant outcomes of patients with similar risk at their center in recent years, specific interventions, ranging from early engagement of specialists, referral to a different center or waiting to list until the patient risk can be lowered, that may improve the likelihood of patient survival may be considered. Another example would be to mandate a systematic evaluation or root-cause analysis of center processes and systems following one of the well-defined early posttransplant events, such as severe rejection, kidney failure, or death, to seek changes in existing processes that may prevent such events in the future.

Limitations
This study has several limitations. First, this was a retrospective study using registry data with inherent limitations of such data. However, submission of these data to UNOS by centers is required, the data are periodically audited by UNOS, and the data used for the current analysis are identical to those used by the regulatory bodies to generate center-specific reports. Second, we excluded centers performing pediatric HT in children on a rare basis, and our findings may not be generalizable to such centers. Third, the OPTN data do not provide information on systems, processes, and clinical practices that allowed high-performing centers to achieve superior outcomes. Understanding these will require studies using different data sets or a prospective design.

Conclusions
This cohort study found that high-performing pediatric HT centers had superior outcomes across recipient risk spectrum, lower cause-specific mortality associated with most causes, a lower incidence of acute kidney failure following transplant, and a higher rate of recovery following serious complications, such as allograft rejection and kidney failure. These findings suggest the presence of

JAMA Network Open | Cardiology
Characteristics Associated With High-Performing Pediatric Heart Transplant Centers