Simple Score to Assess the Risk of Rejection After Orthotopic Heart TransplantationClinical Perspective
Background—The aim of this study was to derive and validate a risk score for rejection after orthotopic heart transplantation.
Methods and Results—The United Network for Organ Sharing registry was used to identify patients undergoing orthotopic heart transplantation between 1998 and 2008. A total of 14 265 eligible patients were randomly divided into derivation (80%; n=11 412) and validation (20%; n=2853) cohorts. The primary outcome was drug-treated rejection within 1 year of orthotopic heart transplantation. Covariates found to be associated (exploratory univariate P<0.2) with rejection were entered into a multivariable logistic regression model. Inclusion of each variable in the model was assessed by improvement in the McFadden pseudo-R2, likelihood ratio test, and c index. A risk score was then generated through the use of relative magnitudes of the odds ratios from the derivation cohort, and its ability to predict rejection was tested independently in the validation cohort. A 13-point risk score incorporating 4 variables (age, race, sex, HLA matching) was created. The mean scores in the derivation and validation cohorts were 8.3±2.2 and 8.4±2.1, respectively. Predicted 1-year rejection rates based on the derivation cohort ranged from 16.2% (score=0) to 50.7% (score=13; P<0.001). In weighted regression analysis, there was a strong correlation between these predicted rates of rejection and actual, observed rejection rates in the validation cohort (r2=0.96, P<0.001). Logistic regression analysis also demonstrated a significant association (odds ratio, 1.13; P<0.001). The c index of the composite score was equivalent in both the derivation and validation cohorts (c=0.67).
Conclusions—This novel 13-point risk score is highly predictive of clinically significant rejection episodes within 1 year of orthotopic heart transplantation. It has potential utility in tailoring immunosuppressive regimens and in research stratification in orthotopic heart transplantation.
Outcomes of orthotopic heart transplantation (OHT) for end-stage heart failure have improved significantly over the past few decades.1 This improvement is due in large part to improvements in immunosuppressive regimens and subsequent reductions in rejection and infection rates. More specifically, the introduction of cyclosporine and later tacrolimus and mycophenolate mofetil has led the way for OHT to become the gold standard therapy for end-stage heart failure.2,3 Despite these improvements, rejection remains a real concern in OHT, affecting roughly 30% of recipients during the first year.1 Moreover, patients with rejection during the first year have worse longer-term survival compared with those who do not. Although much progress has been made with immunosuppression in OHT, it remains an inexact science. A risk stratification tool that could be used to predict rejection rates after OHT could help in research investigations and aid in tailoring immunosuppressive therapy. The aim of this study was to derive and validate a risk score for rejection after OHT.
Editorial see p 2971
Clinical Perspective on p 3021
The Standard Transplant Analysis and Research files as provided by the United Network for Organ Sharing (UNOS) were used for this study. This registry provides deidentified data on all thoracic transplantations performed in the United States. Because of the exclusion of identifiable patient and center information, exempt status was granted by the institutional review board at our institution.
All adult (≥18 years) patients undergoing OHT during the study period (1998–2008) were identified in the UNOS registry. The following patients were excluded: patients with heterotopic heart transplantation, pediatric (<18 years) patients, those with heart retransplantations, those with multivisceral transplantations, and patients with inadequate clinical follow-up. The study population was randomly divided into 2 cohorts. A derivation cohort consisted of 80% of the study population, with the validation cohort comprising the remaining 20%. The primary end point was drug-treated rejection within 1 year of OHT.
Data and Statistical Analyses
The UNOS registry contained 461 variables. A comprehensive list of data that are routinely collected by UNOS can be found on their Web site (http://www.unos.org). The impact of all plausible recipient, donor, recipient-donor matching, and transplant covariates on 1-year rejection rates was evaluated in univariate logistic regression analysis in the derivation cohort. Those associated with rejection (exploratory P<0.2) were entered into the multivariable logistic regression model. To limit the effect of missing data, any covariate with an entry that was missing in ≥20% of patients was excluded from the multivariate model. Potential interactions between covariates were thoroughly tested; variables with a significance that was explained by collinearity with other variables were also excluded. In addition, only covariates that were found to improve the explanatory power of the multivariable model were included in the final iteration. Moreover, the appropriateness of inclusion of each covariate in the multivariate model was tested with the McFadden pseudo-R2, Akaike information criterion, likelihood ratio test, and c index (area under receiver-operating curve). The model was also tested with the Hosmer-Lemeshow goodness-of-fit test. In summary, the final multivariable model included only variables that were significant univariate predictors (P<0.2), had <20% missing data, and did not exhibit significant collinearity and the inclusion of which improved the explanatory power of the model.
The relative magnitudes of the odds ratios from the final multivariate model in the derivation cohort were then used to assign points for each significant covariate. A composite score was calculated for all patients in the derivation and validation cohorts. Weighted regression analysis was used to evaluate the correlation between predicted rejection rates based on the derivation cohort and actual, observed rejection rates for each risk score in the derivation and validation cohorts. Weights were assigned on the basis of the number of patients with each particular risk score. The association of risk score with rejection rates was also assessed in both cohorts with logistic regression analysis. Disjoint categories of risk score also were created, and rejection rates were calculated on the basis of these categories and compared. Additionally, the c index based on receiver-operating-characteristic curves was used to measure the predictive power of the composite score in the derivation and validation cohorts. Means are presented with standard deviations and odds ratios with 95% confidence intervals. All analyses were performed with version 11 STATA statistical software (StataCorp LP, College Station, TX).
A total of 14 381 OHT recipients were identified in the study period, of whom 116 (0.8%) had inadequate follow-up and were excluded. Therefore, 14 265 eligible patients were included in our analysis. The mean age of included patients was 51.9±12.0 years; 10 904 (76.4%) were male. The majority of patients were white (75.9%, n=10 776). The most common indication for OHT was ischemic heart disease (46.2%, n=6594), followed by idiopathic dilated cardiomyopathy (42.8%, n=6107). Only 302 patients (2.1%) underwent OHT for congenital heart disease. A total of 2078 patients (14.6%) were bridged to OHT with older, pulsatile-flow ventricular assist devices and 351 (2.5%) with newer-generation, continuous-flow ventricular assist devices.
The mean donor age in this study population was 31.1±12.3 years. The cause of death in the majority of donors was trauma related (59.8%, n=8525). Similar to recipients, the majority of donors were white (70.9%, n=10 109) and male (71.9%, n=10 253). Recipients and donors were sex matched in 10 274 cases (72.0%) and race matched in 8687 (60.9%). The mean ischemic time was 3.1±1.0 hours, with only 905 (6.3%) being >6 hours. Rejection within 1 year occurred in 5198 patients (36.4%). Those with rejection had a higher all-cause mortality rate than those without (30.3% versus 18.2%; P<0.001) at an overall clinical follow-up of 5.6±3.0 years.
Derivation and Validation Cohorts
Randomly generated derivation (80%, n=11 412) and validation (20%, n=2853) cohorts were created. These cohorts were well matched in key recipient characteristics (Table 1). They were also well matched in donor and transplant variables except for a statistical trend (P=0.05) in different distributions of donor race, although the absolute differences were small and likely clinically insignificant (Table 2).
Logistic Regression Analysis for Rejection in Derivation Cohort
Exploratory univariate logistic regression analysis for 1-year rejection in the derivation cohort yielded 30 potential covariates (P<0.2). The final multivariable model in the derivation cohort consisted of 15 of these variables (Table 3). The remaining 15 variables were excluded because of >20% missing data, collinearity, or failure to improve the explanatory power of the model as tested by various measures (Table 4). Moreover, each covariate in the final model had improved the power of the model as determined by the McFadden pseudo-R2, Akaike information criterion, likelihood ratio test, and c index. The final model had a c index of 0.67. Furthermore, the Hosmer-Lemeshow goodness-of-fit test had a nonsignificant P value of 0.39, suggesting that the final multivariable model appropriately fit the data.
Significant predictors of 1-year rejection in the final model in the derivation cohort included recipient age, sex, race, and HLA matching between recipient and donor. Moreover, older patients had a reduced risk of rejection, such that recipients >60 years of age had a 45.2% risk of 1-year rejection versus 30.1% for those <40 years of age (P<0.001). Additionally, female patients were at higher risk than male patients for rejection (43.5% versus 34.3%; P<0.001). With regard to recipient race, Asians had the lowest (23.0%) and blacks had the highest (40.1%; P<0.001) rate of 1-year rejection. Finally, patients with a high degree of HLA mismatching (ie, 4, 5, or 6 antigen mismatches) also had an increased risk of rejection (36.9% versus 32.1%; P=0.008).
It is important to note that year of transplantation and center volume were also significant in the multivariate analysis (each P<0.001). Despite these findings, these variables were not included in the generation of the score because of their limited utility in prospectively evaluating OHT patients for rejection risk. The other 4 significant variables were adjusted for the year of transplantation and center volume, however.
Generation of Risk Score
Points were assigned to each of the 4 significant covariates in the derivation cohort multivariable model (Table 3). These point values were assigned as whole numbers based on the relative magnitudes of the odds ratios in the multivariate analysis. Whole numbers were used to simplify the calculation of the composite risk score in the clinical setting. Overall, a 13-point composite score was generated with these covariates.
Predictive Value of Risk Score in the Derivation Cohort
The mean score in the derivation cohort was 8.3±2.2, with a range of 0 to 13 (Figure 1A). The predicted rate of 1-year rejection based on risk score ranged from 16.2% in those with a score of 0 to 50.7% in those with a score of 13 (Table 5). As expected, the actual, observed rates of rejection based on risk score in the derivation cohort correlated well with these predicted rates (Table 5). Indeed, in weighted regression analysis, the correlation between predicted and observed rates of rejection in the derivation cohort was highly significant with an r2 of 0.94 (P<0.001; Figure 2A). In addition, logistic regression analysis demonstrated a significant increase in observed rejection risk with increasing score in the derivation cohort (odds ratio, 1.14; 95% confidence interval, 1.12–1.16; P<0.001).
Validation of the Risk Score
The mean score in the validation cohort was 8.4±2.1 with a range of 2 to 13 (Figure 1B). Similar to the derivation cohort, increases in the risk score correlated with significant increases in the risk of rejection in the validation cohort in logistic regression analysis (odds ratio, 1.13; 95% confidence interval, 1.09–1.18; P<0.001). There was also a highly significant correlation between observed rates of rejection in the validation cohort and predicted rates based on the derivation cohort in weighted regression analysis, with an r2 of 0.96 (P<0.001;Figure 2B and Table 5). By disjoint categories, the risk of 1-year rejection in the validation cohort increased from 19.5% in those with a score of 0 to 5 to 43.9% in those with a score of 10 to 13 (P<0.001), which was similar to the increasing trend observed in the derivation cohort when stratified according to these same score categories (Figure 3). Furthermore, the c index was 0.67 in the validation cohort, which was equal to that in the derivation cohort. The Hosmer-Lemeshow goodness-of-fit P value was 0.76, demonstrating that the score also was an appropriate fit for the data in this cohort. A subanalysis was performed to ensure that the risk score was applicable to transplantations performed exclusively in the modern era. When limited to transplantations performed in 2005 or later, the composite score retained its highly significant association with rejection in the validation cohort (r2=0.96, P<0.001; odds ratio, 1.12; 95% confidence interval, 1.05–1.20; P=0.001).
Approximately 30% of OHT recipients in the modern era experience rejection during the first postoperative year.1 This has important implications for healthcare costs and patient survival. Indeed, OHT recipients experiencing rejection within 1 year have worse survival thereafter compared with those who are rejection free during this time interval.
Currently, there is no risk stratification tool to identify which patients are at high versus low risk for rejection after OHT. Thus, the aim of our study was to use the UNOS database to create such a tool. In examining data from >14 000 OHT recipients, we were able to derive and then validate a 13-point risk score for 1-year rejection in independent cohorts.
The overall 1-year rejection rate in our study was 36.4%, which correlates well with published data. It was encouraging to observe that rejection rates were significantly reduced in later transplantation years. This supports the postulate that advancements in drug therapies and overall care of the OHT patient have translated into better outcomes.
Risk Score Components
The components of the risk score included recipient age, race, sex, and HLA matching between the recipient and donor. Although some of these findings were unexpected, others were not. Moreover, older patients in our study had lower rates of rejection. Advancing recipient age is known to correlate with a lower risk of rejection and is thought to be due to a decline in immune function that accompanies ageing.4–6
Race was also found to affect 1-year rejection in our study. Whites were the most common race in our study population and served as the reference. Relative to whites, other races had statistically comparable rejection rates except for Asian recipients, who had significantly reduced rates of rejection. Interestingly, the race of the donor and recipient-donor race matching did not alter rejection rates. Although prior studies have demonstrated worse survival in black OHT recipients, the reasons for this remain unclear.7 Biological and socioeconomic factors have been implicated, as well as differential responses to immunosuppressive regimens.8
In addition to age and race, recipient sex affected rejection rates after OHT; female patients had a higher frequency of rejection. Similar to the effects seen in race, donor sex and sex matching did not independently affect 1-year rejection. Prior data have been mixed with regard to the effects of sex on OHT outcomes, although some studies have indeed suggested higher rates of rejection within the first year in female recipients such as that observed in our study.9,10 This increased risk has been postulated to result in part from the development of anti-HLA antibodies during pregnancy, although prior pregnancy in female recipients was a covariate that was excluded in our multivariate model owing to >20% missing data and therefore we could not evaluate its impact on rejection in this study.
The final component of the risk score was HLA matching between the recipient and donor. A high degree of mismatch (4, 5, or 6 antigen mismatches) correlated with an increased risk of rejection. The impact of HLA matching on rejection, graft survival, and/or patient survival in prior literature has been controversial, with multiple studies presenting conflicting results.11–13
Potential Clinical and Research Utility of the Risk Score
The 13-point risk score that was made up of these 4 variables was highly predictive of 1-year rejection after OHT. Indeed, in both the derivation and validation cohorts, the significant association between score and rejection was demonstrated in weighted regression analysis, in logistic regression analysis, and by c index. Moreover, there was a wide range of predicted rejection rates based on risk score, suggesting true clinical implications in patients with risk scores on either end of the spectrum. In addition to showing strong predictive capability in both the derivation and validation cohorts, this composite score had several other features that would be considered important in clinical application. For instance, it is easily calculable given that whole numbers are assigned to each component and that only 4 components exist. This, plus the fact that each component of the score represents data that are easily attainable in the clinical realm, would make this score easy to calculate and use for the practitioner.
Being able to predict which patients are at higher risk for rejection could aid in tailoring immunosuppressive regimens and rejection surveillance protocols. Furthermore, there are several areas of investigation with regard to immunosuppression in OHT such as the use of induction therapy, which could benefit from having a risk stratification tool. Induction protocols entail intense immunosuppressive regimens at the time of OHT and during the early postoperative period to prevent early acute rejection and to minimize the perioperative administration of nephrotoxic agents.14 Although roughly half of OHT recipients receive induction therapy, its efficacy in reducing early rejection and inducing graft tolerance remains to be elucidated.1 It may be useful to compare the outcomes of induction and other immunosuppressive strategies between low- and high-rejection-risk patients.
There are several limitations to this study. A major limitation is that we could not control for the variables included in the database because this is a nationwide, multi-institutional registry that independently collects this information. Therefore, some variables that may affect rejection rates but are not provided by UNOS could not be assessed. This includes immunosuppression protocols, which are not provided but would clearly be expected to affect posttransplantation rejection. Furthermore, some of the variables had a high percentage of missing data and were therefore not included in the final multivariate model, even though they may in fact be associated with rejection. We tried to limit the adverse effect of missing data on the power of our model by excluding such variables with a relatively high proportion of missing entries. In addition, we could not perform a time-to-event analysis because the number of days between OHT and rejection is not recorded in the UNOS data set. The risk score also was not normally distributed, with few patients having a risk score of <4. This was due to all races having 4 points automatically assigned except for the Asian population. We believe that having white race as the reference population was important, however, because it represented the majority of the study patients.
Another limitation of this study was that an independent validation cohort external to the UNOS database was not used, although we believe our cross-validation method was appropriate. Additionally, this analysis was restricted to patients undergoing OHT in the United States. These findings therefore cannot be extended to the international population with certainty. Thus, validation in OHT recipients outside the United States is also prudent.
In this analysis of >14 000 OHT recipients in the UNOS registry, we evaluated recipient, donor, recipient-donor matching, and transplant data to generate a 13-point risk score for 1-year rejection. This risk score was highly predictive of rejection after OHT. Furthermore, the risk score was validated in an independent set of patients. Therefore, this rejection risk score could have clinical implications in tailoring immunosuppressive therapies and could affect research study design by stratifying OHT recipient subjects into rejection risk categories.
Sources of Funding
This work was supported in part by Health Resources and Services Administration contract 234-2005-37011C. 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 US government.
- Received September 8, 2011.
- Accepted April 18, 2012.
- © 2012 American Heart Association, Inc.
- Stehlik J,
- Edwards LB,
- Kucheryavaya AY,
- Aurora P,
- Christie JD,
- Kirk R,
- Dobbels F,
- Rahmel AO,
- Hertz MI
- Mehra MR,
- Uber PA,
- Scott RL,
- Park MH
- George JF,
- Pamboukian SV,
- Tallaj JA,
- Naftel DC,
- Meyers SL,
- Foushee MT,
- Brown RN,
- Pajaro OE,
- McGiffin DC,
- Kirklin JK
In this study, we have derived and validated a risk score for rejection after orthotopic heart transplantation. We used data from the United Network for Organ Sharing registry on >14 000 orthotopic heart transplantation recipients to achieve this goal. The 13-point risk score incorporated 4 variables (age, race, sex, and HLA matching) and was highly predictive of drug-treated rejection in the first posttransplantation year. Predicted rates of 1-year rejection ranged from 16.2% in those with a score of 0 to 50.7% in those with a score of 13. In weighted regression analysis, there was a strong correlation between these predicted rates based on the derivation cohort and actual, observed rejection rates in the validation cohort (r2=0.96, P<0.001). Logistic regression analysis also demonstrated this significant association (odds ratio, 1.13; P<0.001). The c index for the score was equivalent in the derivation and validation cohorts (c=0.67). This highly predictive, simple-to-calculate risk score has potential utility in tailoring immunosuppressive regimens for orthotopic heart transplantation recipients and in research stratification.