Prediction of Waitlist Mortality in Adult Heart Transplant Candidates

The Candidate Risk Score

Abstract

Background 

The cardiac allocation system in France is currently based on urgency and geography. Medical urgency is defined by therapies without considering objective patient mortality risk factors. This study aimed to develop a waitlist mortality risk score from commonly available candidate variables.

Methods 

The study included all patients, aged 16 years or older, registered on the national registry CRISTAL for first single-organ heart transplantation between January 2010 and December 2014. This population was randomly divided in a 2:1 ratio into derivation and validation cohorts. The association of variables at listing with 1-year waitlist death or delisting for worsening medical condition was assessed within the derivation cohort. The predictors were used to generate a candidate risk score (CRS). Validation of the CRS was performed in the validation cohort. Concordance probability estimation (CPE) was used to evaluate the discriminative capacity of the models.

Results 

During the study period, 2333 patients were newly listed. The derivation (n =1 555) and the validation cohorts (n = 778) were similar. Short-term mechanical circulatory support, natriuretic peptide decile, glomerular filtration rate, and total bilirubin level were included in a simplified model and incorporated into the score. The Concordance probability estimation of the CRS was 0.73 in the derivation cohort and 0.71 in the validation cohort. The correlation between observed and expected 1-year waitlist mortality in the validation cohort was 0.87.

Conclusions 

The candidate risk score provides an accurate objective prediction of waitlist mortality. It is currently being used to develop a modified cardiac allocation system in France.

In Brief

This French study shows that a new Candidate Risk Score predicts waitlist mortality in heart transplant candidates without the need to consider hospitalization or inotropic infusion that depend on medical practices. Also, waitlist mortality of patients on mechanical bridge therapy is similar to that of patients without bridge therapy.

Heart transplantation is considered as the criterion standard for the treatment of advanced heart failure refractory to medical treatment^1,2 for carefully selected patients.³ There is an imbalance between the number of candidates registered on the waiting list and available grafts,^4,5 and the greatest survival benefit from transplantation is seen in patients at highest risk of death from heart failure.^6,7 Cardiac allocation systems around the world are thus primarily based on medical urgency, offering grafts to candidates at the highest risk of waitlist mortality.^8,9

Many models have been developed to predict mortality in ambulatory and hospitalized patients with heart failure,^1,2 but none have been validated on a transplant candidate population, including patients on mechanical circulatory support. Therefore, the priority statuses for heart allocation are worldwide based on therapies and not on candidate characteristics. The current French cardiac allocation system grants priority status to patients on inotrope support or short-term mechanical circulatory support (MCS) (high urgency 1 status) and to candidates on long-term MCS with device-related complications (high urgency 2 status).¹⁰ In 2015, 57% of transplant recipients in France had high urgency status at the time of transplantation. These cardiac allocation models may not properly prioritize candidates and may provide incentives to overuse therapies associated with complications, without protecting the system from being gamed.¹¹

With the use of the French national database, this study aimed to generate and validate a candidate risk score that predicts 1-year waitlist mortality or delisting for worsening medical condition, based on candidate's characteristics.

MATERIALS AND METHODS

Patients

This study was a cohort analysis using the French national database CRISTAL. All newly registered adult patients (16 years or older) on the French national waiting list for first, single-organ heart transplantation between January 1, 2010 and December 31, 2014, in the 25 active heart transplant centers in France were included. A total of 2623 patients were listed for transplantation during the study period. Candidates younger than 16 years (n = 153), as well as those listed for retransplantation (n = 73) or combined solid-organ transplantation (n = 78) were excluded from the analysis. A total of 2333 patients were finally included in the study. This population was randomly divided into a derivation (67%; n = 1555 candidates) and validation cohort (33%; n = 778 candidates).

Data Collection and Variables

CRISTAL is a national database initiated in 1996 and administered by the Agence de la biomédecine that prospectively collects data on all organ transplant candidates in France along with their outcomes. Data are entered into the registry by each center. Data collection is mandatory. Withdrawal from the waiting list and patient death are prospectively notified. The study was conducted according to the French law indicating that research studies based on the national registry CRISTAL are part of the transplant assessment activity and do not require institutional review board approval.

We analyzed variables potentially associated with the outcomes on the waiting list. Data selected included candidate's demographics, primary diagnosis, clinical status, device therapy, and, laboratory parameters at listing. Short-term MCS included venoarterial extracorporeal membrane oxygenation (VA-ECMO), and intra-aortic balloon pump. Long-term MCS comprised left ventricular, right ventricular, or biventricular assist device (VAD) or total artificial heart (TAH). Candidates on long-term MCS together with short-term MCS were considered to be patients on long-term MCS.

Glomerular filtration rate (GFR) was estimated using the Modification of Diet in Renal Disease formula (186.3 × (Creatinine in μmol/L/88.4)^−1.154 × Age^−0.203 (× 0.742 if sex female)). Plasma concentrations of natriuretic peptides (NPs) (B type NP or N-Terminal pro–B type natriuretic peptide) were collected and expressed per deciles. We used imputation of the 10th decile in patients on bi-ventricular assist device, TAH, and VA-ECMO, as these therapies are limited to the most severely ill patients while they modify natriuretic peptides synthesis.

Covariates with more than 20% missing data were excluded from the analyses. For relevant covariates with less than 20% missing data, missing values were substituted with values obtained by multiple imputations. The SAS MIANALYZE procedure was used to combine the results of the analyses.

Current Allocation System in France

The current French cardiac allocation system grants priority status to patients on inotrope support or short-term MCS (high urgency 1 status) and to candidates on long-term MCS with device-related complications (high urgency 2 status). In the geography-based allocation system, grafts are allocated by the Agence de la biomédecine to centers according to local, regional, and national geographic areas.

Statistical Analysis

We compared candidate characteristics at listing of the derivation and the validation cohorts using a χ² test or the 2-sided Fisher exact test for qualitative variables and the Student t test or the Wilcoxon rank sum test for quantitative variables.

Competing outcomes analysis was conducted in the derivation cohort. Overall survival was calculated from the day of listing until death or delisting for worsening medical condition with censoring at transplantation, delisting for other reasons, or lost to follow-up.

The main outcome assessed in this study was 1-year waitlist mortality or delisting for worsening medical condition. Survival curves were estimated using the Kaplan-Meier method and compared using the log-rank test.

Factors associated with 1-year waitlist mortality or delisting for worsening medical condition were determined using a Cox proportional hazards model in the derivation cohort. Multivariable analysis included all variables associated with 1-year death on the waiting list or delisting for worsening medical condition in univariate analysis at a P < 0.2. The study focused on creating an objective candidate risk score. Thus, we excluded the following covariates related to medical practice or medical assessment: need for hospitalization, requirement for inotropic infusion, and presence of ascites. The variables of the final model were selected by means of a stepwise backward procedure. Patients supported with long-term MCS were combined with candidates without MCS in the final model because they had similar waitlist mortality risk in the multivariable model (hazard ratio [HR], 1.5 [0.8-2.6]). Four commonly available variables of the 7 variables associated with 1-year waitlist mortality or delisting for worsening medical condition in multivariable analysis were incorporated into a simplified model. Models strengths were compared with the Akaike Information Criterion. The candidate risk score (CRS) was generated using the parameter estimations of this simplified survival model by summing the products of the factors and their β estimations. Comparison of the CRS distribution between the derivation and validation cohorts was performed using histograms.

Validation of the CRS was performed by calculating the regression coefficient and the HR of the CRS in the validation cohort.¹² In addition, we estimated and compared 1-year survival curves stratified by CRS quartile.

The discriminative capacity of the models from which the score was derived was assessed by the concordance probability estimation (CPE),¹³ and the calibration was tested by calculating the correlation between observed and predicted 1-year waitlist mortality by CRS deciles in the validation cohort. Statistical analyses were performed using SAS guide 7.1. P < 0.05 was considered statistically significant.

RESULTS

Candidate Population

The characteristics of the 2333 patients included in the study are presented in Tables 1a-c. Their median age was 53.0 years, 6.1% (n = 142) were 65 years or older, and 76.8% were men. The 2 main reasons for listing were dilated cardiomyopathy (46.5%) and coronary artery disease (35.4%). The proportion of candidates on MCS at listing was 21.9% (n = 510), with 12.3% (n = 286) of candidates on VA-ECMO, 1.5% (n = 36) on intra-aortic balloon pump, 7.5% (n =175) on VAD, and 0.5% (n = 13) on TAH. Other life supports included inotropic infusion (31.4% [n = 733]), mechanical ventilation (9.6% [n = 223]), and dialysis (1.2% [n = 28]). High urgency 1 status¹⁰ was granted to 44.2% of candidates and high urgency 2 status to 4.5%. Comorbidities included diabetes (15.2% [n = 355]), obesity (14.4% [n = 335]), preexisting cancer (7.3% [n = 171]), chronic renal dysfunction (12.1% [n = 283]), pulmonary disease (10.6% [n = 247]), history of stroke (9.7% [n = 227]), peripheral vascular disease (5.5% [n = 129]), active or former tobacco use (58.1% [n = 1355]), and a history of alcohol abuse (12.1% [n = 283]). A total of 24.5% (n = 571) of candidates had at least 1 previous cardiac surgery.

TABLE 1A: Comparison of candidate demographics at listing between derivation and validation cohorts

TABLE 1B: Comparison of clinical and medical characteristics of candidates at listing between derivation and validation cohorts

TABLE 1C: Comparison of candidate laboratory values at listing between derivation and validation cohorts

There was a similar distribution of variables between the derivation and validation cohorts, except for mechanical ventilation, which was more frequent in the derivation cohort, and being overweight, which was more frequent in the validation cohort.

Competing outcomes analysis demonstrated that 65.3% of candidates from the derivation cohort underwent transplantation within 1 year after listing. Death or withdrawal from the list for worsening medical condition occurred in 201 patients (12.9%) within 1 year after listing (Figure 1). The 1-year overall survival rate on the waiting list, estimated using the Kaplan-Meier method, was 76.9% (95% confidence interval [CI], 73.6%-79.8%; Figure 2).

FIGURE 1: Competing outcomes of candidates on the waiting list—derivation cohort, N = 1555.

FIGURE 2: Kaplan-Meier waitlist survival—derivation cohort.

Score Generation

Of the 32 candidate variables included in the univariate analysis, 22 were associated with 1-year mortality on the waiting list or delisting for worsening medical condition at a P < 0.2 within the derivation cohort: age older than 55 years, history of cancer, previous stroke, hospitalization, mechanical ventilation, dialysis, short-term MCS, inotropic therapy, defibrillator, anticoagulant therapy, hypertension, chronic renal dysfunction, diabetes, ascites, total bilirubin, alanine aminotransferase, GFR, plasma concentrations of NPs, serum sodium, hematocrit, platelet count and serum protein (Table 2). Three variables were not included in the model because they were related to medical practice or medical assessment: need for hospitalization, requirement for inotropic infusion, and presence of ascites.

TABLE 2: Univariate analysis of predictors of 1-year waitlist mortality or delisting for worsening medical condition—derivation cohort, n = 1555

In the multivariable model (Table 3a), factors associated with 1-year waitlist death or delisting for worsening medical condition were short-term MCS (HR, 2.5; 95% CI, 1.6-3.9), diabetes mellitus (HR, 1.5; 95% CI, 1.1-2.1), logarithmic bilirubin level (HR, 1.8; 95% CI, 1.4-2.3), logarithmic GFR (HR, 0.6; 95% CI, 0.5-0.9), NP decile (HR, 1.2; 95% CI, 1.1-1.2), hematocrit (HR, 0.95; 95% CI, 0.92-0.97), and serum sodium (HR, 0.97; 95% CI, 0.95-1.0). The discriminative capacity of this model was 0.74.

TABLE 3A: Multivariable analysis of predictors of 1-year waitlist mortality or delisting for worsening medical condition—complete model-derivation cohort

We constructed a simplified model using the 4 variables, which were the most commonly reported as predictors of mortality in end-stage heart failure: short-term MCS (HR, 3.7; 95% CI, 2.5-5.5), logarithmic bilirubin level (HR, 1.9; 95% CI, 1.5-2.3), logarithmic GFR (HR, 0.6; 95% CI, 0.5-0.8), and NP decile (HR, 1.2; 95% CI, 1.1-1.3; Table 3b).

TABLE 3B: Multivariable analysis of predictors of 1-year waitlist mortality or delisting for worsening medical condition—simplified model—derivation cohort

We generated a CRS using the regression coefficients from the multivariable analysis as follows: CRS = 1.301335 × short-term MCS + 0.157691 × NP decile − 0.510058 × ln(GFR) + 0.615711 × ln(bilirubin). The scores for the derivation cohort ranged from −1.6 to +4.8 with a median score of 0.66 (Figure 3A).

FIGURE 3: A, Candidate risk score histogram for the derivation cohort. B, Candidate risk score histogram for the validation cohort.

The CPE of this score was 0.73 for the total derivation cohort, 0.73 in candidates without long-term MCS, and 0.78 in long-term MCS candidates.

Score Validation

The CRS was applied to the validation cohort. The scores in this cohort ranged from −1.6 to +4.4 with a median score of 0.69 (Figure 3B). The CRS slope assessed using the Cox model was not significantly different from 1 (0.92 [0.7-1.1] (Table 4). The CRS HR was similar for the derivation (2.7 [2.4-3.1]) and validation (2.5 [2.1-3.0]) cohorts. The CPE of the CRS was 0.71 in the validation cohort. The correlation between observed and predicted 1-year waitlist mortality was excellent up to the ninth decile, where small sample size probably skewed the relationship (r = 0.87) and the slope of the calibration straight line was 0.93 (Figure 4).

TABLE 4: CRS validation

FIGURE 4: Plot of observed versus predicated 1-year waitlist mortality or delisting for medical condition by CRS deciles—validation cohort.

For the total cohort, 1-year waitlist survival was 93.4% [89.9%-95.7%] for CRS first quartile, 82.9% [77.9%-86.8%] for the second quartile, 71.7% [65.3%-77.1%] for the third quartile, and 40.5% [32.8%-48.1%] for the fourth quartile (Figure 5).

FIGURE 5: Kaplan-Meier waitlist 1-year survival stratified by CRS quartile—all candidates.

DISCUSSION

In the present study, we derived and validated an easily calculable CRS based on candidates' characteristics predicting 1-year waitlist mortality or delisting for worsening medical condition. By calculating the CRS for each candidate, it would be possible to rank candidates according to their medical urgency. Only a few predictive models of death on the waiting list have been generated and published.^7,14 These models were derived from studies conducted in the late 1990s and included patients registered on the waiting list before the current advances in MCS technology when pretransplant mortality was higher.

Here, we identified variables that were independent predictors for 1-year waitlist mortality or delisting for worsening medical condition. A CRS was derived from a simplified model including 2 markers of the hemodynamic severity of heart failure (short-term MCS use, plasma concentrations of NPs) and 2 markers of end-organ dysfunction (GFR, total bilirubin level). The model from which the CRS was derived performed well with good discriminative ability (CPE = 0.71) and an adequate estimation of mortality (r = 0.87). Existing models for predicting mortality in patients with heart failure seem to have only moderate discriminatory performance, in particular, in listed transplant candidates.^15,16 Moreover, none of the scores tested by Smits et al,¹⁶ predicted waitlist mortality in patients on VAD. Thus, the use of existing prognostic models in heart transplant candidates may inadequately distinguish patients who will die from those who will not. However, NP levels and renal function have consistently been reported to be strong predictors of death across different heart failure populations.¹⁷ Consistent with our findings, Wever-Pinzon et al¹⁸ also reported that patients on short-term MCS had higher waitlist mortality than other candidates, and Smedira et al¹⁹ showed that patients supported to transplant with ECMO had a higher risk of death than those bridged with left ventricular assist device. We recently reported that patients on VA-ECMO at listing showed a poor waitlist outcome compared to other candidates.²⁰ This worse outcome seemed to be due to the poor condition of the patients, as well as to device-related complications. Finally, liver function abnormalities have also been shown to have a prognostic impact on the outcome of patients with advanced heart failure.^21,22 Serum bilirubin levels, as an individual variable or a component of the Model for End-Stage Liver Disease scoring system, have been shown to be predictor for 1-year death, transplantation, and VAD requirement in ambulatory patients referred for transplant evaluation.²³

The principal finding of our study is that waitlist mortality can be predicted without considering the need for hospitalization and the requirement for inotropic infusion, which depend on medical practices. In addition, we found, as others, that waitlist mortality of patients bridged with long-term MCS, almost exclusively with new-generation continuous-flow left ventricular assist device, was similar to that of patients without MCS.^18,24 A notable finding of the study is that the CRS exhibited good discriminatory power for candidates supported with long-term MCS (CPE = 0.78).

The CRS seems to be an inappropriate measure of medical urgency in several subcategories of candidates. The study was restricted to candidates 16 years or older, and we are thus unable to ensure that waitlist mortality is accurately assessed using CRS in pediatric candidates. The well-known high risk of mortality in candidates supported with bi-VAD and TAH could not be captured by the CRS because of the small number of candidates in that subgroup. The discriminative capacity of the CRS remains questionable in candidates on VAD with device-related complications, as their cause of death is usually unrelated to heart failure. Finally, the CRS does not take the urgency of patients with life-threatening arrhythmias into account.

In light of these results, the French cardiac transplant community is developing a new graft allocation model based on the CRS. The new allocation system will allocate grafts to patients according to their CRS while considering blood type, morphology, and age donor-recipient matching. Exceptions to the CRS-based allocation system will be allowed for the subcategories of candidates for whom the CRS seems to be an inappropriate measure of urgency as pediatric candidates, patients on long-term MCS with device-related complications, and patients with contraindication to VAD implantation. Another consideration is that a cardiac allocation system based on the use of short-term MCS and renal and liver dysfunction may negatively affect posttransplant outcomes, as these factors are predictors of mortality after cardiac transplantation.^25,26 Posttransplant outcomes have to be balanced with waitlist mortality. The creation of a survival benefit-based cardiac allocation score has been proposed,^16,27 similar to the lung allocation system. A more pragmatic approach would be to preclude graft allocation to candidates with a posttransplant survival estimate below the threshold that defines futile transplantation. In the future French allocation system, access to transplantation will be denied to candidates with an expected 1-year posttransplant survival of 50% or lower. The posttransplant survival will be estimated using a model that has been developed from the French data including recipient and donor factors.²⁸

This study has several limitations. The analysis was based on data collected through a national registry subject to missing data and coding errors. Some variables not incorporated in the database may affect waitlist mortality and skew the study results. The validation cohort in this study was a random subset of the total cohort. External validation in another candidate population would be useful. The key limitation is that waitlist mortality differs from one country to the other. There were 10.7 deaths on the waiting list per 100 waitlist years in the United States in 2012-2013,⁵ whereas this figure was 23.8 in France in 2013.⁴ This difference may be driven by differences in graft availability and allocation algorithms as well as candidate characteristics and management. The CRS may therefore not be generalizable to other countries.

In conclusion, the CRS is a novel tool predicting poor waitlist outcome based primarily on candidate laboratory parameters. It does not incorporate the most frequently used medical and MCSs used to determine the priority status of candidates. This score can serve to develop an objective and robust urgency-based cardiac allocation algorithm.