This study used data from the SERVE trial. The responsible investigator at each site ensured that approval from an appropriately constituted Competent Ethics Committee was sought for the clinical study. All changes in the research activity (including the case of planned or premature study end), all unanticipated problems involving risks to humans and the final report were reported to the ethics committee.

The study was carried out in accordance with the protocol and with the principles outlined in the current version of the Declaration of Helsinki. The ethics committee and regulatory authorities received annual safety and interim reports and were informed about study stop/end in agreement with local requirements.

The study population consisted of participants from the SERVE trial with valid baseline NT-proBNP values. The detailed methodology of the study has already been published.14,15 In short, the SERVE trial is a double-blind, randomized, placebo-controlled, multicenter trial comparing placebo vs tadalafil in a 1:1 ratio. All participants started on tadalafil 20mg or placebo once daily, or both. At baseline, a clinical examination, a 12-lead electrocardiogram, and an echocardiogram were performed. Neurohormonal activation, detailed cardiac anatomy, biventricular volumes, and function, as well as exercise capacity and cardiopulmonary fitness were assessed by means of blood tests, cardiac magnetic resonance (CMR) or coronary multidetector computed tomography (CMDCT) if CMR was contraindicated, and a cardiopulmonary exercise test, respectively. These examinations were repeated after 1 year and at the end of the study follow-up 3 years later. For the purpose of this analysis, data on 98 patients with no missing NT-proBNP values at baseline from a total of 100 were analyzed.

Concentrations of NT-proBNP (ng/L) were analyzed in a core lab with expertise in biomarker analysis: the Cardiovascular Research Institute Basel (CRIB) in Basel, Switzerland. An electrochemiluminescence immunoassay (Elecsys, Roche Diagnostics, Switzerland) was used. Specific patient sets with plastic tubes and barcodes corresponding to the individual patient numbers and time points in the study were used. Samples were collected locally, followed by centrifugation, aliquoting, and initial storage at−80°C at the participating centers. Routine sample collection at the local study sites and refrigerated transportation to the dedicated biobank at the CRIB were provided by the core lab. The data obtained from the analysis were then centrally entered into the electronic database of the SERVE trial.

The combined primary outcome of the current study (from now on also referred to as cardiac complications) was defined as the occurrence of either clinically relevant arrhythmias (requiring hospitalization and/or therapeutic intervention), hospitalization for HF, or all-cause death.

Secondary outcomes investigated the pathophysiological determinants of NT-proBNP in patients with an sRV by assessing the correlations of this biomarker with baseline biventricular end-diastolic and end-systolic volumes, stroke volumes, and ejection fraction assessed by CMR/CMDCT.

Continuous variables are presented as medians (interquartile range [IQR]), and categorical variables as counts (percentages). Baseline characteristics were stratified by NT-proBNP quartiles and compared using the Kruskal-Wallis test for continuous variables and the chi-square or Fisher exact test for categorical variables, as appropriate. Confidence intervals were computed as recommended.16

Friedman 2-way ANOVA was used for the comparison of dependent samples. Multivariable linear regression models (age-, sex-, and creatinine-adjusted) were fitted for each baseline CMR/CMDCT variable related to biventricular volumes and function. To avoid dichotomizing continuous variables and imposing linearity, continuous variables were modeled using restricted cubic splines. Three spline knots were placed at 0.1, 0.5, and 0.9 percentiles of each variable marginal distribution, following Harrell's recommendations.17 Linear regression assumptions were checked using diagnostic plots. Consequently, the dependent variable (NT-proBNP) in each model was log-transformed to satisfy the assumption of homoscedasticity. To avoid dichotomizing our main variable of interest (NT-proBNP), another Cox proportional hazards model corrected for age and sex was fitted. NT-proBNP was then modeled with a restricted cubic spline function (3 knots) to take possible nonlinearity into consideration and to treat the main variable of our study continuously. Because NT-proBNP was modeled with a restricted cubic spline, we expressed its hazard ratio (HR) using the 75th percentile of our NT-proBNP values (429 ng/L) as the reference value. We calculated the HR of several levels of NT-proBNP compared with this specific reference value. The proportional hazard assumption was tested by means of the Schoenfeld residuals. The assumption was not violated (overall P=.73; figure 1 of the supplementary data). In addition, to investigate the association of clinically relevant parameters with the occurrence of the primary outcome, an age- and sex-corrected Cox proportional hazards model was fitted for each of the following variables: New York Heart Association (NYHA) functional class> I, NT-proBNP, predicted peak oxygen uptake (predicted peak VO2 [%]), and right ventricular ejection fraction (RVEF), as %. To test the added value of NT-proBNP on top of RVEF and peak VO2, a sequential modeling approach starting with RVEF, adding peak VO2, and ending with NT-proBNP was performed. Due to the limited sample size, we were unable to adjust the base model for more prior information.

Survival analysis for the occurrence of the combined primary outcome of patients with NT-proBNP values lower vs higher than the cutoff value (> 429ng/L) was assessed using the Kaplan-Meier estimator. If a patient experienced multiple events, event-free survival time was calculated from the date of inclusion until the occurrence of the first event. The statistical significance for the difference between groups was assessed by means of the log-rank test. The individual prognostic accuracy of NT-proBNP and those of other established markers for predicting the primary outcome was assessed by means of the concordance statistic (c-statistic). To compare the prognostic value of NT-proBNP, we built a model adjusted for age and sex and compared the c-statistic with adjusted models for RVEF and predicted peak VO2.

All hypothesis testing was 2-tailed, and P values <.05 were considered statistically significant. There was 1 missing value for the variable “predicted peak VO2 (%)”. This patient was excluded from the corresponding analysis. Data analysis was performed in R, version 4.1.3 (R Foundation for Statistical Computing, Austria).

There were 24 primary outcome events in 20 (20%) patients. Specifically, there were 2 deaths, 5 hospitalizations due to HF, and 17 clinically relevant arrhythmias (15 supraventricular arrhythmias and 2 ventricular arrhythmias). The occurrence of the combined primary outcome increased significantly within the fourth quartile compared with quartiles 1-3 (for each of quartile 1-3: 3 events (12%), compared with 11 events in quartile 4 (46%); P=.012). Levels of NT-proBNP were significantly higher among patients meeting the combined primary outcome (816 [194-1094] vs 205 [122-357]; P=.003). The median [IQR] of NT-proBNP levels for arrhythmia and HF were 322 [190-1095] and 1202 [928-3980], respectively. The 2 (male) patients who died had NT-proBNP levels of 773 and 932 ng/L, respectively. The occurrence of the primary outcome did not significantly differ among cardiac anatomy groups (6 [24%], 5 [22%], and 9 [17%]; P=.768 for ccTGA, d-TGA after Mustard operation, and d-TGA after Senning operation).

HRs with 95% confidence intervals (95%CI) for other established risk factors for increased risk of cardiovascular events during the 36-month follow-up are shown in table 2. The association between NT-proBNP levels and the risk of 36-month complications (primary outcome) is depicted in figure 1. A significant nonlinear dose-response association is shown. HRs (95%CI in gray) of NT-proBNP compared with the cutoff value of 429 ng/L for the occurrence of the primary outcome are also outlined in figure 1. Event-free survival (median [IQR] follow-up duration=37 [36-37]) was significantly higher in patients with NT-proBNP values below the cutoff value (429 ng/L) when compared with patients above this level (figure 2). The unadjusted c-statistic for NT-proBNP and other established markers for poor prognosis are depicted in table 3. There was no difference in the prognostic performance of NT-proBNP, RVEF, NYHA class> I, and peak VO2 (predicted) for the occurrence of the primary outcome (table 3). When comparing Model I to Model II and Model III (table 4), there was no significant difference in their prognostic value (P=.815 and .935, respectively). The base model with RVEF yielded a c-statistic of 0.69 (95%CI, 0.56-0.82). Incorporating peak VO2 into the model resulted in a c-statistic of 0.7 (95%CI, 0.56-0.84). Finally, the addition of NT-proBNP to the model comprising RVEF and peak VO2 increased the c-statistic to 0.71 (95%CI, 0.57-0.85). These results are depicted in table 5.

Figure 1.

NT-proBNP and the occurrence of the primary outcome. This dose-response plot shows the association between NT-proBNP levels at baseline and the risk of complications at 36 months. We calculated age- and sex-adjusted hazard ratios (95%CI) for several levels of NT-proBNP and compared them to our reference value of 429ng/L. 95%CI, 95% confidence interval; HR, hazard ratio; NT pro-BNP, N-terminal pro–B-type natriuretic peptide.

(0.21MB).

Figure 2.

Survival analysis stratified by a NT-proBNP level of 429ng/L. Kaplan-Meier curve showing the occurrence of the combined primary outcome in patients with NT-proBNP values above and below the median. There was a significantly higher occurrence in patients with levels at or above 429ng/L. 95%CI, 95% confidence interval; HR, hazard ratio; NT pro-BNP, N-terminal pro–B-type natriuretic peptide.

(0.32MB).

In a sensitivity analysis assuming linearity for the association between NT-proBNP and the outcome and for the association between NT-proBNP and the magnetic resonance imaging parameters, the results were comparable (figure 2 and 3 of the supplementary data).

Measurements of biventricular volumes and function are shown in table 1. Systemic RV end-systolic and end-diastolic volumes, ejection fraction and stroke volume were significantly associated with NT-proBNP (P <.03 for all correlations).

A graphic representation of the multivariable linear regression models for the correlation of NT-proBNP levels with CMR- or CDMCT-derived biventricular indexed volumes and function are depicted in figure 3. A statistically significant positive correlation between NT-proBNP levels and right ventricular end-systolic and end-diastolic volume index was observed (coefficient of determination: +0.481; P <.001, and +0.427; P <.001 respectively). There was a statistically significant negative correlation between NT-proBNP levels and RVEF (coefficient of determination: −0.561; P <.001). For the remaining parameters, no statistically significant correlation with NT-proBNP was observed.

Figure 3.

Correlation of NT-proBNP levels with variables regarding the right ventricular size and function. Multivariable linear regression models (age-, sex-, and creatinine adjusted) were fitted for baseline biventricular end-diastolic and end-systolic volumes, stroke volumes and ejection fraction assessed by either cardiac magnetic resonance or coronary multidetector computed tomography. There was a significant association between NT-proBNP levels and right ventricular volumes and function. LV, left ventricular; NT pro-BNP, N-terminal pro–B-type natriuretic peptide; RV, right ventricular.

(0.58MB).

Previous studies analyzing the prognostic impact of NT-proBNP were either performed in heterogeneous cohorts of patients with adult congenital heart disease or were retrospective in nature.

Popelovà et al.11 investigated the prognostic value of NT-proBNP in 646 adults with various congenital heart lesions. These authors found a significant correlation between NT-proBNP and all-cause mortality. Of these, 84 patients (13%) had transposition of the great arteries after Mustard or Senning operation and 22 patients (3%) had ccTGA.11 Baggen et al.12 also examined the prognostic value of serial NT-proBNP measurements in 602 ACHD patients, including 65 (11%) patients with d-TGA after atrial switch operation and 20 (3%) patients with ccTGA. They found an association between elevated NT-proBNP measurements and a combined endpoint of death, HF, hospital admission, arrhythmia, thromboembolic event, and cardiac intervention. However, neither study reported lesion-specific outcomes. A retrospective study of 87 patients after atrial switch operation found NT-proBNP to be the most reliable prognostic marker for mortality.18 Another retrospective cohort study of 89 TGA patients with prior atrial switch operation showed that BNP had a high predictive value for distinguishing patients with and without critical cardiac events.13

The strength of our analysis is that measures of neurohormonal activation, ventricular volumes and function derived from CMR and CMDCT, as well as determinants of exercise capacity, were obtained from dedicated core labs, used in the SERVE trial.14,15 While our analysis supports findings from previous studies regarding the prognostic value of NT-proBNP levels, the determined NT-proBNP cutoff> 429 ng/L derived from our data may serve as a clinical risk marker, although this should be confirmed in larger patient cohorts. Additionally, our data strongly support the prognostic importance of NT-proBNP levels, which are comparable to RVEF and peak oxygen uptake on exercise testing, and may therefore help in tailoring the frequency and type of follow-up studies in clinically stable patients.

The role of NT-proBNP in patients with a systemic left ventricle has been thoroughly investigated, and its importance in the evaluation and management of these patients, for example in the setting of the emergency department, is undisputed.19

While the importance of NT-proBNP in predicting adverse events in ACHD is mentioned in the current guidelines, there is no specific recommendation for patients with an sRV.20 Our study showed that NT-proBNP may help to discriminate patients who are at risk of clinically relevant arrhythmia, HF, or death. Values of NT-proBNP> 429 ng/L may now be seen as a screening tool for closer follow-up or even as a turning point to consider starting HF medication in asymptomatic patients with an sRV. The correlations of NT-proBNP with sRV volumes and function points toward a comparable pathophysiological pathway to that seen in subaortic left ventricles. Furthermore, the prognostic accuracy of NT-proBNP alone was comparable to that of models using RVEF or predicted peak VO2. Thus, this widely available and inexpensive biomarker may be seen as an alternative to other more costly, time-consuming, and less universally available tools.

Only 24 cardiac events occurred during our 3-year follow-up period. Of these, only 2 were deaths. Furthermore, the overall number of patients was low. Given the relatively small sample size of the cohort, the suggested cutpoint for NT-proBNP should be interpreted with caution. Studies with longer follow-ups and a larger number of patients are needed to assess the prognostic value of NT-proBNP with regard to mortality and to validate the cutpoint and assess its extrapolation and robustness. For the same reason, we were only able to adjust for some selected variables, and the effect of other variables on NT-proBNP levels and their prognostic value is still unknown. Additionally, Kaplan-Meier curves should be interpreted with caution due to potential residual confounding.