We approached 91 SV patients meeting inclusion criteria and presenting to the UCSF Pediatric Heart Center between February 2007 and December 2010 for enrollment. Fourteen (15%) declined. Of the 77 remaining subjects, 6 (8%) were subsequently excluded because of missing data. Thus, 71 children were studied (). Of the single RV patients, 28 had hypoplastic left heart (10 with mitral/aortic atresia, 8 with mitral stenosis and aortic atresia, and 10 with mitral/aortic stenosis), 13 had a right dominant atrioventricular canal defect, 6 had double-outlet RV, 2 had interrupted aortic arch with ventricular septal defect and small LV, and one patient each had aortic stenosis and severe LV dysfunction or L-transposition of the great arteries with severe pulmonary stenosis and small LV. Of the single LV patients, 10 had tricuspid atresia, 3 had pulmonary atresia with intact ventricular septum, 2 had double-inlet LV, 2 had D-transposition of the great arteries with straddling mitral valve and small RV, and one had critical pulmonary stenosis with small RV. The patients with indeterminate ventricular morphology had primitive ventricles with heterotaxy syndrome.
Patients were assigned Ross scores as a measure of clinical HF as previously described1
() and the surgical stage of palliation was noted (). Of the 49 children free of clinical HF, the median Ross score was 1±0.8. For the 22 children with clinical HF, the median Ross score was 4±1.7. The overall prevalence of HF in the included sample was 31%, with morphology-specific prevalence of 30% in those with single RV and 39% in those with single LV.
A scatter plot of BNP versus Ross score was used to assess the relationship between the biomarker and clinical status. After log2 transformation, BNP demonstrated a roughly linear relationship with Ross score (). A doubling of BNP was associated with an odds ratio for HF of 2.20 (95% CI 1.36–3.55, p=0.001) when side and stage were held constant. Utilizing univariate logistic regression, and stratifying the subjects into morphology and stage, there was a statistically significant correlation between BNP and HF in the single RV group with an odds ratio for HF of 2.03 (95% CI 1.24–3.31, p=0.005), and in all patients at stage II with an odds ratio of 1.5 (95% CI 1.03–2.16, p=0.03). In SV patients at stage I, the correlation between BNP and HF approached statistical significance with an odds ratio of 2.92 (95% CI 0.90–9.51 p=0.08) as did the correlation between BNP and HF among patients at stage III (odds ratio 3.51 [95% CI 0.75–16.5, p=0.11]) We cannot preclude the possibility that our failure to detect a significant association in these two groups may be due to insufficient power. There was no statistically significant correlation between BNP and HF in the single LV group (odds ratio 1.94 [95% CI 0.89–4.22, p=0.095]) although this too was suggestive of an association.
Evaluation of BNP versus Ross scores
ROC curve analysis of BNP data for the entire cohort met our pre-specified threshold of ≥75% area contained by the curve (). ROC analysis stratified by ventricular morphology demonstrated that the right SV group exceeded the threshold of ≥75% (c-statistic= 81%), however, left SV patients did not (c-statistic=69%) (). ROC analysis of right SV patients stratified by stage of surgical palliation demonstrated that patients at all 3 stages of palliation exceeded the usefulness threshold (c-statistic=82%, 77%, and 92%, for stage I, II, and III, respectively) (). Together, logistic regression and ROC curve analysis supported the conclusion that BNP is a useful biomarker of HF (Ross score ≥3) for single RV patients at all stages of palliation.
Receiver Operator Characteristic curves for BNP by Patient Group
We determined sensitivity and specificity for various BNP levels as markers of HF for the entire cohort and groups stratified by ventricular morphology and stage of surgical palliation that demonstrated c-statistic values ≥75%. A cut point of ≥45 pg/ml correctly classified 75% of SV patients for all patients considered in total, with both sensitivity (62%) and specificity (81%). Right SV morphology also demonstrated a cut point of ≥45 pg/ml that correctly classified 76% of patients (sensitivity 71%, specificity 77%). When stratified by stage, a cut point of 45 pg/ml correctly classified 73% of stage I patients (sensitivity 100%, specificity 60%), 73% of stage II patients (sensitivity 57%, specificity 80%), and 88% of stage III patients (sensitivity 100%, specificity 83%). This analysis supports a cut point of 45 pg/ml for predicting HF in single RV patients, regardless of stage of palliation.
BNP is rapidly cleared from the circulation through a process of cell surface binding and internalization, with a half-life of 20 minutes8
. The BNP threshold for predicting HF in this and our previous pilot study was significantly lower than for predicting HF in both pediatric and adult patients with biventricular hearts. Since it is possible that SV patients produce less BNP precursor (pre-BNP), we wanted to determine whether NT-proBNP, another product of pre-BNP processing cleared by renal excretion with a longer half-life (1–2 hours)8
, might be more useful for predicting HF in SV patients. In addition, NT-proBNP is stable at room temperature for 72 hours, while BNP is stable for 24 hours, making NT-proBNP a more stable biomarker target9
Scatter plot analysis of log2 NT-proBNP versus Ross score demonstrated a roughly linear relationship (). Multivariate logistic regression of plasma NT-proBNP levels demonstrated that a doubling of NT-proBNP was associated with an odds ratio for HF of 1.92 (95% CI 1.17–3.14, p=0.009) when side and stage are held constant. Univariate logistic regression of groups stratified by ventricular morphology and stage showed a statistically significant correlation between NT-proBNP and HF in the single RV group, with an odds ratio for HF of 1.77 (95% CI 1.08–2.88, p=0.02). In SV patients at stage I, the correlation between NT-proBNP and HF approached statistical significance with an odds ratio of 2.63 (95% CI 0.96–7.26 p=0.06). The correlations between BNP and HF in patients at stages II and III were not significant, with few patients at stage III. There was no statistically significant correlation between NT-proBNP and HF in the single LV group (odds ratio 1.19 [95% CI 0.69–2.05, p=0.53]).
Evaluation of NT-proBNP versus Ross scores
ROC curve analysis of NT-proBNP data for the entire study cohort gave a c-statistic of 63%, which did not exceed our pre-specified threshold of ≥75% area contained by the ROC curve (). ROC analysis stratified by ventricular morphology demonstrated that the right SV group exceeded the threshold of ≥75% (c-statistic=78%), however, left SV group did not (c-statistic=57%) (). ROC analysis stratified by stage of palliation demonstrated that patients at stages I and III exceeded the usefulness threshold (c-statistic=85% and 100%, respectively), however, those at stage II did not (c-statistic=46%) ().
Receiver Operator Characteristic curves for NT-proBNP by Patient Group
The test characteristics for various plasma levels of NT-proBNP as markers of HF were examined for the single RV group and for those at stages I and III (those that demonstrated c-statistic values ≥75%). For all patients with single RV, a cut point of ≥1100 pg/ml showed both sensitivity (78%) and specificity (83%), thus correctly classifying 82% of patients. Interestingly, when single RV patients were stratified by stage, the threshold for classifying HF decreased with stage. A cut point of 1900 pg/ml was needed to correctly classify 80% of patients at stage I, with sensitivity 100% and specificity 70%, while this decreased to 1100 pg/ml for stage II patients, and 300 pg/ml for stage III patients (sensitivity 100%, specificity 75%). Logistic regression, ROC curve analysis, and evaluation of test characteristics suggest that NT-proBNP can be a useful biomarker of HF (Ross score ≥3) in single RV patients at all stages of palliation.