We approached 39 single ventricle children meeting the inclusion criteria for this study and presenting to the UCSF Pediatric Heart Center between February 2007 and June 2008 for enrollment. Four (10%) declined, and of the 35 remaining subjects, 6 (15%) were subsequently excluded from analysis because of missing data. Thus, 29 children were studied (). We obtained blood samples at the time of cardiac catheterization in 26 of 29 (90%). Of the remaining 3, 2 were drawn immediately prior to cardiac surgery, and 1 was drawn in the course of medical admission for HF.
Of 29 children, 22 (76%) had systemic non-left ventricles. The median Ross score in the overall sample was 2, with a standard deviation of 2.3 Ross units. Among the 20 children free of clinical HF, the median Ross score was 1, with a standard deviation of 0.8 Ross units. Among the 9 children with clinical HF (Ross score >2), the median score was 4, with a standard deviation of 2.3 Ross units and a range of 3 to 9. The overall prevalence of HF in our included sample was 31%.
Only 4 of the 29 children (14%) were taking no cardiac medications at the time of the study. Diuretics were the most common cardiac medications (69%), followed by angiotensin converting enzyme inhibitors (55%). Two children were on inotropic therapy at the time of the study. Not surprisingly, these children were both in HF. Digoxin use was more prevalent in children free of HF.
shows scatter plots of our candidate biomarkers versus the Ross HF score. The scatter plots provide a visual appraisal of the linearity of the relationship between the candidate biomarkers and the Ross HF score. Notably, after log2 transformation, BNP demonstrated a roughly linear relationship with the Ross score. A one unit increase on this scale corresponds to a doubling on the linear scale. Simple linear regression analysis on the transformed variable revealed that a doubling of plasma BNP was associated with a 0.7 unit increase in Ross score (95% CI 0.4, 1.0; p<0.001). Component-plus-residual plots demonstrated that the assumption of linearity had not been violated, and the residuals were normally distributed, suggesting that the linear model is a good fit for these data. Although DFBETA statistics revealed two potentially influential points, exclusion of these points preserved the linear association: a doubling of plasma BNP was still associated with a 0.4 unit increase in Ross score, with the 95% CI encompassing zero (95% CI −0.05, 0.8; p=0.082).
Scatter plots of biomarkers versus Ross scores
For hsCRP, a 0.2 mg/L increase was associated with a 0.8 unit increase in Ross score (95% CI −0.03, 1.6; p=0.058), however, DFBETA statistics and component-plus-residual plots revealed three highly influential points. Exclusion of these points abolished the association. Neither ET1 (20 unit increase associated with 0.07 decrease in Ross score, 95% CI (−0.3, 0.15), p=0.50) nor cTnI (0.1 microgram/L increase associated with a 0.07 unit increase in Ross score, 95% CI (−0.2, 0.3), p=0.61) was associated with increased Ross score in our sample. Exclusion of influential data points for these assays did not alter the results.
Using single-predictor logistic regression, a doubling of plasma BNP was associated with an odds ratio for HF of 2.17 (95% CI 1.10, 4.3; p=0.026). A 0.2 mg/L increase in hsCRP was associated with an odds ratio for HF of 1.76 (95% CI 0.69, 4.5; p=0.23), suggesting that elevations in hsCRP may be associated with increased odds of HF, but this result had poor precision (31–350% increase in odds). Likewise, neither cTnI (OR=0.72, 95% CI 0.05, 10; p=0.81) nor ET1 (OR=0.86, 95% CI 0.54, 1.36; p=0.52) was associated with clinically relevant alterations in the odds of HF.
We generated Receiver Operator Characteristic (ROC) curves, which plot a test's true positives against its false positives, for each of the four potential biomarkers (). Only one protein, BNP, exceeded our pre-specified threshold of ≥75% area contained by the ROC curve. The area under the curve for the plot of BNP against HF was 80.3%. cTnI failed to distinguish children in HF from those free of failure. Similarly, hsCRP and ET1 demonstrated relatively poor sensitivities and specificities for HF throughout their curves.
Receiver Operator Characteristic curves for BNP, cTnI, hsCRP and ET1
The test characteristics for various plasma levels of BNP as markers of current clinical HF were determined. A cut point of ≥30 pg/mL showed both sensitivity (88.9%; 95% CI 51.8% – 99.7%) and specificity (75%; 95% CI 50.9% – 91.3%). The wide 95% confidence intervals demonstrate that our results are consistent with sensitivities as low as 51.8%. However, our sample consists of predominantly mild HF, and as such, we would expect this to be a relatively difficult population to detect. Given the roughly linear relationship between BNP doubling and raw Ross score, we expect BNP values to be higher in moderate than in mild failure. Thus, lowering the cut point further would increase the false positive rate without increasing detection of moderate HF. As such, we believe a cut point of ≥30 pg/mL presents the best threshold for identifying HF in young children with single ventricle physiology.