Response to Letter Regarding Article, “Serial Biomarker Measurements in Ambulatory Patients With Chronic Heart Failure: The Importance of Change Over Time”
We thank Drs Ballo, Bocelli, and Mondillo for their interest in our recent article.1 They raise an interesting issue relating to the finding in our study cohort that cardiac resynchronization therapy (CRT) was associated with adverse effects. They point out correctly that in several reports, CRT and brain natriuretic peptide (BNP) are associated with clinical benefit and that over time, as in our longitudinal study, a reduction in BNP levels occurs (CARE-HF).2 Thus, they express a concern that this known collinearity of BNP and CRT, albeit for different outcomes, could have confounded the multivariate model relating baseline risk factors to the outcome of death and cardiac transplantation in this patient cohort. This is a reasonable possibility to explore, as we have done.
It is important to keep in mind the context of our findings and to appreciate that only 27 patients (14% of the study cohort) received CRT, making a more comprehensive analysis of this subset problematic. In addition, many of these patients received CRT early after its introduction, given the relatively long timeline of our study. Thus, some of these patients likely were those with more severe disease that characterized the patients selected at that time for this type of therapy. However, statistical confounding is also possible. To address the concern that a strong association between the independent variables of BNP elevation and CRT (collinearity) could have affected the resulting coefficient in the model (Table 3 in our article1), we re-analyzed the multivariate model without the CRT variable. If over-adjustment was present due to collinearity, it would be expected that the hazard ratio for an elevated BNP level would change significantly and likely become a more prominent predictive factor. However, re-analysis showed the hazard ratio for BNP to be 1.58 (P=0.114), which is only a minor change from the hazard ratio of 1.53 (P=0.146) with CRT included in the model. Accordingly, although the findings from this multivariate model of baseline risk factors are unexpected, given the present state of the art, they appear to be valid within the timelines and context of this patient cohort. Obviously in this study,1 as with all observational data, caution is needed in attributing causation to these correlations.
Dr Miller has received a research grant from Dade Behring. Dr Jaffe has received a research grant from Dade Behring and has served as a consultant to Roche Diagnostics and Dade Behring. The other authors report no conflicts.