Unlike, the relatively slow evolving trends (over days or weeks) in plasma BNP observed in HFrEF patients, some HFpEF patients exhibited rapid spikes and falls in plasma BNP occurring at intervals of hours to one or 2 two days- without clinical acute decompensated heart failure.^8-10 Some have suggested that this pattern makes NT-proBNP an unreliable marker of clinical status in HFpEF, because the levels of NT-proBNP are “spontaneously” variable in the absence of any discernible clinical change in HFpEF patients.⁸

We clearly demonstrate that similar to HFrEF patients, up-trending NT-proBNP levels are associated with clinical deterioration in HFpEF patients. (Figure 2). We disprove the hypothesis that rising NT-proBNP levels are somehow “decoupled” from clinical deterioration in the HFpEF due to kinetics of the molecule in this condition, as some prior studies have postulated.⁴ Using Cox regression that treats NT-proBNP as a time dependent covariate, we demonstrate that for both subsets of HF (HFpEF and HFrEF): a two-fold increase in the NT-proBNP levels (above normal) is associated with an increased the HR of death within 6 months by approximately 45-50% (Table 3).

There are several caveats to consider when interpreting the data. Our data should not be interpreted to prove that trends in NT-proBNP can be used to predict the clinical trajectory of HF. It is likely that patients with progressively rising NT-proBNP levels had certain subtypes of disease that was more likely to be refractory to treatment and thus the HF was more likely to progress. The comorbidity profiles, albumin, creatinine and the Charlson co-morbidity score were not significantly different across the groups at baseline. (Table 1) We did not match the cohorts at baseline in terms of functional status or other variables- it is likely that patients with up-trending NT-proBNP levels had a poorer functional status (NYHA class) at the time of entry into the study. Because our study was not prospective or randomized, the frequency and the period at which serial NT-proBNP measurements were obtained was not fixed and were obtained at varying periods by the patient’s physicians — we used statistical methods that account for this. Although multivariable statistical models were used to adjust for heterogeneity between groups in this observational study, residual unmeasured confounding factors may be present.