To the Editor Jia et al 1 investigated the associations between 6-year change in N-terminal pro-B-type natriuretic peptide (NT-proBNP) and the risk of heart failure (HF) hospitalization and all-cause death among adults without prior HF, using the data from the Atherosclerosis Risk in Communities (ARIC) study. The investigators provide a successful example of using large-scale observational data with long-term followup. Along with a previous study on elderly patients without HF, 2 the current study extends our knowledge of serial NT-proBNP measurements in the patients on the spectrum from clinical HF to pre-HF. This promising approach may aid in stratifying patients for precise primary HF prevention.Although we agree that serial NT-proBNP provides additional information beyond the baseline measurement, in our opinion, the authors' emphasis on the role that the change of NT-proBNP plays in prognosis prediction is not substantiated by their data and may be misleading. First, based on Table 2 and eFigure 2 in Supplement 1, 1 patients in the high/high category were of the highest probability of events, followed by those in low/high, high/low, and low/low categories. 1 However, patients in the low/high category, but not the high/high category, are expected to have the greatest increase in NT-proBNP levels. This contradicts their conclusion on the association between the change of NT-proBNP and events. Second, the associations between the change in NT-proBNP and HF hospitalization and all-cause mortality were substantially attenuated, with most of them neutralized, after adjusting for the latest NT-proBNP level, as shown in Table 3, 1 leaving only the categorized model predicting all-cause death barely significant. The above 2 points argue against the conclusion that change of NT-proBNP matters and instead support the hypothesis that the latest measurement is of more predictive value, as raised in a recent study by Wolsk et al. 3 Another concern in this study and in other studies on serial NT-proBNP measurement is the intra-individual variability and measurement error, which can account for up to 40% of its change in patients without HF and up to 60% in patients with stable HF per previous reports. 4,5 This complicates the distinction of the true change from the noise and may be one reason why the continuous models demonstrated nonsignificant results after adjustment. 1 In the current study, the investigators set the cutoff as a 25% change of NT-proBNP in the categorized model. Additional information on how this cutoff was established would further help readers translate the current results into their clinical practice.