Prognostic relevance of visit‐to‐visit office blood pressure variability in Systolic Blood Pressure Intervention Trial: Same data, different conclusions?

We read with interest the article by Mezue et al,1 which reports a positive association between visit-to-visit office blood pressure (BP) variability and cardiovascular outcomes using data from Systolic Blood Pressure Intervention Trial (SPRINT). We published an article on the same topic last year, also using data from SPRINT,2 and reported no association of office BP variability with cardiovascular outcomes. Because the main conclusions of our two articles appear to be contradictory, and because differences between our two articles were not discussed by Mezue et al, we would like to take this opportunity to highlight key factors that may have contributed to our disparate conclusions. First, we focused on systolic BP variability, defined as the coefficient of variation, and found no significant association of systolic BP variability with the primary cardiovascular composite outcome, hospitalized heart failure, or stroke. Mezue et al examined the association of systolic BP variability, diastolic BP variability and pulse pressure variability (also defined as the coefficient of variation) with the primary cardiovascular composite outcome and hypoperfusion-related adverse events, but their presentation of results was incomplete; for cardiovascular events, they reported results only for diastolic BP variability. We are left to assume that, despite the title of their article stating that “Blood Pressure Variability Predicts…Cardiovascular Outcomes in SPRINT,” Mezue et al found no significant association of systolic BP variability (or pulse pressure variability) with the primary cardiovascular composite outcome. Second, we assessed BP variability with a uniform number of BP measurements during a period relatively free of active BP medication titration or other interventions, as these extrinsic factors can themselves influence BP variability.3 In contrast, Mezue et al used all available measurements after a “landmark date” (defined as the date of the first recorded BP that reached treatment systolic BP target). The Authors provided no information on the number of BP measurements and do not appear to have included this variable in their regression models. Moreover, because some of the measurements included by Mezue et al may have been obtained after a non-fatal event, the BP variability estimate from the regression model may reflect the effects of a non-fatal event on BP variability, rather than the effects of BP variability on the event. Third, we selected variables for inclusion in the models a priori based on known confounders, which included age, sex, race, randomized group, baseline chronic kidney disease, coronary artery disease, smoking status, mean systolic BP, and the number of antihypertensive medication classes received at the 3-month study visit. In our analysis, we confirmed that many of these variables are significantly associated with BP variability. Mezue et al present unadjusted Kaplan-Meier survival curves in figure 2, and did not include information on which variables comprised their final adjusted regression models. Understanding how the authors adjusted for potential confounders is of particular importance regarding diastolic BP variability, given that lower diastolic BP in SPRINT was associated with not only higher risks of adverse outcomes but also higher absolute risk reductions for measured cardiovascular outcomes.4 In summary, we appreciate the additional analyses on BP variability using SPRINT data reported by Mezue et al, but we believe that the Authors should provide clarification of the aforementioned methodological issues so that the scientific community can appropriately interpret their findings. None of the authors has any conflict of interest related to this document.