We were interested to read the paper by Ryuno et al.1 that was published in Hypertension Research in July 2016. The authors aimed to assess the association between hypertension and the cognitive functioning in participants aged 70–80 years in the Japanese study of Septuagenarians, Octogenarians and Nonagenarians Investigation with Centenarians (SONIC). The result of multiple regression analyses demonstrated that a 1 unit increase in systolic blood pressure (SBP) yielded predicted decreases of −0.10 or −0.05 in the mean for the Japanese version of the Montreal Cognitive Assessment (MoCA-J) in participants who were 70 years old with uncontrolled or controlled blood pressure, respectively. These results in participants who were 80 years old with uncontrolled or controlled blood pressure were −0.05 or 0.04, respectively.

Although interesting data were presented on the association between hypertension and cognitive functioning in elderly people, some methodological and statistical issues should be considered to avoid misinterpretation. In the Ryuno et al.1 study, clinical judgments were overlooked in the interpretation of the regression coefficients because clinically a −0.10 and −0.05 decrease in the mean of MoCA-J is negligible. When interpreting results, clinical importance carries more weight than statistical significance. A larger effect size, lower variation in the variables studied in the study population and, in particular, a larger sample size can easily lead to a significant P-value. It is crucial to emphasize that the predicted regression coefficients in the Ryuno et al.1 study should be interpreted with caution because unbiased predictions cannot be guaranteed by cross-sectional studies; we also need to internally or externally validate the predictions.2

Moreover, reporting only P-values for regression coefficients is a common mistake. As is the case in previously published studies about the association of hypertension and cognitive impairment in elderly population,3 the authors should report confidence intervals for readers so that they can understand the magnitude and direction of the association, and the random variability of the regression coefficients.4