We read with great interest the article ‘Left atrial dimension is related to blood pressure variability in newly diagnosed untreated hypertensive patients’ by Cipollini et al.,1 which was published in Hypertension Research in March 2016. The authors argue that 1-unit increases in mean 24 h systolic blood pressure (BP) variability and mean 24 h diastolic BP variability are predictive of increases of 0.018 and 0.022 in the left atrial posteroanterior diameter index for height (LADi) and the left ventricular mass index for body surface area (LVMi), respectively, after adjusting for sex, age, body mass index, heart rate, diastolic function and estimated glomerular filtration rate.1

This interpretation appears to be correct; however, it is important to emphasize that differences of 0.018 and 0.022 in LADi and LVMi, respectively, are clinically negligible and even unimportant. Some factors, including large sample sizes, small variations in the studied variable in study populations and large mean differences, can lead to significant P-values.2 As a general rule, clinical importance carries more weight than statistical significance.

The authors note that multiple linear regression models were used to estimate regression coefficients. However, it seems that confounders were not efficiently addressed in the statistical methodology. Investigators usually use statistical methods, such as stepwise selection and/or prior knowledge, to adjust for confounders in multivariable analyses. There are advantages and disadvantages to using these methods.3 Here, it is unclear which method was used to address confounders in the multivariable analyses of the study by Cipollini F et al.1

The take-home message for readers is that they should distinguish between statistical significance and clinical importance. Using stepwise selection or definitive prior knowledge acquired from previous empirical studies can guarantee that the most relevant variables will be included in the final models in multivariate analyses.