Spotlights on Ambulatory Measures of Arterial Stiffness

To the Editor: Arterial stiffness is an independent predictor of cardiovascular morbidity and mortality.1 It should come as no surprise that researchers are currently looking for measures of arterial stiffness that are easy to record even under ambulatory conditions. In this Journal, Gosse and colleagues2 recently compared three indexes, derived from individual ambulatory blood pressure (BP) recordings, namely, the QRS-Korotkoff delay index standardized to a systolic BP of 100 mm Hg and a heart rate of 60 beats per min (QKD100–60), 2 the ambulatory arterial stiffness index (AASI),3 and the slope of pulse pressure on mean BP (PP/MBP). Gosse and colleagues confirmed that AASI provides an indirect estimate of arterial stiffness, but they also proposed that QKD100–60 might be a better index of arterial stiffness than AASI. To date, several cross-sectional studies4 and three prospective cohort studies5–7 demonstrated association of AASI either with signs of target organ damage in never-treated hypertensive patients,4 or with the incidence of cardiovascular mortality and morbidity.5–7 AASI is particularly predictive of stroke,5–7 even at levels of BP within the normotensive range.5,6 When adjusted for PP5–7 or aortic pulse wave velocity,8 AASI retained its predictive value, in particular for stroke.8 We believe that over-interpretation of weak results and methodological issues go against Gosse’s proposition that the QKD100–60 outperforms AASI in the prediction of cardiovascular events.2 The first argument of the French investigators in favor of the QKD100–60 was the higher reproducibility of this index.2 In 38 volunteers undergoing repeat ambulatory recordings within 2 weeks, the coefficient of variability was 4% for QKD100–60, 25% for AASI, and 61% for PP/MBP.2 The authors did not present any test statistics or associated P-values for the comparison of the coefficients of variability between the three ambulatory indexes of arterial stiffness. Moreover, the coefficient of variability is the s.d. of the differences between paired measurements (SDD) divided by the mean of all measurements. It does not account for the possible range of biological variation in the ambulatory measures of arterial stiffness. We, therefore, proposed to express the SDD as a percentage of near maximal variation in a measurement as given by four times the s.d.9 Had reproducibility been expressed in this way with the application of a proper test statistic, the conclusion of Gosse’s reproducibility study2 might have been quite different. The second argument of the French investigators hinged on the prediction of cardiovascular events in 440 hypertensive patients followed up for 7 years. In unadjusted analyses, all three ambulatory indexes of arterial stiffness predicted outcome (P < 0.001).2 However, with adjustments applied for age and 24-h PP, only QKD100–60 remained prognostically significant. The correlation coefficient with the 24-h PP was –0.51 for QKD100–60, +0.36 for AASI, and +0.33 for PP/MBP.2 We suspect that the use of highly correlated variables, all reflecting arterial stiffness, might result in unstable Cox models. Gosse and colleagues did not account for established risk factors, such as sex, smoking, body mass index, serum cholesterol, or diabetes mellitus. In contrast to the authors’ opinion,2 62 cardiovascular events, including 13 deaths, were sufficient to allow adjustment for this number of covariates. Furthermore, the association of cardiovascular risk with the QKD100–60 and AASI are continuous functions. Nonetheless, the authors chose to evaluate the predictive value of these indexes across tertiles of their respective distributions. The cardiovascular endpoints were unevenly distributed among the bottom, middle, and top tertiles of QKD100–60 (3, 19, and 40, respectively) and AASI (10, 17, and 35). The lower number of events in the bottom tertile of the QKD100–60 than AASI favors the former index. However, the survival function plotted in Figure 2 of Gosse’s report2 shows the incidence of at least five cardiovascular events in the lowest tertile of QKD100–60. Thus, there is inconsistency in the distribution of the number of events reported in tabular and graphical format. In age-adjusted analyses with the bottom tertile as the reference, the relative risks (RRs) of a cardiovascular event in the middle and top tertiles were 4.6 (95% confidence interval, 1.3 – 15.7) and 6.9 (2.0 – 23.0) for the QKD100–60; the corresponding RRs for AASI were 1.5 (0.7 – 3.3) and 2.8 (1.3 – 5.8), respectively. Unfortunately, Gosse and colleagues did not report the RRs for AASI with adjustments applied for both age and the 24-h PP.2 For the point made in Gosse’s report,2 it is less relevant whether RRs are significant. What the authors should have shown is that estimates of RRs differed significantly between QKD100–60 and AASI. Point estimates of the RRs for AASI fall within the confidence intervals for the corresponding RRs for the QKD100–60. This observation strongly suggests that QKD100–60 and AASI were equally predictive of cardiovascular complications. We have additional concerns about the validity of the conclusions of the French investigators. From a mathematical point of view, the correlation between PP and MBP is spurious and must be positive, because PP is used in the calculation of MBP. We doubt that standardizing QKD to a systolic BP of 100 mm Hg is meaningful in hypertensive patients, whose 24-h systolic BP averaged 133 mm Hg. Furthermore, Gosse and colleagues did not specify the nature of the cardiovascular complications and the procedures put in place for their adjudication. This is not without importance, because in our hands AASI is most predictive of stroke, not including transient ischemic attacks.5–8 Finally, practitioners will also consider the practicability of the various ambulatory measures of arterial stiffness. The QKD approach requires the [Q1]