Components of variability in the systolic blood pressures of preschool children.

The term "horse-racing effect" refers to a positive correlation between the slope at which blood pressure increases with age and blood pressure level at a baseline. Previous studies have reported such an effect in adults, while studies in children have found a negative correlation (termed "the Jenss effect"). These studies used analytic methods in which it was assumed that the blood pressure slopes were constant or the blood pressure profiles were linear. In this study, the authors used a components-of-variance approach that did not require this assumption in order to model serial blood pressure measurements made in 216 US preschool children (mean age at first analyzable blood pressure observation, 58.9 months) at 6-13 visits over a 3-year period (1986-1989). Measurements were made using an automated blood pressure monitor. Values from the second and third measures at each visit were averaged to obtain each observation. Data from 2,203 blood pressure observations were available for analysis. For the full group, over the mean period of observation of 21.1 months, the mean rate of increase was 3.45 mmHg/year for systolic blood pressure and 0.06 mmHg/year for diastolic blood pressure. In the authors' model, 33 percent of the total marginal variability in systolic blood pressure was attributed to random error (visit-to-visit variability); average (constant) subject-specific difference from the population slope accounted for 42 percent, and a nonlinear component of variability around the subject-specific average accounted for the remaining 25 percent. All three components were statistically significant. Models which assumed that the slopes were constant did not fit these data, and fitting these models to the data led to an artifactual negative correlation between subject-specific slopes and intercepts. An implication of this is that the concepts of "horse-racing" and "the Jenss phenomenon," which have been operationalized by testing the covariance of the subject-specific slopes and intercepts using models which assume that the blood pressure slopes are constant over time, should not be applied to data that contain this nonlinear component of variance.