Abstract
Random fluctuations in blood pressure complicate the design and evaluation of screening and treatment programs. The coefficient of reliability (
G) indicates the severity of such fluctuations and is used to correct for their effects. This coefficient is defined as the ratio of variance of individuals' true pressures to the variance of their measured pressures under some specified procedure. To estimate
G, a statistical model was developed in which true blood pressure drifts according to a type of diffusion process. Correlations between blood pressure measurements at different examinations up to 18 yr apart in the Framingham Study showed a pattern consistent with the model. For diastolic pressure in males, for example, the two-parameter model explained 94% of the variance among 45 correlations and estimated
G for a single reading as 0.67. The model implies that elevations in true blood pressure may decline spontaneously over several years in some hypertensives. The model also shows how the reliability of an average of several readings increases (and random error decreases) with the number and type of replications. For example, 2 measurements at separate visits generate the same reliability (0.80 for diastolic BP inmales) as 6 readings at the same visit. This coefficient can be used to decide how many measurements of blood pressure should be obtained in screening, treatment, and compliance programs, and to refine before-and-after evaluations of such programs by correcting the statistical artifact of regression to the mean.