Reliability of longitudinal ultrasonographic measurements of carotid intimal-medial thicknesses. Asymptomatic Carotid Artery Progression Study Research Group

Abstract

Background and purpose: Serial ultrasonic B-mode measurements of intimal-medial thickness (IMT) of the carotid artery are commonly used as surrogates for describing atherosclerosis progression. This report describes the longitudinal reliability of IMT measurement during a multicenter clinical trial, quantifies the error attributable to differences among readers, and discusses how studies can be efficiently designed.

Methods: Serial B-mode measurements of carotid IMT from the 3-year Asymptomatic Carotid Artery Progression Study (ACAPS; formerly Asymptomatic Carotid Artery Plaque Study) were used to estimate the contributions to longitudinal measurement error of systematic reader effects, nonvisualization, and nonsystematic error and to describe the distribution of "true" progression rates that underlie the observed data. Variance components were estimated from random-effects models fitted to outcome measures formed by averaging IMTs from different sets of carotid artery walls. These were used to contrast the relative efficiency of study designs.

Results: Of the total variance of measured IMT, 11% was attributable to systematic differences among readers. Nonvisualization contributed less than 7%. Thus, the predominant source of error was unaccounted for (ie, random error or "noise," which in our analyses included any drift, nonlinearity, and sonographer differences). For studies with measurement protocols similar to ACAPS, follow-up times of 2 years or more are desirable for describing the mean progression rates of cohorts, and of 6 years or more for categorizing progression within individuals. In 3-year studies, sample sizes as low as 237 provide 90% statistical power for detecting risk factors that have correlations with IMT progression of .50 or greater.

Conclusions: The ACAPS measurement protocol provided highly reliable serial IMT data. Moderate-sized multicenter studies using B-mode outcomes are feasible.