Reproducibility of Ultrasonographically Determined Intima-Media Thickness Is Dependent on Arterial Wall Thickness
The Tromsø Study
Background and Purpose We compared the reproducibility of B-mode ultrasonographic measurements of intima-media thickness (IMT) in various segments of the right carotid artery and examined whether measurement error was associated with IMT or cardiovascular risk factor levels.
Methods In 1994/1995 a total of 6676 participants in the Tromsø Study underwent ultrasound examination of common carotid artery IMT. Reproducibility of measurements was assessed by inviting 111 participants to a second ultrasound scan within 3 weeks of the first scan. On each occasion the subjects were examined by three sonographers.
Results The mean between-observer absolute differences in IMT in the far wall of the bifurcation and the near and far walls of the common carotid artery were 0.15, 0.10, and 0.08 mm, respectively. The corresponding within-observer differences were 0.15, 0.10, and 0.06 mm, respectively. Approximately 70% to 80% of total measurement variability was due to differences among sonographers; the rest was attributable to within-reader variability. Measurement error increased significantly with increasing IMT: the increase was more than twofold over the range of measurements. Cardiovascular risk factor levels were not associated with measurement variability when we controlled for IMT.
Conclusions We conclude that B-mode ultrasound provides reproducible estimates of the IMT in both the near and far walls of the carotid artery. Although measurement error is generally small, it increases proportionally with the level of IMT.
Measurements of carotid artery IMT by use of ultrasound are increasingly used to assess the extent and severity of atherosclerosis in the carotid arteries,1 2 3 4 and several lines of evidence suggest that ultrasonographically determined IMT is a valid way to study early atherosclerosis.5 6 7 8 9 Ultrasound is preferable to arteriography because it is noninvasive, carries no risk for the examined subject, and can detect atherosclerosis as an increase in arterial wall thickness before a reduction in lumen diameter occurs. The usefulness of ultrasound therefore appears to depend critically on the precision with which the IMT can be measured. A number of factors contribute to measurement reliability, including ultrasound instrumentation and reader performance, but the dependence of B-mode image quality on interrogation angle and fine adjustment of instrument controls makes the role of the sonographer crucial to the measurement process.
Previous studies have reported that the IMT of the far wall of the CCA can be measured with mean absolute differences on repeated measurements ranging from 0.09 to 0.18 mm, and there is a general agreement in the literature that ultrasound provides highly reproducible data.10 11 12 13 However, reproducibility studies on the ultrasound method have mainly looked at average performance and with a few exceptions have not examined the possibility of patient factors being associated with measurement error. Results from the Rotterdam Study indicate that the scatter of the differences increases with increasing thickness of the carotid far wall intima-media complex.14 Reproducibility of ultrasonography may therefore critically depend on the characteristics of the study population.
The Tromsø Study was started in 1974 and is a prospective population-based study with repeated assessment of cardiovascular risk factor levels in the population.15 16 17 In 1994/1995 ultrasonography of the carotid artery was performed on 6720 subjects. High-quality images of the CCA and the carotid bifurcation were obtained in 6676 (99.3%) and 6314 subjects (94.0%), respectively. To determine the effect of sonographer and reader performance on IMT measurements, we assessed both within- and between-observer reproducibility in different locations of the carotid artery. We also examined the determinants of measurement variability. All images were analyzed in Gothenburg (Chalmers University of Technology) with the aid of a newly developed computerized automatic reading system developed at the Wallenberg Laboratory for Cardiovascular Research.18 19
Subjects and Methods
Study Design and Subjects
The Tromsø study was started in 1974 and is a single center prospective follow-up study of inhabitants in the municipality of Tromsø, Norway. The aims of the study are to investigate, by means of epidemiological, clinical, and basic research, determinants of chronic diseases to assess etiologic significance and to investigate potentially modifiable determinants that may be developed into preventive or therapeutic strategies. The main focus is on cardiovascular diseases. The study design includes repeated population health surveys to which total birth cohorts and random samples are invited. The study was approved by the regional board of research ethics, and each subject gave informed consent.
The fourth survey of the Tromsø population started in September 1994 and was completed in October 1995. The survey was conducted by the University of Tromsø in cooperation with the National Health Screening Service and comprised two screening visits with an interval of 4 to 12 weeks. All inhabitants older than 24 years were invited to the first visit, and 27161 subjects (77% of the eligible population) participated. A protocol similar to that used in the previous surveys in this population15 16 17 and the Norwegian county studies20 was followed. The examination included standardized measurements of height, weight, blood pressure, nonfasting serum lipids, serum calcium, γ-glutamyltransferase, hemoglobin and blood cell counts, and a 20-second ECG of lead I. Two questionnaires covered previous and present diseases and symptoms, use of drugs, lifestyle (physical activity, smoking, alcohol intake) and dietary habits, and socioeconomic situation. All subjects aged 55 to 74 and random 5% to 10% samples in the other age groups were invited to the second visit. A total of 6891 subjects (98% of those who came to the first visit and were eligible for the second visit) attended. The second visit comprised ultrasonographic examination of the carotid artery and the abdominal aorta, echocardiography, a 12-lead resting ECG, a 90-second eight-lead rhythm ECG during standardized deep breathing, measurements of bone density, body fat composition, waist and hip circumference, sitting and standing blood pressures, and urine and blood sampling.
The Reproducibility Study
The reproducibility study was designed to study variability in IMT measurements between sonographers (different sonographers on the same occasion), within sonographers (same sonographer on two separate occasions), and within reader (same reader on two separate occasions) in the beginning (first reproducibility study) and the end (second reproducibility study) of the survey period. We use the term “observer” variability to describe the sum of measurement error due to sonographers (ie, the scanning process) and the reader (ie, the off-line reading of images). A total of 111 subjects were invited to the two reproducibility studies. All of them participated, but some met only once. The total number of subjects with paired measurements was 101 (75 in the first and 26 in the second reproducibility study).
The subjects were examined by the same three sonographers on two occasions with an interval of 1 week (weeks 10 and 11) in the first and 3 weeks (weeks 37 and 40) in the second reproducibility study. The sonographers were blinded to each other’s results and to medical information about the participants. One of the sonographers was a neurologist with 10 years’ experience in ultrasound examination of the carotid artery, the second was a physician, and the third was a specially trained technician. A 2-month training session during which the neurologist instructed the other sonographers according to a protocol described by Wendelhag et al10 was completed before the survey started. In addition, the first 200 survey examinations conducted by the physician and the technician were supervised by the neurologist.
To assess the within-reader variability, images from the first scanning of the 80 subjects in our first reproducibility study were reread by the same reader who performed the initial reading. He was blinded to the participants’ characteristics and the results of previous readings. A separate comparison study has been performed at the Wallenberg Laboratory to compare this reader with other experienced readers. The coefficients of variation for IMT measurements in the far walls of the CCA and the carotid bifurcation were 1.3% and 2.3%, respectively, indicating satisfactory agreement.
IMT measurements of the right carotid artery were obtained with the use of a high-resolution ultrasound Acuson 128 XP/10 c scanner equipped with a linear transducer with 7 MHz in B-mode and 5 MHz in pulsed-Doppler mode. The system provides an optimal axial resolution of approximately 0.20 mm. The subjects were examined in the supine position with the head turned approximately 45 degrees to the left. The angle of examination was mainly antero-oblique and lateral, occasionally postero-oblique, depending on which angle gave optimal visibility of the thickest part of the arterial wall. The scanning procedure lasted 15 to 20 minutes.
The boundaries of the different layers of the carotid artery wall can be visualized with high-resolution B-mode ultrasonography. On a longitudinal ultrasound image of the carotid artery, both the near and the far walls have a double-line pattern with two parallel echogenic lines separated by a small hypoechogenic space (Fig 1⇓). The first echo on the far wall arises from the lumen-intima interface, while the second arises from the media-adventitia interface. The upper demarcation line of an echo is defined as the “leading edge” (near edge) and the lower demarcation line as the “trailing edge” (far edge). The distance between the leading edge of the first line (line 4 in Fig 1⇓) and the leading edge of the second line (line 5 in Fig 1⇓) of the far wall corresponds to the combined IMT.
Initial cross-sectional and longitudinal scans of the common, the external, and the internal carotid arteries were made to investigate anatomic relations. The carotid artery was then scanned for the presence of plaques (defined as a focal widening of the IMT relative to adjacent segments) starting approximately 3 cm proximally to the bifurcation and extending approximately 2 cm distally to the tip of the flow divider into the internal carotid artery. Atherosclerotic plaques in the near or far wall of the internal carotid artery (the starting point of the internal artery was defined as the tip of the flow divider), the bifurcation, and the CCA were recorded, and the maximal plaque thickness was measured on-line by use of electronic calipers.
The carotid artery was then scanned to find optimal images for IMT measurements. If possible, plaques were included in the images of the IMT. We attempted to visualize both the near and the far walls of the artery simultaneously. To guarantee true proportions of the IMT and lumen diameter, we aligned the ultrasound transducer so that it crossed the axis of the vessel, since only in this projection are the inner echoes of both the near and far walls clearly visible. The loss of parallel configuration of the near and far walls of the CCA served as a reference point for the start of the carotid bifurcation. This point was marked on-line by an arrow (Fig 1⇑). Three frozen images of IMT from both the bifurcation and the CCA were stored on high-resolution S-VHS videotape for off-line analysis. To minimize variability in IMT during the cardiac cycle, image capturing was standardized by recording images at the top of the R wave in an ECG signal. A short sequence of real-time images from the bifurcation and the CCA was also stored on videotape.
Automated Measurement of IMT and Lumen Diameter
The stored ultrasonic images were analyzed off-line by a newly developed computerized technique for automated ultrasonic image analysis.18 19 The video images were digitized by a frame grabber (VideoRaptor, BitFlow Inc) attached to a personal computer and stored on disk for subsequent automated analysis. The automated analysis system combines echo density, edge strength, and edge continuity of the boundaries to decide the exact location of the boundaries.18 Estimated values of the boundary features are included as weighted terms in a cost function. The boundary detection algorithm inspects all points (pixel) in the image and gives favor to those boundary pixels that minimize the cost function. The precision with which an individual cursor line can be placed on a boundary is defined by the pixel dimension (approximately 0.07 mm) of the digitized image. The interfaces were marked automatically and were, if necessary, manually modified (Fig 1⇑). The real-time images were available on a second monitor so that interpretation of the static images could be integrated with the moving images on tape.
A 10-mm wall segment on both sides of the demarcation arrow separating the CCA and the bifurcation was measured, including the near and far walls of the CCA and the far wall of the bifurcation. IMT of the far wall is defined as the distance between the leading edge of the lumen-intima interface and the leading edge of the media-adventitia interface (Fig 1⇑, lines 4 and 5). In the near wall the adventitia-media interface is not easily identifiable because it interferes with the periadventitia-adventitia interface.10 To try to get an estimate of IMT in the near wall, we measured IMT as the distance between the trailing edge of the periadventitia-adventitia interface and the trailing edge of the intima-lumen interface (Fig 1⇑, lines 1 and 3). The computer program estimated the maximum, minimum, and average IMT along the 10-mm segment. The mean of the average, maximum, and minimum IMT of the three frozen images was calculated and used in the analysis. We also measured the lumen diameter of the CCA, defined as the distance between the leading edge of the near wall intima-lumen interface and the leading edge of the far wall lumen-intima interface (Fig 1⇑, lines 2 and 4).
Cardiovascular Risk Factors
Height and weight were measured with the subjects in light clothing without shoes. Blood pressure was recorded by a specially trained technician using an automatic device (Dinamap Vital Signs Monitor 1846, Critikon Inc) before the ultrasound examination and blood sampling. Nonfasting serum total cholesterol and triglycerides were analyzed by enzymatic colorimetric methods with commercial kits (CHOD-PAP for cholesterol and GPO-PAP for triglycerides; Boehringer-Mannheim). Serum HDL cholesterol was measured after the precipitation of lower-density lipoprotein with heparin and manganese chloride. The analyses were done at the Department of Clinical Chemistry, University Hospital of Tromsø. Information about current cigarette smoking was obtained from a self-administered questionnaire.
Between- and within-observer variability was estimated by calculating the mean arithmetic difference and the mean absolute difference between repeated measurements on the same subject. ANOVA (the GLM-procedure in SAS21 ) with IMT as dependent variable and subject (n=111), time (two levels), observer (three levels), and interaction terms of time by subject and observer by subject was used to determine the SDs of between- and within-observer differences (Tables 2⇓ and 3⇓). This procedure was used to exclude intersubject variability from variability caused by measurement error. To examine whether the mean and SD of the arithmetic difference were reasonably constant throughout the range of measurements, we plotted the arithmetic difference between repeated measurements against their average according to Bland and Altman.22 If the differences are normally distributed, 95% of the differences will lie within a range of ±1.96 SD of the mean arithmetic difference. This range will be referred to as “the limits of agreement”. The CV describes measurement error as a percentage of the pooled mean value (x̄) and is calculated according to the formula CV=(s*100/x̄)%. The standard deviation (s) of the measurement error is calculated according to the formula s=SD/. Variance component analysis was used to determine which part of the variability could be attributed to between subject variability and which part to measurement imprecision. The association between measurement error (mean absolute difference in IMT) and cardiovascular risk factor levels was examined by computing Spearman rank order correlation coefficients. Means were compared by ANCOVA with age as covariate. Two-sided values of P<.05 were considered to indicate significance. The SAS software package21 was used.
Measurements of IMT of the CCA and the bifurcation could be obtained in all subjects participating in the reproducibility study. The mean (range) IMT in the reproducibility-study was 0.71 mm (0.44 to 1.74 mm) in the far wall of the CCA, 0.77 mm (0.45 to 1.70 mm) in the near wall of the CCA, and 1.00 mm (0.46 to 2.23 mm) in the far wall of the carotid bifurcation (Table 1⇓). Participants in the reproducibility study were slightly younger than the total study population (Table 1⇓). The reason is that very old persons were not invited to participate in the reproducibility study because the subjects had to undergo two extensive examinations. When we controlled for age there were no significant differences across the two groups.
The mean arithmetic differences among observers were generally small and showed little variation between vessel wall locations or between the two reproducibility studies (Table 2⇓). The mean (SD) arithmetic differences in IMT, computed as the average of three locations, for the three pair of observers were −0.02 (0.17 ) (technician versus physician), 0.01 (0.17 ) (physician versus neurologist) and −0.01 (0.16) mm (technician versus neurologist), indicating that there were no systematic differences between observers. The mean absolute difference and the SD were higher in the bifurcation than in the CCA (Table 2⇓), suggesting greater scattering (ie, lower precision) of the measurements with increasing IMT. Interestingly, the measurement precision values of the near and far walls of the CCA were similar. Lumen diameter was measured with low between-observer variability. Absolute differences between observers in measurements of lumen diameter were significantly correlated with mean lumen diameter, but this association disappeared with logarithmic transformation of the lumen diameter values (data not shown).
Figs 2⇓ and 3⇓ show that between-observer variability is skewed and increases with increasing level of IMT. For the far wall of the CCA the CV is 4.8 when IMT is below the median value (0.68 mm), whereas the CV is 8.4 at IMT above the median value (Fig 2⇓, top panel). The limits of agreement, defined as the range around the mean arithmetic difference where 95% of the differences lie, are ±0.08 mm when IMT is below the median value, whereas they are ±0.20 mm when IMT is greater than the median value (Table 4⇓). The same trends are found with measurements in other locations. Generally the limits of agreement for measurements of IMT thicker than the median value were about twice as large as when IMT was thinner than the median value (Fig 2⇓, Table 4⇓). The strong relationship between measurement error and the level of IMT is further illustrated in Fig 3⇓, which shows that the absolute difference between observers increases significantly with IMT (top panel). The association remained significant after we standardized the measurement error by taking the absolute difference as a percentage of the average IMT (bottom panel). Also, the measurement error remained significantly associated with the level of IMT after logarithmic transformation of the IMT values (r=.33, P=.0001) (data not shown). The between-observer limits of agreement for IMT in the far wall of the carotid bifurcation were ±0.35 mm in subjects with atherosclerotic plaque(s) and ±0.20 mm in subjects without plaque(s).
Within-observer variability tended to be lower than between-observer variability, and the mean arithmetic differences between repeated measurements of the same subject by the same observer were small (Table 3⇓). The mean (SD) arithmetic differences in IMT (average of three locations) were −0.01 (0.18), −0.02 (0.15), and −0.02 (0.15) mm for the neurologist, the physician, and the technician, respectively, indicating that there was no systematic difference between measurements by the same observer on two occasions. Within-observer variability was similar in the beginning and end of the survey period (Table 3⇓). All measures of scatter, such as the mean absolute difference and the SD, increase with increasing IMT (Table 3⇓, Fig 4⇓). In Fig 4⇓ the within-observer difference is plotted against the mean of replicate measurements for each observer. For the far wall of the CCA the CV is 3.6 when IMT is below the median value, whereas the CV is 5.9 when IMT is above the median value (Fig 4⇓, top panel). The limits of agreement increase from ±0.06 mm for IMT below the median value to ±0.14 mm for IMT above the median value (Table 4⇓). The same trends are found with measurements in other locations. Generally the limits of agreement for measurements of IMT thicker than the median value were approximately twice as large as for when IMT was thinner than the median value (Fig 4⇓, Table 4⇓). There was a significant association between within-observer measurement error and the level of IMT before (r=.32; P=.0001) and after (r=.26; P=.0002) logarithmic transformation of the IMT values. The within-observer limits of agreement for IMT in the far wall of the carotid bifurcation were ±0.41 mm in subjects with atherosclerotic plaque(s) and ±0.21 mm in subjects without plaque(s).
The mean (SD) maximum IMT was 0.89 (0.24) mm in the far wall of the CCA and 1.41 (0.66) mm in the far wall of the bifurcation. The mean absolute difference between observers was 0.13 and 0.24 mm for the CCA and the bifurcation, respectively. The corresponding CVs were 12.6 and 13.6, respectively. The within-observer CVs for measurements of maximum IMT were 9.1 and 15.6 for the CCA and the bifurcation, respectively.
The mean (SD) absolute within-reader differences for measurements of IMT in the far and near walls of the CCA and the far wall of the carotid bifurcation were 0.02 (0.02), 0.02 (0.03), and 0.03 (0.03) mm, respectively (n=80). These values may be subtracted from the mean absolute differences shown in Tables 2⇑ and 3⇑ to obtain the proportion of measurement error that is attributable to the sonographers. The within-reader variability did not increase with increasing IMT in the CCA, but in the carotid bifurcation there was a statistically significant correlation between mean absolute difference and mean IMT (r=.49, P=.0001).
Measurement Variability and Cardiovascular Risk Factor Levels
Measurement variability was not associated with cardiovascular risk factors (age, sex, smoking (yes/no), total cholesterol, HDL, systolic blood pressure, body mass index, waist-hip ratio) in any location of the carotid artery when we controlled for mean IMT (data not shown).
The variance component analysis revealed that 83.5% of the total variance in the IMT in the far wall of the CCA was caused by differences between subjects, whereas 16.5% was caused by between- and within-observer variability.
The present study shows that the IMT of the carotid artery can be measured without systematic difference between observers, whereas the precision of the measurements appears to be highly dependent on the level of IMT. The trend of poorer precision with increasing IMT was present in all segments of the carotid artery and was found both for between- and within-observer variability. In the carotid bifurcation, where the severity of atherosclerotic involvement is more advanced, there was a more than twofold increase in measurement error over the range of measurements. We studied a fairly large number of subjects in the setting of a population survey. The sonographers completed an extensive standardization program before the start of the reproducibility study and were blinded to each other’s results, and the images were read with the use of an automatic program. The generalizability of the results therefore appears sound.
The measurement variability in our reproducibility study was small in relation to the biological variability between subjects. The values of the mean absolute difference between repeated measurements in the far wall of the CCA ranged between 0.06 and 0.10 mm, whereas IMT ranged between 0.44 and 1.74 mm in the same location. Reproducibility, measured in terms of both absolute difference and CV, was better for the CCA than for the carotid bifurcation. The arterial walls in the bifurcation are thicker and more prone to plaque development than in the CCA, and this apparently creates a higher degree of measurement variability. In addition, the arterial walls in the bifurcation are curved and therefore may create more scattering of the ultrasound beams. This leads to darker and less detailed images and occasionally nonvisualized interfaces. Within-observer variability was only slightly smaller than between-observer variability. This indicates that in our single center study with three sonographers, the variability in IMT on repeated measurement to a small extent is influenced by different sonographers.
There are several sources of measurement error. The most important factor is probably the use of different interrogation angles during sonography, which can lead to different judgments of where the vessel wall is thickest. Another source of error is different opinions among sonographers regarding the location of the beginning of the bifurcation (Fig 1⇑). This may be difficult to decide when the bifurcation shows little divergence of the near and far walls. Since the arterial wall in the carotid bifurcation is thicker than in the CCA and plaques are more often located in the bifurcation, misinterpretation of the beginning of the bifurcation can lead to measurement variability. This is exemplified in the top panel of Fig 2⇑. The outlier at the lower right quadrant is caused by different opinions among two sonographers about the location of the start of the bifurcation in a subject with little divergence of the near and far walls.
Reproducibility of IMT measurements in the near wall of the CCA has to our knowledge previously only been presented in one study.13 Interestingly, both that and the present study indicate that reproducibility of the near and far walls is quite similar. Prior studies have shown that the ultrasound-measured IMT of the far wall reflects the anatomic intima-media layer.5 23 The IMT of the far wall is measured by using the leading edges of an echo, and the location of the leading edge is on the same level as the level of the interface that creates the echo.10 The IMT of the near wall, on the other hand, is measured using the trailing edges of an ultrasound echo. Prior studies have shown that the trailing edge of an echo is not dependent on the anatomic structure only but also depends on gain setting and other properties of the reading system.10 24 Gamble et al24 showed that a 10-fold increase in gain (from 0.2% to 2% of maximum gain) resulted in a 50% increase in the width of an ultrasound echo in the carotid artery but did not affect the position of the leading edges. In our study gain setting was continuously changed during the scanning procedure to obtain optimal images, but despite this the reproducibility of the near wall IMT measurements was good. We kept the gain setting at the lowest possible level maintaining good-quality images, and it is possible that our small changes in gain did not affect the width of the ultrasound echo. An alternative explanation is that the two trailing edges in the near wall (lines 1 and 3 in Fig 1⇑) are displaced parallel when gain is varied, so that the thickness of the intima-media complex is not altered significantly. The present study suggests that the influence of gain setting on near wall IMT measurements may not be as important as previously thought. However, since the measurements of IMT in the near wall lack an anatomic substrate, near and far wall measurements should be analyzed separately. A multiple regression analysis of the present data showed that age, sex, systolic blood pressure, smoking (yes/no), and total cholesterol accounted for 31% and 34% of the explained variability in mean IMT of the far and near walls of the CCA, respectively. The present results as well as findings showing that the variability of IMT progression estimates may be lowered by 30% to 40% by using near wall measurements,25 indicate that examination of the near wall may provide additional information and should be considered for inclusion in future studies.
Previous studies have used different summary statistics to present the reproducibility of the data, such as the mean arithmetic or absolute differences, the correlation coefficient, or the CV. Comparison of results is therefore complicated. Also, the use of a single summary statistic provides limited information on the amount of agreement and may be misleading if the differences vary systematically over the range of measurements, as shown in the present study. The mean arithmetic differences will, if there is no systematic difference in measurement levels between repeated examinations, add up to a value near zero. However, the SDs of the mean arithmetic differences combined with the absolute differences provide a good estimate of the range of measurement error. The correlation coefficient measures the strength of a linear relationship between replicate measurements, not the agreement between them,22 and is therefore not a suitable indicator of reproducibility.
Table 5⇓ shows a comparison of reproducibility studies of carotid artery IMT. The CVs of the IMT of the far wall and lumen diameter of the CCA are quite similar among the studies. This indicates that the variability of measurements of IMT only to a limited degree depends on differences in ultrasound equipment and protocols across studies. In Table 5⇓ the Tromsø values include within-reader variability. In our study 70% to 80% of total variability is due to sonographers, and the rest is attributable to the reading process. The within-reader mean (SD) absolute difference of the average of three locations in the present study was 0.02 (0.03) mm compared with 0.06 (0.05) mm in the Asymptomatic Carotid Artery Plaque Study,26 and 85% of the differences in maximum IMT in the carotid bifurcation were less than 0.13 mm, whereas in the Atherosclerosis Risk in Communities study 85% of the differences were less than 0.20 mm.27 This may indicate that the automated computerized image analyzing system developed by the group at the Wallenberg Laboratory at Sahlgrenska University Hospital, Gothenburg, Sweden, presents less variability than other reading systems.
IMT is being used both as an exposure variable to predict cardiovascular disease and as an outcome variable to study the determinants of atherosclerosis. When measurement imprecision of the exposure variable occurs randomly (ie, independent of the outcome under study), the magnitude of an association between exposure and outcome may be underestimated. Statistical methods can to some degree correct for measurement error in the exposure to obtain more reliable estimates.28 29 In the present study measurement error did not occur completely at random but increased with increasing level of IMT and was associated with the presence of atherosclerotic plaques. Thus, our data show that when IMT is used as an exposure to predict atherosclerotic cardiovascular disease, errors in the classification of individuals by exposure produce a differential accuracy depending on disease status. Unfortunately, it is difficult to estimate the precise effect of nonrandom misclassification; it can result in a biased risk estimate that is an overestimate, an underestimate, or, by chance, the same as the true measure of association.30 It has been suggested that IMT values be transformed logarithmically,14 but our data indicate that this will not remove the association between measurement error and IMT level. When IMT is used as an outcome variable, measurement imprecision may lead to misclassification of subjects. In the present study measurement imprecision was not associated with risk factor levels, and misclassification will therefore be nondifferential and tend to weaken the association between risk factor and outcome.
In conclusion, our findings suggest that reproducibility of IMT measurements of the carotid artery is acceptable, but measurement error increases considerably with increasing IMT. Computer algorithms for automatic reading of images may be used to minimize reading error and improve reliability. Ultrasound may be a useful tool in epidemiological studies on the determinants of atherosclerosis and cardiovascular disease. However, the precision of the method must be improved before it can be used in the monitoring of atherosclerosis on an individual level. Performing replicate measurements and/or using a fixed angle of interrogation during sonography can do this, although the use of a fixed angle will probably lead to more missing values.
Selected Abbreviations and Acronyms
|CCA||=||common carotid artery|
|CV||=||coefficient of variation|
This study was supported by grants from the Norwegian Research Council. We are indebted to John Wikstrand and Inger Wendelhag at the Wallenberg Laboratory for Cardiovascular Research, Gothenburg University, Sweden, for introducing us to their ultrasound scanning methods. We are also indebted to Tomas Gustavsson and Quan Liang, Department of Applied Electronics, Chalmers University of Technology, Gothenburg, Sweden, for valuable help with the reading of the ultrasound images.
- Received February 28, 1997.
- Revision received May 12, 1997.
- Accepted June 24, 1997.
- Copyright © 1997 by American Heart Association
Heiss G, Sharrett R, Barnes R, Chambless L, Szklo M, Alzola C. Carotid atherosclerosis measured by B-mode ultrasound in populations: associations with cardiovascular risk factors in the ARIC Study. Am J Epidemiol. 1991;134:250-256.
Crouse J, Toole J, McKinney W, Dignan M, Howard G, Kahl F, McMahan M, Harpold G. Risk factors for extracranial carotid artery atherosclerosis. Stroke. 1987;18:990-996.
Howard G, Sharrett R, Heiss G, Evans G, Chambless L, Riley W, Burke G. Carotid artery intimal-medial thickness distribution in general populations as evaluated by B-mode ultrasound. Stroke. 1993;24:1297-1304.
Pignoli P, Tremoli E, Poli A, Oreste P, Paoletti R. Intimal plus medial thickness of the arterial wall: a direct measurement with ultrasound imaging. Circulation. 1986;74:1399-1406.
Wendelhag I, Wiklund O, Wikstrand J. Arterial wall thickness in familial hypercholesterolemia: ultrasound measurement of intima-media thickness in the common carotid artery. Arterioscler Thromb. 1992;12:70-77.
O’Leary D, Polak J, Kronmal R, Kittner S, Bond MG, Wolfson S, Bommer W, Price T, Gardin J, Savage P. Distribution and correlates of sonographically detected carotid artery disease in the Cardiovascular Health Study. Stroke. 1992;23:1752-1760.
Salonen R, Tervahauta M, Salonen JT, Pekkanen J, Nissinen A, Karvonen MJ. Ultrasonographic manifestations of common carotid atherosclerosis in elderly eastern Finnish men: prevalence and associations with cardiovascular diseases and risk factors. Arterioscler Thromb. 1994;14:1631-1640.
Persson J, Stavenow L, Wikstrand J, Israelsson B, Formgren J, Berglund G. Noninvasive quantification of atherosclerotic lesions: reproducibility of ultrasonographic measurement of arterial wall thickness and plaque size. Arterioscler Thromb. 1992;12:261-266.
O’Leary D, Polak J, Wolfson S, Bond MG, Bommer W, Sheth S, Psaty B, Sharrett R, Manolio T. Use of sonography to evaluate carotid atherosclerosis in the elderly: the Cardiovascular Health Study. Stroke. 1991;22:1155-1163.
Mercuri M, Bond MG, Nichols FT, Carr AA, Flack JM, Byington R, Raines J. Baseline reproducibility of B-mode ultrasound imaging measurements of carotid intima media thickness: the Multicenter Isradipine Diuretic Atherosclerosis Study (MIDAS). J Cardiovasc Diag Procedures. 1993;11:241-255.
Bønaa KH, Thelle DS. Association between blood pressure and serum lipids in a population: the Tromsø Study. Circulation. 1991;83:1305-1314.
Bønaa KH, Arnesen E. Association between heart rate and atherogenic blood lipid fractions in a population: the Tromsø Study. Circulation. 1992;86:394-405.
Wendelhag I, Liang Q, Gustavsson T, Wikstrand J. A new automated computerized analyzing system simplifies readings and reduces the variability in ultrasound measurements of intima-media thickness. Stroke. In press.
Gustavsson T, Liang Q, Wendelhag I, Wikstrand J. A dynamic programming procedure for automated ultrasonic measurement of the carotid artery. In: Program and abstracts of the meeting of Computers in Cardiology; September 25-28, 1994; Washington, DC; 297-300.
Bjartveit K, Foss O, Gjervig T, Lund-Larsen P. The cardiovascular disease study in Norwegian counties: background and organization. Acta Med Scand. 1979;634:1-70.
SAS Institute Inc. SAS/STAT Guide for Personal Computers, Version 6 Edition. Cary, NC: SAS Institute; 1987.
Wong M, Edelstein J, Wollman J, Bond MG. Ultrasonic-pathological comparison of the human arterial wall: verification of intima-media thickness. Arterioscler Thromb. 1993;13:482-486.
Riley W, Barnes R, Applegate W, Dempsey R, Hartwell T, Davis V, Bond MG, Furberg C. Reproducibility of noninvasive ultrasonic measurement of carotid atherosclerosis. Stroke. 1992;23:1062-1068.
Willet W. Correction for effects for measurement error. In: Willet W, ed. Nutritional Epidemiology. New York, NY: Oxford University Press; 1990.
Hennekens C, Buring J, Mayrent S, ed. Epidemiology in Medicine. Boston, Mass: Little, Brown & Co; 1987.