Introduction

Dysfunction of the vascular endothelium is considered to be a precursor of atherosclerosis [1], and Type 2 diabetes is associated with both endothelial dysfunction [2] and increased risk of cardiovascular disease [3]. However, dietary [4] and pharmacological [5] interventions significantly improve endothelial function, even in this high-risk group. One non-invasive technique for assessing endothelial health uses ultrasound to examine the magnitude of the increase in diameter in the brachial artery after an increase in blood flow [6, 7], a phenomenon called flow-mediated dilation (FMD) [8]. Impaired FMD may be an independent predictor of coronary events [9], and FMD scores are inversely correlated with a number of coronary risk markers [6], including insulin resistance [10], high WHR [11] and dyslipidaemia [6].

FMD can be measured using ultrasound equipment found in most medical centres and involves little discomfort or risk for patients. However, given the small diameter of the brachial artery (2–5 mm) and the large number of extraneous variables that can affect FMD [12, 13], it is challenging to measure it reliably. Published estimates of within-subject variability in FMD differ substantially (Table 1), and no studies to date have measured reliability in patients with diabetes. Variability is often attributed to sonographer technique and measurement error; however, it is also possible that day-to-day variations are caused by fluctuations in biological parameters such as glucose, insulin and triglycerides. In the present study, we examined the stability of FMD estimates in healthy adults with diabetes and examined the implications of any variability for future intervention studies in this population.

Table 1 Studies reporting the CV for FMD and artery diameters in healthy adults
Full size table

Subjects and methods

Subjects

We recruited 18 adults (aged 41–73 years) with Type 2 diabetes for a study of the acute effects of unsaturated fatty acids on vascular reactivity [14]. Patients were not insulin dependent and were otherwise reasonably healthy. Only the fasting data from the study are reported here. The sample included two postmenopausal women, one premenopausal woman and 15 men. Because hormone fluctuations across the menstrual cycle affect endothelial function, the premenopausal woman was examined during the early follicular phase of three consecutive menstrual cycles and women using oral contraceptives or hormone replacement therapy were excluded. Table 2 shows the cardiovascular risk characteristics of the subjects. The average BMI of the subjects in this sample was 29.3 kg/m2.

Table 2 Vascular and metabolic parameters measured under fasting conditions across the three testing sessions
Full size table

Study design

Each participant was tested after a 12-h fast on three occasions, separated by at least 1 week (with the exception of the premenopausal female participant, all three tests were completed within 60 days). Participants were asked to avoid alcohol for 48 h, to discontinue medications the night before each test and to maintain similar exercise and diet habits throughout the study. The study was approved by the Office of Regulatory Compliance at the Pennsylvania State University and written informed consent was obtained from each participant.

Ultrasound assessments

An Acuson 128XP duplex ultrasound imaging system (Siemens Medical Solutions, Malvern, Pa., USA) with a 10-MHz linear array transducer was used to measure brachial artery diameter at baseline during reactive hyperaemia according to published guidelines [7]. Ischaemia was induced by inflating a BP cuff on the forearm (distal to the target artery) to 50 mm Hg above systolic BP using an automated device (D. E. Hokanson, Bellevue, Wash., USA). Continuous, longitudinal, two-dimensional images of the brachial artery at 5 to 10 cm above the elbow of the right arm were obtained and stored on SVHS tape during quiet rest (1 min), cuff occlusion (5 min) and reactive hyperaemia (2 min). The ultrasound examinations were performed by a single registered vascular technologist (P. Wagner). Flow velocity was measured using duplex pulsed Doppler with the ultrasound beam at two time points: at the beginning of baseline and immediately after cuff release. Flow (ml/min) was calculated using the following equation: velocity time integral × cross-sectional area of the vessel (πr2)×heart rate.

Analysis of arterial diameters

Images for analysis were sampled at end diastole using Brachial Imager software (Medical Imaging Applications, Iowa City, Iowa, USA). This yielded approximately 50 to 60 frames at baseline and 100 to 140 frames during the post-deflation sequence. Within each frame, diameters were repeatedly measured along a segment of vessel 2 to 8 mm long using automated edge-detection software (Brachial Analyzer; MIA, Iowa City, Iowa, USA) [15]. Each sequence of ultrasound images was reviewed by a single technician and scores were confirmed by a second observer (S. L. Schoemer). If FMD estimates differed by more than 1.5%, consensus was reached by a third scorer (S. G. West). The same arterial landmark [16] was used for all scans collected from a single patient.

The average value of all interpretable images over the 1-min baseline period was used for resting arterial diameter. Peak FMD (FMDpeak) was calculated by identifying the post-deflation image with the largest average diameter and using this value to calculate percentage change from baseline. We examined whether increasing the number of frames used to estimate the peak diameter would produce more consistent estimates. FMDpeak±5 was calculated using the average of the peak diameter, the five frames preceding the peak and the five frames following the peak. Similar calculations were performed using the average of two, four, six or eight additional frames surrounding the peak. Unless otherwise specified, FMDpeak±5 was reported here.

Lipids, lipoproteins, glucose and insulin were measured using conventional methods as described previously [14]. The homeostasis model assessment for insulin resistance (HOMAIR) was calculated using the following formula: fasting plasma glucose (mmol/l) × fasting insulin (µU/ml) / 22.5 [17]. We also calculated the quantitative insulin sensitivity check index (QUICKI), a recently proposed indicator of insulin sensitivity using the following equation: 1 / (log insulin × log glucose), where glucose is measured in mg/dl and insulin is measured in µU/ml [18].

An oscillometric monitor (Dinamap Pro 100, Critikon, Tampa, Fla., USA) was used to measure BP. Stroke volume, cardiac output and total peripheral vascular resistance were estimated every other minute during resting periods via impedance cardiography as described previously [19].

Statistical analyses

We conducted a series of repeated-measures ANOVA using mixed models (SAS Version 8.2, SAS Institute, Cary, N.C., USA) to examine within-subject change in the outcome variables over repeated visits. Pearson correlations, adjusting for basal artery diameter, were used to test whether average fasting FMD was correlated with any of the demographic or cardiac-risk parameters.

The CV was calculated for each subject from the mean and SD for the three measurements as follows: CV = (100 × SD) / mean. Herrington et al. [12] noted that many published studies have used an alternative method of calculating CV, which they termed CV’ : CV’ = (100 × SD) / 100 + mean. The CV was calculated for all variables, whereas CV’ was only reported for FMD. Additional variability measures included: (i) correlations across successive visits; and (ii) mean variability (the difference between pairs of measurements, expressed as the absolute value, averaged across all three pairs of measurements from an individual). We examined whether the mean variability in FMD was correlated with the mean variability in haemodynamic measures (blood flow, BP, heart rate and total peripheral vascular resistance) or with variations in glucose, insulin or lipids/lipoproteins. Multivariable regression analysis was used to estimate how much of the variance in FMD over the three visits could be explained by the biological predictors. A p value less than or equal to 0.05 was considered statistically significant.

Finally, we estimated the sample sizes for crossover and parallel-arm designs using two-sided tests, power = 0.80 – 0.90 and alpha = 0.05. The within-subject SD was estimated using the equation SDwithin = √MSE (where the MSE is the mean square error from ANOVA with subject as a main effect) [20].

Results

There were few systematic changes in the outcome variables across the three measurements (Table 2). There were no significant changes in FMD or hyperaemic-flow volume across the three visits. Diastolic BP (−1.4 mm Hg), triglycerides (−16 mmol/l) and hyperaemic-flow velocity (−22%) decreased significantly (p≤0.02) from the first to the third visit. In contrast, fasting glucose was 0.39 mmol/l higher at the third visit (p = 0.04).

Predictors of fasting FMD

Lower FMD was associated with increasing age (r = −0.72, p = 0.001) and higher systolic BP (r = −0.50, p = 0.04). Subjects with high fasting triglycerides (≥1.69 mmol/l) had lower FMD scores than subjects with low or normal fasting levels (mean FMD = 3.76±0.74 vs 6.31±0.66, p = 0.03). None of the other demographic, metabolic or cardiovascular-risk variables were significant predictors of fasting FMD, and fasting glucose was not related to fasting FMD in individuals either across visits or in the sample as a whole (Fig. 1).

Fig. 1
figure 1

Fasting glucose concentrations and brachial artery FMD values in individuals with Type 2 diabetes (n=18, three visits per subject). Open circles, visit one; open squares, visit two; filled triangles, visit three

Full size image

Reliability of vascular ultrasound measures

Reliability statistics for FMD, artery diameter, arterial-blood-flow volume and blood-flow velocity are shown in Tables 3 and 4. The CV for FMD was 29.7%, while the mean value for CV’ was 1.2%. The absolute diameters at baseline and peak dilation were measured more reliably than FMD itself was, and FMD variability was similar to that reported by Herrington et al. [12].

Table 3 Test–retest reliability of FMD-related variables
Full size table
Table 4 Test–retest reliability of blood flow and blood-flow velocity measurements
Full size table

The primary measure of variability was the mean difference between three pairs of measurements, expressed as an absolute value (Table 3). Variability in FMD was positively correlated with variability in glucose (r=0.52, p=0.03), insulin (r=0.47, p=0.05) and heart rate (r=0.48, p=0.04), despite the fact that FMD was not related to absolute levels of these parameters. In multivariate analysis, only mean variability in glucose remained an independent predictor, accounting for 35% of the variance in FMD variability. FMD variability was unrelated to fasting glucose or any of the other health-related variables examined (HbA1c, age, BMI, BP and lipids). Results were unchanged when adjusted for the number of days between visits.

Effects of peak selection on the stability of FMD

As expected, FMD scores were higher when a single frame was used to calculate the peak diameter (Table 5). However, the SD of the repetitions remained stable irrespective of the number of adjacent frames used to define the peak. The lowest CV (indicating higher reliability) was observed when the mean diameter of a single frame was used.

Table 5 Effect of increasing the number of frames used to calculate the peak diameter after cuff deflation on the stability of FMD estimates
Full size table

Sample size

Table 6 shows sample size requirements for crossover and parallel-arm designs when the magnitude of the treatment effect on FMD can be estimated. If an intervention is expected to increase fasting FMD by 25% (power = 0.90), 30 subjects would be needed to find significance in a crossover design, while a parallel-arm design would require 65 subjects.

Table 6 Sample sizes required to determine significant effects on FMD in crossover and parallel-arm studies
Full size table

Discussion

Experts in the measurement of FMD suggest that individual laboratories document test–retest reliability of their measurement protocols by repeatedly testing the same individuals in the absence of changes in medication status, diet or exercise regimen. We have shown that FMD can be reliably measured in healthy individuals with diabetes, and that fasting glucose and insulin concentrations are not correlated with FMD scores measured simultaneously. In the present study, fasting glucose varied by more than 5.56 mmol/l between individuals and by up to 2.78 mmol/l within individuals. Thus, we cannot comment on whether or not glucose variations over a broader range would impact FMD scores. Our estimates of test–retest reliability in patients with Type 2 diabetes are in keeping with those in samples of healthy adults [12], and our estimates of the CV for FMD are considerably lower than in several other studies.

The variability estimates reported here may be considered conservative given the controlled conditions under which the measurements were collected. Participants were instructed to avoid changes in exercise, diet or medication regimen during the study. Before each test, subjects fasted for 12 h (confirmed by glucose measurement). Ultrasound studies were conducted after a 45-min habituation period and 20 min of quiet rest. A single sonographer performed all FMD assessments, and arterial diameters were measured using automated scoring software that had been validated for this purpose [15]. Our design controlled for diurnal variability in FMD [21] and hormone fluctuations caused by the menstrual cycle [22], and there were only minimal changes in heart rate and BP over the three visits.

Nevertheless, the reported CVs for FMD (in this study, and in many others) are higher than those conventionally accepted for biochemical assays. As noted previously, variability in FMD is likely to result from the combined effects of measurement error, real biological variability and subtle differences in the application of the technique over repeated testing sessions [20]. We found that subjects with more variable values for glucose, insulin and heart rate also showed the greatest variation in FMD across testing days. Variability in glucose was the only independent predictor of variability in FMD.

These results suggest that true biological variability is a significant source of variation in FMD scores. However, the lack of a correlation between fasting glucose concentrations and fasting FMD scores indicates that the relationship between glucose and FMD is complex. It is possible that larger day-to-day fluctuations in glucose are a marker of more advanced disease, and that FMD variability may be even greater in individuals with poorly controlled diabetes. In as far as heart rate, glucose and insulin levels can be controlled by careful subject selection and instruction, researchers are advised to account for these variables in their study design and analyses.

In this study, measurements of hyperaemic flow were also highly variable from one testing session to the next. However, neither FMD nor the volume of blood flow during reactive hyperaemia were significantly altered with repeated testing. Our results showed an unexpected decrease in peak flow velocity during hyperaemia over time. These results highlight the importance of careful study design, including counterbalanced presentation of treatments in crossover studies.

There appears to be no additional benefit (in terms of lower variability) from including diameters from multiple adjacent frames in the calculation of FMD scores. In fact, CV estimates actually increased when more diameters were used to determine peak diameter. The use of automated edge-detection software in this study meant that the value for average diameter within each frame was actually based on dozens of measurements across the segment of vessel in the region of interest. This method substantially reduces the influence of a single aberrant diameter on FMD estimates, and FMD estimates determined using this software may be more reliable than those obtained by traditional methods for measuring arterial boundaries [15].

Although nutritional interventions have been shown to improve FMD in smokers and subjects with elevated cholesterol [13], relatively few of these studies have included patients with diabetes. Using the same study population, we recently reported that individuals with both diabetes and hypertriglyceridaemia showed substantial FMD improvements 4 h after a meal containing omega-3 fatty acids [14]. The success of future intervention studies is dependent on accurate estimates of within-subject variability and control of extraneous variables that are known to influence FMD. In this study, we have demonstrated that FMD can be measured as reproducibly in patients with diabetes as in persons without diabetes, even in the presence of glucose and insulin fluctuations across testing days.