Measurement Repeatability of Central and Peripheral Blood Pressures: The ARIC Study

Abstract

BACKGROUND

Central systolic and pulse pressures are markers of risk for small vessel disease in the brain and kidneys. The extent to which these markers are reproducible in the setting of population studies is less well established. We estimated short-term repeatability of central systolic and pulse pressures, and those of their peripheral measures for comparison.

METHODS

Participants aged 65 years and over (n = 79, 56% women) were drawn from the 2011–2013 examination of the ARIC cohort. Measurements were obtained with automated devices in the supine position, except for conventional sitting pressures, from paired measurements at each of 2 visits separated by 4 to 8 weeks. Three-level variance component models with between-participant, between-visit, and within-visit components estimated reliability metrics.

RESULTS

Mean central systolic and pulse pressures were higher than conventional brachial measures, yet their 4 to 8 week measurement repeatability was similar: reliability coefficients were 0.62 (95% confidence interval: 0.49, 0.74) and 0.63 (0.51, 0.76) for central and sitting brachial systolic pressures, and 0.66 (0.54. 0.77) and 0.73 (0.63, 0.82) for their corresponding pulse pressures. Between-participant variation contributed to two-thirds of the short-term repeatability for all measures. Within-visit variation remained uniformly low across visits.

CONCLUSIONS

Our results indicate that the average of 2 standardized measurements obtained at a single visit can provide reliable estimates of central systolic and pulse pressures. The reliability coefficients of central and peripheral blood pressure measures were comparable. Estimates are presented of minimal detectable change and difference to aid in study design and evaluation of analytic results.

Central systolic and pulse pressures have been studied as measures of the central pulsatile hemodynamic load imposed on the cardiac, cerebral, and renal vasculatures, and to assess their association with small vessel disease.^1–3 Central blood pressure (BP) levels derived from noninvasive pulse recordings either at the radial or carotid artery have also been proposed for hypertension reclassification and/or management among adults at high risk for microvascular damage attributable to hypertension.^4–6 The extent to which these measurements are reproducible in the setting of population studies is not well established, although measurement reproducibility influences parameter estimates of associations, the magnitude of intervention effects, and a study’s statistical power.

Our goal was to quantify the 4 to 8 week repeatability of central systolic BP (cSBP) and central pulse pressure (cPP) in a sample of community-dwelling older adults. For comparison, repeatability was assessed also for peripheral pressures: conventional sitting brachial, and supine bilateral brachial and posterior tibial BPs. The comprehensive assessment of these BP measurements is in line with a recommendation of current guidelines for older adults.^7,8

We further set out to estimate the main sources of the 4 to 8 week measurement variability in terms of between-participant, between-visit, and random error components. Lastly, we wished to incorporate the information on the short-term repeatability of cSBP and cPP into metrics that aid in the interpretation of treatment effect and observed associations with these physiologic parameters, and for an optimal design of clinical trials and observational studies.

METHODS

Study population

This study was nested in the Atherosclerosis Risk in Communities (ARIC) study. Details of the ARIC study design and of the design of this repeatability study have been published.^9,10 Unselected cohort members taking part in the 5th ARIC examination (2011–2013) were invited for a repeat examination (n = 80, equally distributed across the 4 ARIC field centers). Study participants were excluded from the repeatability analyses for conditions that could influence measurement quality, such as evidence of major cardiac arrhythmias (n = 1), aortic stenosis (n = 1), and a body mass index ≥40 kg/m² (n = 1). Participants provided written informed consent, and the study was approved by the Institutional Review Boards at each of the field centers, coordinating center, and central labs and reading centers.

Measurement procedures

BPs were measured twice following a 5-minute rest, at each of 2 visits separated by 4 to 8 weeks (mean follow-up: 40 ± 10 days). Trained and certified technicians followed the standardized ARIC study protocol using validated, automated devices¹¹ (see Manual 2: Cohort Procedures Visit 5 for detail at https://www2.cscc.unc.edu/aric/cohort-manuals).¹² Briefly, participants were asked to not consume food or drinks and to refrain from vigorous physical activity and smoking for 8 hours prior to each visit. Participants rested for 5 minutes before each measurement during the first visit, and repeated this procedure at a second visit.

Central systolic and pulse pressures.

Supine cSBP levels were derived from direct pulse recording via a tonometry sensor positioned on the left common carotid artery, using the Omron VP-1000 plus device (Omron Healthcare, Kyoto, Japan). The sensor was secured and its holding pressure was standardized via a customized neck collar. Recorded carotid waveforms were calibrated with simultaneously measured brachial mean arterial pressure and diastolic BP (DBP) with an oscillometric cuff on the right arm. The calibration method assumes that mean arterial pressure and DBP are largely constant between the brachial and carotid arteries.¹³ cPP was defined as the difference between cSBP minus supine right brachial DBP, with the assumption that DBP values are largely uniform throughout the arterial tree. The device automatically set extreme values as missing, which resulted in 6% of missing information.

Peripheral systolic and pulse pressures: conventional sitting brachial and four-limb BPs.

Measurements of peripheral pressure included conventional sitting BP, and supine bilateral brachial and posterior tibial BPs. The latter (i.e., supine peripheral BPs) was taken simultaneously with central BP. Sitting BP was measured following a 5-minute rest with a validated, automated oscillometric device (Omron HEM-907 XL). Standardized cuff sizes were selected according to arm measurements.¹²

Statistical analysis

Summary statistics and Bland–Altman plots described the distributional properties of the data. To assess repeatability, mean differences were calculated for paired measurements within-visit (5 minutes apart) and between-visits (4 to 8 weeks apart). The differences were compared using 3-level variance component models in which the amount of measurement variability was decomposed into 3 variance components (i.e., between-participant, between-visit, and within-visit).^14,15 The conventional sitting BP was used as a reference to evaluate measurement repeatability since the paucity of reports on the measurement properties of central BP makes it difficult to set a priori reference values. Reliability metrics included intraclass correlation coefficients (ICC) with 95% confidence intervals, coefficient of variation, and SEM.^16,17 We also calculated minimal detectable change based on 95% confidence intervals estimates and minimal detectable difference with a 90% confidence level to detect true mean BP change for 1- and 2-sample study designs, respectively. The conventional 2-sided P-value of 0.05 was used and statistical analyses were performed using SAS 9.4 (Cary, NC).

RESULTS

A total of 79 participants (56% women) were available for the analysis (Table 1), whose mean (±SD) age was 76 (±5) years with a range from 68 to 86 years. The mean heart rate was 64 bpm, and 77% were hypertensive.

Table 1.

Characteristics of the study participants^a (n =79)

Variable	Mean ± SD or n (%)
Age (year)	75.8 ± 4.7
Women	44 (56)
African American	24 (30)
Ever smokerb	44 (56)
Current cardiovascular conditions
Hypertension	61 (77)
Use of antihypertensive medication	54 (74)
HDL cholesterol (mmol/l)	1.3 ± 0.4
LDL cholesterol (mmol/l)	2.6 ± 0.7
Triglycerides (mmol/l)	1.4 ± 0.7
Heart rate (bpm)	64 ± 10
Diabetesc	33 (42)
Anthropometric measures
Body mass index (kg/m2)	29.7 ± 4.3
Abdominal circumference (cm)	102.4 ± 12.7

Abbreviations: HDL, high-density lipoprotein; LDL, low-density lipoprotein.

^aParticipants’ characteristics were measured at the study intake examination, ARIC Cohort Visit 5, 2011–2013.

^bEver smoker included 1% current smoker.

^cDiabetes was defined as hemoglobin A1c ≥ 6.5%.

Summary statistics of central and peripheral BP measures are shown in Table 2 (and Supplementary Table 1). The mean cSBP level at visit 1 was higher than the sitting and supine brachial systolic pressures on the right arm, respectively (147.2 ± 22.5 vs. 130.4 ± 18.2, and 138.4 ± 16.0 mm Hg), but not higher than the corresponding SBP at the ankle (165.1 ± 22.4 mm Hg). The mean sitting SBP level was lower (by 7–9 mm Hg) and showed a higher dispersion than the supine SBPs. This pattern was observed also for the PP measures (Table 2). The mean values at visit 1 were somewhat higher than at visit 2 although the SDs were similar between visits 1 and 2. Examination of quantiles showed that the study data were reasonably symmetric, without major outliers (Supplementary Table 1).

Table 2.

Mean levels at visit 1, and mean differences for paired measurements within-and between-visits (n = 79)

			Within-visita		Between-visitsb
		Visit 1	Average	Absolute	Average	Absolute
	n	Mean ± SD	Mean difference ± SD		Mean difference ± SD
SBP
R brachial SBP-sitting	74	130.4 ± 18.2	0.5 ± 5.5	5.4 ± 3.7	−0.7 ± 14.6	11.4 ± 9.4
Central SBP-supine	67	147.2 ± 22.5	−5.3 ± 8.9	10.4 ± 6.5	−0.5 ± 18.8	15.5 ± 12.1
Rc brachial SBP-supine	75	138.4 ± 16.0	−3.2 ± 4.8	5.6 ± 3.5	−2.6 ± 13.3	10.6 ± 8.6
L brachial SBP-supine	76	139.1 ± 15.2	−3.3 ± 5.0	5.6 ± 3.7	−3.0 ± 12.5	11.0 ± 7.2
R ankle SBP-supine	76	165.1 ± 22.4	−3.6 ± 8.9	8.6 ± 6.9	−3.2 ± 18.7	16.0 ± 10.9
L ankle SBP-supine	76	163.8 ± 23.2	−0.4 ± 4.9	4.7 ± 3.2	0.1 ± 9.0	7.6 ± 5.4
PP
R brachial PP-sitting	74	64.8 ± 16.3	0.5 ± 5.8	5.5 ± 3.9	−1.6 ± 10.7	8.8 ± 7.3
Central PP-supine	64	73.8 ± 19.0	−4.2 ± 7.8	9.4 ± 5.2	−0.2 ± 14.3	13.1 ± 8.4
R brachial PP-supine	75	66.0 ± 11.0	−2.3 ± 4.3	5.3 ± 3.3	−2.5 ± 8.5	7.9 ± 5.3
L brachial PP-supine	76	66.4 ± 11.7	−1.6 ± 4.4	4.9 ± 3.1	−2.5 ± 8.2	7.9 ± 4.3
R ankle PP-supine	76	92.6 ± 17.0	−3.0 ± 9.3	8.7 ± 6.9	−3.1 ± 14.3	13.4 ± 8.8
L ankle PP-supine	76	89.2 ± 18.6	−2.7 ± 7.4	8.5 ± 4.7	−2.7 ± 12.8	12.0 ± 7.4

Abbreviations: PP, pulse pressure; SBP, systolic blood pressure.

^aWithin-visit mean difference = {[(2^nd measure at visit 1) − (1^st measure at visit 1)] + [(2^nd measure at visit 2) − (1^st measure at visit 2)]}/2.

^bBetween-visit mean difference = {[(1^st measure at visit 2) − (1^st measure at visit 1)] + [(2^nd measure at visit 2) − (2^nd measure at visit 1)]}/2.

^cR = right; L = left.

As evaluated by Bland–Altman plots (Supplementary Figures 1 and 2), the dispersion of between-visit measurements relative to 95% limits of agreement indicated constant variances of the between-visit mean differences overall, across the 2 study measures. The variability of the mean differences conditional on mean levels found in some measures such as central BPs was negligible.

Table 2 also shows the mean change of each paired measurement within- and between-visits, respectively. The mean within-visit and between-visit measurement differences were of small magnitude, and as expected, the mean between-visit differences were larger than the within-visit differences. Larger mean within- and between-visit measurement differences were observed for cSBP and cPP than for the measurements obtained at the ankles. Also consistent with expectations, the negative signs on the mean differences indicate that first readings were slightly higher than the second, even after the 5-minute rest before each measurement.

Between-participant variability was the largest source of variation in measurements (Table 3). The between-participant variation contributed to two-thirds of the short-term repeatability both for central and peripheral SBPs (range: 62–69% of the total variance) and for pulse pressure measures (range: 66–73%). Random measurement error (i.e., within-visit variation) explained the smallest portion for SBPs (range: 9–17%) and remained uniformly low from visit 1 to visit 2. The proportions of variance explained by random error and by between-visit variation for the PP measurements were similar to those of the SBP measurements.

Table 3.

Variance components for short-term repeatability measurements of systolic and pulse pressure (n of observations = 316)^a

	R Brachial SBP-sitting		Central SBP-supine		R Brachial SBP-supine		L Brachial SBP-supine		R Ankle SBP-supine		L Ankle SBP-supine
Source of variation	SDb	%c	SD	%	SD	%	SD	%	SD	%	SD	%
Between-participants	14.3	63	17.8	62	13.4	63	13.5	66	19.7	65	19.4	69
Between-visits	9.5	28	10.6	22	8.6	28	8.2	24	11.8	23	10.7	21
Within-visit	5.3	9	9.3	17	5.0	9	5.2	10	8.6	12	7.3	10
Total	18.0		22.7		16.7		16.6		24.5		23.3
	R Brachial PP-sitting		Central PP-supine		R Brachial PP-supine		L Brachial PP-supine		R Ankle PP-supine		L Ankle PP-supine
Between-participants	13.5	73	16.1	66	9.7	67	10.1	71	17.1	68	16.6	71
Between-visits	6.8	18	8.6	19	5.0	18	4.9	16	8.9	18	7.3	14
Within-visit	4.7	9	7.9	16	4.6	15	4.3	13	7.6	14	7.8	15
Total	15.8		19.9		11.8		12.0		20.8		19.8

^a n of observations was obtained from 2 series of paired measurements over 4–8 weeks on the same participants (79 × 2 × 2 = 316).

^bVariance component estimate was expressed as SD.

^dPercentage of variation; percentages were rounded to the nearest whole number.

^eR = right; L = left; SBP = systolic blood pressure; PP = pulse pressure.

Within-visit and between-visit ICCs were estimated (Table 4). The within-visit 1 ICCs ranged from 0.84 to 0.94 for SBPs and 0.80 to 0.92 for PPs. The between-visit ICCs (or short-term repeatability), based on between-visit variation, ranged from 0.62 to 0.69 for SBPs and from 0.66 to 0.73 for PPs. The ICC estimates and the width of their confidence intervals were similar across all measurements both within-visit 1 and between-visits.

Table 4.

Indices of repeatability for systolic and pulse pressure measures

	Within-visit 1 (n = 158)				Between-visit (n = 316)
	ICC (95% CI)	CoV	SEM	MDC95	ICC (95% CI)	CoV	SEM	MDC95
SBP
R Brachial SBP-sitting	0.94 (0.91, 0.97)	14.5	4.7	13	0.63 (0.51, 0.76)	13.8	10.9	30
Central SBP-supine	0.84 (0.77, 0.91)	15.8	9.4	26	0.62 (0.49, 0.74)	15.5	14.1	39
R Brachial SBP-supine	0.89 (0.84, 0.94)	12.3	5.6	16	0.65 (0.53, 0.77)	12.1	9.9	28
L Brachial SBP-supine	0.89 (0.84, 0.93)	11.9	5.6	15	0.66 (0.54, 0.77)	12.0	9.7	27
R Ankle SBP-supine	0.86 (0.80, 0.92)	15.0	9.3	26	0.65 (0.53, 0.76)	15.2	14.6	40
L Ankle SBP-supine	0.89 (0.85, 0.94)	14.6	7.8	22	0.69 (0.59, 0.80)	14.5	12.9	36
PP
R Brachial PP-sitting	0.92 (0.89, 0.96)	25.7	4.6	13	0.73 (0.63, 0.82)	24.8	8.3	23
Central PP-supine	0.82 (0.74, 0.90)	27.7	8.7	24	0.66 (0.54, 0.77)	26.9	11.6	32
R Brachial PP-supine	0.80 (0.71, 0.88)	17.4	5.2	14	0.67 (0.56, 0.78)	18.2	6.8	19
L Brachial PP-supine	0.87 (0.81, 0.92)	18.0	4.4	12	0.71 (0.60, 0.81)	18.3	6.5	18
R Ankle PP-supine	0.84 (0.77, 0.91)	21.4	8.0	22	0.68 (0.58, 0.79)	23.1	11.7	33
L Ankle PP-supine	0.81 (0.74, 0.89)	21.9	8.4	23	0.71 (0.61, 0.80)	22.3	10.7	30

Abbreviations: CoV = coefficient of variation (%); CI, confidence intervals; ICC = intraclass correlation coefficient (range: 0–1); L = Left; MDC₉₅ = minimal detectable change value based on 95% confidence interval (mm Hg); R = right.

As shown in Table 4, the measurement SEs followed a similar pattern in that between-visit SEMs were of greater magnitude than within-visit SEMs, but of comparable magnitude for the central and peripheral pressure measurements. When measurement variability was expressed as coefficients of variation (i.e., the SE of measurements as a percentage of the mean), little difference was observed comparing within-visit to between-visit estimates, but the magnitude of the coefficients of variation was consistently greater for the pulse pressure than for the systolic pressure measures.

The measurement variability patterns described above are reflected in the minimal detectable change at 95% confidence estimates (MDC₉₅) also shown in Table 4. MDC₉₅ values were smaller within-visit than between-visits, and central BP measures required larger MDC₉₅ values than other measures. As estimated in this sample of older adults, the MDC values attributed to an intervention were 39 mm Hg for cSBP and 32 mm Hg for cPP, whereas for the conventional sitting SBP and PP, the MDC₉₅ values were 30 mm Hg and 23 mm Hg, respectively.

Estimates of the minimum detectable mean difference with a 90% confidence level (MDD₉₀) for 2 samples are presented in Figure 1 (and Supplementary Table 2). For parsimony, MDD₉₀ values for the central, conventional sitting, and supine right-arm SBPs and PPs are shown by study size of the comparison groups. For illustration, the MDD₉₀ of cSBP indicates that for a study size of 100 in each of 2 samples, a mean difference in cSBP less than 12 mm Hg is attributable to random measurement error. After the sample size of about 500 per group, the required MDD₉₀ for all measures was less than 5 mm Hg.

Figure 1.

Minimal detectable difference for central and peripheral systolic and pulse pressure measures for a 2-sample design by sample sizes per group. *MDD₉₀ = minimal detectable difference with a 90% confidence level. Abbreviations: SBP, systolic blood pressure; PP, pulse pressure.

DISCUSSION

The profound effects of the central (aortic) pressure and its pulsatility on small vessel disease in the heart, brain, and kidneys highlight the importance of understanding the measurement properties of cSBP and cPP outside of the vascular laboratory.^18,19 We quantified the repeatability of cSBP and cPP estimates on 2 series of paired measurements separated by 4 to 8 weeks in a multicenter population study setting, using automated noninvasive devices. We also compared estimated repeatability of central pressure to that of standard peripheral pressures.

Our results indicate that the average of 2 measurements obtained with a standardized protocol and automated devices at a single examination can provide reliable estimates of central BP. The reliability coefficients of central and peripheral BP measures were comparable. Specifically, the ICC for the short-term (4 to 8 weeks) repeatability of cSBP and cPP were 0.62 and 0.66, respectively. The benchmark for ICC can be seen to vary in the literature^20,21 and no conventional ICC reference norms are available for BP measures. Despite the variable nature of BP and considering that the absolute difference of SBPs were observed to range from 7.5 to 15.5 mm Hg, the short-term (40 day) repeatability ICCs of cSBP and cPP was reasonable-to-good and the same-day repeatability ICCs (0.84 and 0.82) was good-to-high according to convention.²⁰ The similarity in the repeatability metrics across the central and peripheral measures is in part attributable to the lack of independence of their measurements (i.e., the calibration of the recorded carotid waveform with the simultaneously measured brachial pressure, and the estimation of cPP as the difference between cSBP minus supine right brachial DBP).

The estimated ICCs of cSBP and cPP observed in this study are generally lower than those reported by others (Supplementary Table 3), although previous studies used different observation times, measurement devices, and analytic techniques, making cross-study comparison difficult. Current guidelines do not specify a desirable interval length for repeated BP measurements to avoid external influences on BP levels.^7,8 We used a common short-term treatment period (4 to 8 weeks) as our follow-up period, and our mean follow-up time (40 days) is longer than most other studies (e.g., 10 days, or 2 consecutive days under laboratory conditions) which likely accounts for the difference in the magnitude of estimated ICCs. Also to be noted, we estimated 3 main sources of measurement variation whereas other studies estimated the 2 classic sources of error, namely intra- and inter-individual variability.²² While most previous repeatability studies used an indirect method of pulse recording at the radial artery calibrated with SBP and DBP, we used the direct method of pulse recording over the carotid artery, calibrated with mean arterial pressure and DBP. Although some systematic error likely occurs with all noninvasive devices, systematic errors should not affect measurement repeatability, the focus of this report.

Provided the use of a standardized protocol implemented by centrally trained technicians and following a rest period, hemodynamic variables are influenced by at least 3 factors: natural physiological variability of the variables over time, variability introduced by observers, and error introduced by the measurement device or its algorithms. It is apparent in our results that the impact of the second and third factors mentioned above is considerably smaller than the first. Further, considering that the majority of the observed short-term measurement variability was attributable to between-participant variability, central and peripheral pressure measurements appear to share similar sources of biologic, behavioral, and environmental sources of variability that warrant attention by researchers and clinical practitioners.

Although the pathophysiologic role of central hemodynamic measures and their clinical relevance are greatest in older age, the fast-growing demographic between 65 and 85 years is largely missing from the literature on the measurement properties of central pressure. This report is one of very few studies focused on this age group, as reflected in Supplementary Table 3. The information presented in this report contributes new information to clinical researchers and trialists whose field of work includes central pressures and their properties, particularly if focused on older adults.

Observed BP levels in our study can be further addressed by body posture and number of measurements used at a single examination. The sitting brachial BP levels were lower than those measured in the supine position as observed in other studies.^23,24 The short-term measurement variability did not differ visibly for measurements taken in the sitting or supine positions; however, concordance correlation between seated and supine right brachial SBP was 0.6 (95% confidence intervals: 0.4, 0.7) (data not shown). This observation further supports the recommendation that body posture and arm position should be reported for BP measurements.

As shown in Table 2, mean differences between paired measurements were of small magnitude and centered on zero but absolute differences in within-visit paired measurements ranged from 4.7 mm Hg to 10.4 mm Hg for SBP and 4.9 mm Hg to 9.4 mm Hg for PP. It is worth noting that the 2013 ESH/ESC guidelines recommend taking an additional reading if the first 2 sequential readings taken on the same day differ by ±5 mm Hg.⁸ No additional measures were indicated by the design of this repeatability study.

Overall these results support the use of well-calibrated automated devices in multicenter epidemiologic studies in order to reduce the observer-dependent measurement error associated with manual devices.²⁵ In this regard, the device used to estimate central BP is considered less operator-dependent than devices that use a hand-held tonometer.^26,27 It is important to note however that the noninvasive assessment of central BP is a relatively new technique, estimated at present using a number of different measurement devices and methods, without the benefit of consensus guidelines from international groups of experts. Evaluations of the various techniques in use and their measurement properties are still few.^13,28

We calculated MDC₉₅ and MDD₉₀ from our results, as statistics useful in the estimation of a study power, for study design and evaluation of study results. MDC₉₅ and MDD₉₀ set boundaries of the true change or difference in measured values as a function of variability of BP and the number of observations in 1- and 2-samples, respectively.

When BP measurements serve as an exposure, postestimation at the analysis stage such as regression calibration and the use of instrument variables, have been used to correct for bias attributable to nonsystematic measurement variability.²⁵ Concerns related to overcorrection using these approaches also have been raised.²⁹ Thus, knowledge of the potential sources of error, its magnitude and the direction of the deviation from the true value can additionally inform the design, conduct, and interpretation of clinical and epidemiologic studies.³⁰

The results of our study should be interpreted considering the following 2 potential limitations. Since our goal was to estimate repeatability in a general population of individuals aged 65 and over, our results may not generalize to younger ages nor to subpopulations such as persons with diabetes or other morbid conditions. A further constraint is that in the absence of the central DBP measurements, we used the right brachial DBP to calibrate cSBP and to approximate cPP. Although this may reduce the accuracy of the central PP values, the literature shows DBP values to be relatively uniform throughout the arterial tree, providing support for this approximation.

In conclusion, our results indicate that the average of 2 measurements obtained at a single visit using a standardized protocol and an automated oscillometric device can provide reliable estimates of cSBP and cPP. The number of reports on the measurement properties of central BP based on repeat measurements is small, and they are predominantly based on healthy, young, or middle-aged adults evaluated in a laboratory setting. We contribute new information on the reproducibility of central and peripheral BP among older adults, under multicenter field conditions. Our results can aid in the design of studies and the evaluation of their data.

SUPPLEMENTARY MATERIAL

Supplementary data are available at American Journal of Hypertension online.

DISCLOSURE

The authors declared no conflict of interest.