PULSE WAVE VELOCITIES DERIVED FROM CUFF AMBULATORY PULSE WAVE ANALYSIS: EFFECTS OF AGE AND SYSTOLIC BLOOD PRESSURE

Abstract

Pulse wave velocity (PWV), a measure of arterial stiffness, is an independent risk factor for cardiovascular morbidity and mortality. We investigated the relationship of ambulatory brachial cuff-based oscillometric PWV (oPWV) to two known correlates: age and brachial systolic blood pressure (SBP).

In 234 participants in the Masked Hypertension Study, we analyzed 7,284 validated hourly ambulatory SBP and oPWV readings using the Mobil-O-Graph monitor, which employs a proprietary pulse wave analysis (PWA) algorithm to determine oPWV. Carotid-femoral PWV (cfPWV) was also measured. Mixed linear models were developed to estimate oPWV from age and ambulatory SBP.

Participants were 34% male, with mean (SD) age 52.8 (9.9) years, SBP 123.8 (18.4) mmHg, and oPWV 7.6 (1.3) m/sec and cfPWV of 7.7 (1.7) m/sec. The relationship of oPWV to age and SBP was:

$$oPWV=\mathit{1.0796}-\mathit{0.0317}*\mathit{\text{Age}}+\mathit{0.0015}*{\mathit{\text{Age}}}^{\mathit{2}}+\mathit{0.0325}*SBP$$

Age uniquely accounted for an estimated 75% of the total variation of oPWV, while SBP uniquely accounted for 20%; these findings were confirmed in an external validation dataset. Together, age and SBP accounted for 99.1% of the total variance of oPWV, but (only) 40.2% of the variance of cfPWV. The correlation between oPWV and cfPWV was 0.58, but was only 0.11 after controlling for age and SBP.

We conclude that the Mobil-O-Graph’s oPWV is nearly completely explained by age and SBP and its relationship to cfPWV is due to their shared associations with age and SBP. Other hemodynamic variables derived from oscillometric PWA may be useful and deserve additional scrutiny.

Graphical Abstract

The progressive age-related increase in systolic blood pressure (SBP) and cardiovascular (CV) disease morbidity and mortality has been attributed, at least in part, to increasing arterial stiffness.^1,2,3,4,5 On the other hand, arterial stiffness is not constant, varying within individuals with fluctuating blood pressures (particularly SBP) and longer-term, with age; thus age, SBP, and arterial stiffness are involved in a “vicious cycle” or “feed-forward” relationship,^6,7,8 so that arterial stiffness not only causes an increase in SBP but SBP is an antecedent of arterial stiffness.^9,10 With increasing arterial stiffness, there is not only an increase in central and peripheral SBP and pulse pressure (PP) but also an increase in arterial wall pulsation and altered microcirculatory flow². PWV may thus be a global indicator of macro- and microcirculatory health³ as suggested by observations that high PWV predicts adverse CV events independent of traditional risk factors as well as chronic kidney disease progression,^11,12,13,14 cognitive function decline,^15,16,17,18 and incidence of glucose intolerance or diabetes.¹⁹

Interest in arterial stiffness has prompted the development of simple clinical indicators such as arterial pulse wave velocity (PWV), usually estimated from simple time and distance calculations between 2 easily accessible arterial sites, most commonly the carotid and femoral arteries (cfPWV).^1,20 These 2-point PWV techniques require considerable time, are subject to substantial errors in arterial length when determined via superficial anatomical landmarks, and remain operator-dependent. Thus, other approaches to arterial stiffness measurement have arisen, including arterial tonometry,^21,22 accelerometry,²³ Doppler flow,²⁴ piezo-electronic pressure transducers,²⁵ cardiac MRI,²⁶ and oscillometry.^{26,27,28,29,30,31}

The Mobil-O-Graph device (I.E.M., Industrielle Entwicklung Medizintechnik und Vertriebsgesellschaft mbH, Stolberg, Germany) conveniently combines cuff oscillometry and pulse wave analysis (PWA) to measure oscillometric PWV (oPWV) during 24-hour ambulatory BP monitoring.^28,29,32 Mobil-O-Graph oPWV results have been validated against intra-aortic catheter measurements²⁷ and correlate with cfPWV.^28,33,34 A consistent problem with this and other proprietary algorithmic methods is and the attendant lack of specific information on the actual determinants of each derived variable. The current study was undertaken to determine the degree to which Mobil-O-Graph oPWV values depend on 2 major determinants: age and cuff oscillometric SBP.

Methods

The data for the primary analyses and the external validation analysis, as well as the programming code used to perform the analyses for this study, are available from the corresponding author upon reasonable request.

As part of an NIH-funded study, 234 participants from Phase 2 of the Masked Hypertension Study³⁵ completed a 40-hour ambulatory blood pressure (ABP) monitoring study using the Mobil-O-Graph device. After being fitted with an appropriate-sized cuff on the non-dominant arm, the device measured their brachial blood pressure, heart rate, and central hemodynamics once an hour. Valid oPWV and brachial SBP values were present in 78% of the maximum of 9,360 readings (7,287 readings) including 7,182 observations with oPWV>5.0 m/sec and 105 with oPWV=5.0 m/sec, the recorded value for all assessments of ≤5.0 m/sec.

As part of a small pilot study, 35 individuals completed a 24-hour ABP recording using the Mobil-O-Graph device with readings taken every 30 minutes (the more common technique). A total of 1077 valid assessments of oPWV (>5.0 m/sec) were obtained and these data were used as an independent, external validation of the primary analysis.

Of the 269 individuals in the combined Masked Hypertension Substudy and external validation datasets, 188 had cfPWV assessments performed by trained technicians with a SphygmoCor device (AtCor Medical) that met recommended quality control criteria. These data were used to examine the association between cfPWV and oPWV, and their respective relationships to age and SBP.

The protocols for both studies were reviewed and approved by the Institutional Review Boards of Columbia University Medical Center and Stony Brook University, and all participants provided written informed consent.

Analysis.

Due to the repeated measures nature of the data (multiple ABP readings nested within persons), we used a multilevel model with oPWV as the outcome and no predictors in order to decompose the total variance into between-person and within-person components. Subsequently, two models were created that included age (a between-person factor) and brachial SBP (a within-person factor) as predictors; the first included only the linear effect of age, while the second added a quadratic term to account for non-linearity. The extent to which the model fit could be improved was then explored by the addition of other parameters assessed by the device. Model performance was examined in an independent external validation dataset of 35 persons who wore the device for 24 hours, with readings taken at 30-minute intervals. Finally, as a sensitivity analysis, the 105 observations with oPWV=5.0 (the left-truncated values) were included for estimation of a multilevel Tobit regression model to determine if the coefficients of age and SBP remained stable (reported in Supplemental Materials).

Standard linear regression methods were used to estimate the relationship of cfPWV to age, age², and SBP, and the proportion of variance explained by these three variables. Further examination assessed the correlation between the predicted cfPWV values based on this regression equation and the values that would be predicted based on Eq. 1 (i.e., the equation derived from the oPWV data). All analyses were performed using SAS, version 9.4 (SAS Institute, Cary, NC).

Results

Demographic characteristics of the study population are shown in Table 1.

Table 1.

Characteristics of the Population, N=234

Attribute	N (%)	Mean (SD)	Range
Male	79 (33.8%)
Black/African-American	20 (8.5%)
Hispanic	35 (15.0%)
Age, years		52.8 (9.9)	27, 78
BMI, kg/m2		27.9 (5.5)	18.7, 46.9
Systolic BP, mmHg*		124.2 (18.3)	71, 225
Diastolic BP, mmHg*		78.5 (14.2)	37, 140
oPWV, m/sec*		7.6 (1.3)	5.1, 13.2
cfPWV, m/sec†		7.7 (1.7)	4.2, 16.2

^*Mean of all 7,182 ambulatory readings where PWV>5.0 m/sec

^†N=164 with carotid-femoral PWV measurements meeting quality control criteria

In the 7,182 observations with oPWV>5.0 m/sec, mean (SD) oPWV was 7.64 (1.31) m/sec with a range of 5.1 to 13.2 m/sec. The mixed model with no predictors provided estimates of the between-person variance (1.57 m²/sec²) and within-person variance (0.21 m²/sec²); intra-class correlation was 0.885.

The mixed model predicting oPWV from age and SBP was extremely robust (R²=97.4%, Table 2, Model 1), but the residuals revealed a non-linear effect of age that was best captured by the addition of a quadratic term (Age²). The resulting equation,

$$oPWV=\mathit{1.0796}-\mathit{0.0317}*\mathit{\text{Age}}+\mathit{0.0015}*{\mathit{\text{Age}}}^{\mathit{2}}+\mathit{0.0325}*SBP$$ [Eq 1]

and the corresponding model characteristics are shown in Table 2 (Model 2). This equation accounts for 99.1% of total oPWV variability; 99.8% of the between-person variability, and 93.4% of the within-person variability. Race, ethnicity, sex, and BMI had no meaningful effect on this predictive model.

Table 2.

Parameter Estimates of Multilevel Linear Regression Models Predicting PWV from Age and Systolic Blood Pressure

Model	Prediction equation*				Residual variance
	Constant	BAge	BAge2	BSBP	Between- person	Within- person	Total†	R2
Model 0	7.5957	---	---	---	1.5718	0.2051	1.7769	---
Model 1	−2.5089	1.1596		0.3256	0.0334	0.0136	0.0470	0.974
Model 2	1.0796	−0.3172	0.1461	0.3253	0.0025	0.0136	0.0161	0.991

^*The coefficients of Age and Age² represent the estimated increase in PWV per 10-year increase in age; the coefficient of SBP represents the estimated increase in PWV per 10 mmHg increase in SBP. Model 0 includes no predictors (only the intercept; i.e., constant term); Model 1 includes Age and SBP as predictors; Model 2 also includes Age².

^†Total variance is calculated as the sum of the between- and within-person variances R² is calculated as 1 – (Total residual variance / Model 0 total variance)

The curvilinear relationship between age and oPWV is shown in Figure 1. The SBP-adjusted mean oPWV (for SBP=120 mmHg) for each person was obtained by removing the estimated effect of SBP (i.e., subtracting 0.0325×[SBP-120]), and calculating the mean of the results.

Figure 1.

Relationship of SBP-adjusted oPWV* with Age, N=234

* Oscillometric PWV (oPWV) measurements are adjusted to a systolic BP of 120 mmHg and the mean of these adjusted measurements for each individual is plotted against age.

Figure 2 shows histograms of the raw oPWV data and of the residuals from the above Model 2. Fully 36.6% of the residuals differ from zero by <0.05 m/sec, the rounding error in oPWV scores, which are only recorded to one decimal place. The mean absolute deviation of observed oPWV values from their predicted values was 0.095 m/sec and the equation predicted PWV to within ±0.5 m/sec for 99.8% of observations.

Figure 2.

Histograms of A) observed oscillometric PWV measurements and B) residuals after removing the effects of age and systolic blood pressure

To illustrate how oPWV varies as a near-perfect linear function of systolic BP for all individuals of the same age, we plotted the data for the 14 participants at the modal age of 55; the relationship of oPWV to SBP was essentially identical (Supplemental Figure S1). Another illustration of the impact of the interaction of age and SBP on oPWV is a graph of the within-individual results by decade of age (a 30-year old, a 40-year old, a 50-year old, a 60-year old, and a 70-year old; Supplemental Figure S2). The vertical difference between the resulting 5 parallel lines increases with each decade, reflecting the contribution of [age + age²] to oPWV.

Eq. 1was able to predict 99.1% of the variance in oPWV, which is very near the rounding error of ±0.05 m/sec. The addition of most other hemodynamic variables reported by the Mobil-O-Graph – brachial diastolic and mean arterial pressure (MAP), heart rate, central SBP and DBP, augmentation pressure, augmentation index, total vascular resistance, and cardiac output – had little effect on the model; however, addition of brachial PP, central SBP, and reflection magnitude improved the overall R² from 0.991 to 0.998.

In a sensitivity analysis, we computed 40-hour means of SBP and oPWV and estimated a linear regression model predicting mean oPWV from age, age², and mean SBP. The coefficients were very similar to those of Equation 1, each differing by less than half of one percent, with R²=0.998 for this analysis of aggregated data.

The relative contributions of age and SBP to the reported oPWV values in this sample were determined individually and together. Age directly accounted for 74.9% of the total variance, while SBP directly accounted for 20.0%. Age and mean ambulatory SBP were also correlated, accounting for an additional 4.2% of the total variance (thus explaining 99.1% of the total oPWV variance). The direct contribution of SBP could be split further into the portion due to between person-variation in mean SBP (9.2%) and within-person variation in SBP during the monitoring period (10.8%).

External validation

Compared to the main study population, the external validation set (Mobil-O-Graph for 24 hours with readings taken every 30 minutes) was younger (mean of 39.2 vs. 52.8 years old) and more diverse (26% vs. 8% black and 51% vs 15% Hispanic); mean SBP and DBP were similar but mean oPWV was lower (6.5 vs. 7.6 m/sec). oPWV predicted from the Model 2 equation also predicted their respective oPWV values, with nearly identical coefficients for [age + age²] as in the original model (overall R² = 98.5%, with the model accounting for 99.8% of the between-person variability and 90.2% of the within-person variability). The two sets of predicted scores (one using the original Equation 1 and the other from the corresponding equation for the external validation dataset) were nearly perfectly correlated (r=0.9999) and the mean absolute deviation of residuals from the refitted equation (0.102 m/sec) were also nearly identical to the main dataset (0.104 m/sec). Table 3 shows the distribution of these residuals, along with those from the original model.

Table 3.

Distributions of 3 sets of residuals

Residual Category (m/sec)	Percent* (N=7,182)	Percent† (N=1,077)	Percent‡ (N=1,077)
−0.75 to −0.85	0.01	---	---
−0.65 to −0.75	0.01	---	---
−0.55 to −0.65	0.04	---	---
−0.45 to −0.55	0.17	0.09	---
−0.35 to −0.45	0.64	0.28	0.37
−0.25 to −0.35	2.37	1.95	1.39
−0.15 to −0.25	7.39	8.64	4.64
−0.05 to −0.15	20.25	24.88	20.24
−0.05 to +0.05	36.65	31.94	32.87
0.05 to 0.15	22.58	19.78	24.70
0.15 to 0.25	7.41	7.99	9.84
0.25 to 0.35	1.85	2.97	3.53
0.35 to 0.45	0.49	1.02	1.58
0.45 to 0.55	0.13	0.37	0.65
0.55 to 0.65	0.01	0.09	0.09
0.65 to 0.75	---	---	0.09

^*Residuals in original data for model fit on original data (N=234 participants)

^†Residuals in validation data for model fit on validation data (N=35 participants)

^‡Residuals in validation data for model fit on original data (N=35 participants)

Prediction of cfPWV by age and SBP

In the pooled dataset (N=188), the Pearson correlations of age and SBP with cfPWV were 0.51 and 0.54, respectively (both p<0.001). In this sample, our model with age, age², and SBP accounted for 40.2% of the variance in cfPWV, leaving 60% unexplained by these variables. Supplemental Figure S3 presents residual plots for oPWV (panel A) and cfPWV (panel B) after removing the effects of age, age², and concurrent SBP from each. Clearly there is much more variance in cfPWV that is independent of age and SBP.

BP and oPWV assessment were also taken in the clinic setting as the first value of the ABPM series, with the participant seated. Figure 3, Panel A shows a scatterplot of these assessments versus cfPWV, with a correlation coefficient of r=0.58 (p<0.001). Panel B shows the “same” scatterplot after the effects of age, age², and SBP have been removed from each of the two measures. The remaining variability in oPWV was very poorly correlated with the remaining variability in cfPWV (r=0.11, p=0.09). However, one limitation of this comparison is that the two assessments were not taken on the same day; (mean interval between assessments was 149 (SD=124) days). After restricting the analysis to participants whose interval between assessments was less than 3 months (N=88), the correlation of oPWV and cfPWV improved from r=0.58 to r=0.63, but when the effects of age, age², and SBP were removed from both, there was essentially no correlation (r=0.06, p=0.49).

Figure 3.

Relationship of oscillometric PWV with carotid-femoral PWV, before (A) and after (B) the effects of age, age2, and systolic BP have been removed (N=188)

Discussion

Data from 40-hour studies with the Mobil-O-Graph in 234 subjects, yielding 7,284 adequate readings, demonstrated that both age and SBP were major determinants of oPWV according to the equation: Predicted oPWV = 1.0796 – 0.0317*Age + 0.0015*Age² + 0.0325*SBP. The multiple R² for this relationship was 0.991, indicating that<1% of the observed variation in oPWV was independent of age and measured ambulatory SBP. Overall, age accounted for about 75% of the variation, while SBP accounted for about 20%; roughly half of the latter was attributable to between-person variation in mean SBP, while the other half was attributable to within-person variation in SBP during the monitoring period. Most of the total residual variance of oPWV (0.0161 m²/sec² for this model) was found to be within-person (0.0136 m²/sec²) rather than between-person (0.0025 m²/sec²).

Other studies support the present findings. The ARCsolver algorithm for oPWV, which is used in the Mobil-O-Graph, has previously yielded very high correlations of oPWV with age (r=0.94) and BP category.³¹ Furthermore, the observed association of SBP with oPWV is strikingly similar to that from a preliminary report in a small convenience sample; when this preliminary study was later expanded, a pattern extremely similar to the present study was revealed.³⁶ Further evidence that our results can be generalized can be found in the manufacturer’s “arterial stiffness” nomogram, which clearly demonstrates a tight upwardly concave relationship of oPWV with age, with very narrow “normal range” around the age-related continuous mean value curve; this nomogram is highly consistent with our model.³² Since there is so little “left over” variation in oPWV, it is extremely unlikely that oPWV offers any additional clinically relevant information, including incremental prognostic information about CV outcomes, beyond that already offered by age, age² and SBP.^10,37,38

We have not studied extensively the relationship between oPWV and traditional 2-point (time and distance) techniques such as cfPWV. Present results, however, strongly support the idea that cfPWV yields different information than oPWV. In a large meta-analysis, cfPWV demonstrated independent risk predictive power even after adjustment for age and SBP but no outcome study has been performed with oPWV that controlled for age and SBP. It is still hoped that a simple, quick method for measuring PWV might improve risk stratification and influence treatment decisions³⁹ but the latter in particular remains elusive.

Decisions about whether a given clinical indicator is useful in clinical and research settings cannot depend on simple correlations with other methods.⁴⁰ Indeed, correlations among different PWV methods have in general been somewhat inconsistent, as exemplified by PWV values derived from radial tonometry being relatively poorly correlated (r=0.28 to 0.67) with standard time-distance methods;⁴¹ the correlation coefficient of oPWV with cfPWV is within this range. A correlation of cfPWV and mean BP has been reported and a quadratic relationship with age was also noted,⁴² but the variance of cfPWV was much larger in that study than we found. A possible limitation of this study is that both the primary sample and the external validation sample were younger and healthier than prior PWV study populations cited.^28,33,34

In summary, the extremely strong relationships of age and SBP to oPWV in this study, coupled with the virtual absence of additional unexplained variation and the very weak partial correlation of oPWV with cfPWV after controlling for age and SBP make it unlikely that oPWV could possibly provide useful incremental information, including for CV outcomes, beyond that provided by age and SBP.

Perspectives

Studies of pulse wave velocity (PWV) are of interest to researchers and clinicians because PWV is believed to yield unique information about the stiffness of large arteries. It would be convenient if a cuff-based system could replace the historical but inconvenient time-and-distance methods for PWV but the current study clearly shows that the oscillometric-based PWV measure employed in this cohort is nearly perfectly predicted (>99%) by an algorithm that includes only the patient’s age and the simultaneously assessed systolic BP.

Novelty and Significance.

What is new? This study shows the major limitations of a novel oscillometry-based measure of pulse wave velocity provided by an automated brachial cuff blood pressure monitor. Although pulse wave velocity, a measure of arterial stiffness, has been shown to be an independent predictor of cardiovascular risk, assessment by this method is a near-perfect function of the patient’s age (which is entered when initializing the device) and his/her brachial systolic blood pressure at the time of the reading.

What is relevant? This study shows that the use of this oscillometry-based pulse wave velocity measure, albeit convenient, provides virtually no useful information independent of age and brachial systolic blood pressure.

Summary. We conclude that the oscillometry-based pulse-wave velocity measure should not be relied on to predict outcomes, including cardiovascular risk, beyond what is already provided by age and brachial systolic blood pressure.