Prognostic significance of mechanical biomarkers derived from pulse wave analysis for predicting long-term cardiovascular mortality in two population-based cohorts

Abstract

Background:

Numerous mechanical biomarkers derived from pulse wave analysis (PWA) have been proposed to predict cardiovascular outcomes. However, whether these biomarkers carry independent prognostic value and clinical utility beyond traditional cardiovascular risk factors hasn’t been systematically evaluated. We aimed to investigate the additive utility of PWA-derived biomarkers in two independent population-based cohorts.

Methods:

PWA on central arterial pressure waveforms obtained from subjects without a prior history of cardiovascular diseases of two studies was conducted based on the wave transmission and reservoir-wave theory: firstly in the Kinmen study (1272 individuals, a median follow-up of 19.8 years); and then in the Cardiovascular Disease Risk Factors Two-Township Study (2221 individuals, median follow-up of 10 years). The incremental value of the biomarkers was evaluated by net reclassification index (NRI).

Results:

In multivariate Cox analyses accounting for age, gender, body mass index, systolic blood pressure, fasting glucose, high-density- and low-density-lipoprotein cholesterol, and smoking, only systolic (SC) and diastolic rate constant (DC) of reservoir pressure could independently and consistently predict cardiovascular mortality in both cohorts and the combined cohort (SC: hazard ratio 1.18 [95% confidence interval 1.08–1.28, p < 0.001; DC: 1.18 [1.09–1.28], p < 0.001]. Risk prediction estimates in traditional risk prediction models were significantly more accurate when incorporating peak of reservoir pressure (NRI = 0.049, p = 0.0361), SC (NRI = 0.043, p = 0.0236) and DC (NRI = 0.054, p = 0.047).

Conclusions:

Of all PWA-derived biomarkers, SC and DC were consistently identified as valuable parameters for incremental cardiovascular risk prediction in two large prospective cohorts.

1. Introduction

The arterial system functions both as a conduit and as a reservoir [1]. Vascular aging and its associated arterial stiffening have a major impact on the arterial pressure waveform and cardiovascular risk. The peak and trough of arterial pressure waveforms, systolic and diastolic blood pressure (BP), have been linearly associated with the risk of cardiovascular events [2]. The Windkessel and wave transmission theories have been used to model the complex interactions between the heart and the arterial system, and both distinctively provide explanations on the changes in the arterial pressure waveform with aging and diseases. Based on the wave transmission theory, the arterial pressure waveform is composed of a forward wave and a reflected wave. Increased wave reflection measured by the pulse wave analysis (PWA) is related to the extent of myocardial ischemia in patients with [3] and without [4] obstructive coronary artery disease. The pressure wave reflection is also directly related to left ventricular hypertrophy [5] and its regression with treatment [6], to left atrial size [7], and is inversely related to left ventricular diastolic function at rest [7] and during exercise [8]. An increase in wave reflection has been associated with a decreased renal function [9] and with an impaired outcome following acute ischemic stroke [10]. Using a triangulation method, we were able to decompose the central pressure waveforms derived by tonometry into their forward wave amplitudes and backward wave amplitudes (Pb) and demonstrate that Pb predicts long-term cardiovascular mortality in men and women independent of arterial stiffness [11,12].

Recently, the reservoir function of vascular systems has been modeled by Wang et al. as the volume related pressure changes in arterial [13] and venous vascular systems [14]. Aortic stiffening with advancing age not only leads to an accelerated wave speed and more pronounced wave reflections [11], but also to a reduced reservoir function. Parameters based on the reservoir-wave concept that combines elements of wave transmission and Windkessel models of arterial pressure generation were shown to predict clinical outcomes in hypertensive patients in two randomized control trials, ANBP2 [15] and ASCOT-CAFÉ studies [16].

As a result, current PWA can provide parameters based on both the wave transmission theory and the reservoir-wave analysis. However, whether these biomarkers confer independent prognostic value beyond traditional cardiovascular risk factors, has not been comprehensively and systematically evaluated.

Although numerous novel biomarkers have been proposed in various study populations, reproducibility is vital to assure that previous results are reliable and valid and could be applied to new situations. Results should be repeated in studies with different subjects and experimenters. Therefore, the aim of the present study was to examine the prognostic value and the incremental clinical utilities of these PWA-derived mechanical biomarkers in two independent population based cohorts.

2.1. Study population

The characteristics of participants and patient enrollment process of the first cohort from Kinmen have been detailed elsewhere [17]. In brief, a previous community-based survey was conducted in 1992–1993 in Kinmen [18], from which a total of 1272 normotensive and untreated hypertensive Taiwanese subjects (47.0% women, mean age 52 ± 13 years old, range 30–79 years) were drawn after excluding participants with a previous history of diabetes mellitus, angina pectoris, peripheral vascular disease, and any clinical or echocardiographic evidence of other significant cardiac diseases. The study protocol was approved by the institutional review board at the Johns Hopkins University and Informed consent was obtained from all participants.

To demonstrate the reproducibility, the prognostic value of all PWA-derived biomarkers was then examined in the second cohort from the “Cardiovascular Disease Risk Factors Two-Township Study” (CVDFACTS), a community-based follow-up study focusing on risk factor evaluation and cardiovascular disease development in Taiwan [19, 20]. Of the participants in CVDFACTS, a total of 2211 individuals without a prior history of cardiovascular diseases had undergone measurements of arterial tonometry during their cycle 4 examination (1997–1999).

2.2. Data collection

After being seated for at least 5 min, measurements of brachial systolic BP (SBP) and diastolic BP (DBP) from the right arm of subjects were manually taken with a mercury sphygmomanometer and a standard-sized cuff (13 cm × 50 cm). Reported blood pressures represent the average of at least two consecutive measurements separated by at least 5 min. Brachial pulse pressure (PP) was calculated as [brachial SBP – brachial DBP]. Overnight fasting serum and plasma samples were drawn for blood chemistry analysis.

2.3. Central arterial pulse wave analysis (Fig. 1)

Fig. 1.

(A) Parameters obtained from the triangulation method for wave separation. Dashed line: forward pressure wave; dash-dotted line: backward pressure wave. (B) Parameters obtained from the arterial reservoir model. Solid line: carotid pressure waveform; dashed line: reservoir pressure. XSPI = excess pressure integral; PRI = reservoir pressure integral. Solid line: carotid pressure waveform; dashed line: forward pressure wave (Pf); dashed-dot line: back pressure wave (Pb).

In the Kinmen study, right carotid artery pressure waveforms, which have been demonstrated to closely resemble central aortic pressure waveforms [21–23], were registered noninvasively with a tonometer [18,24]. The common carotid artery pressure waveforms were calibrated with brachial mean BP (MBP) and DBP to obtain the carotid BP [21]. Recently, it has been argued that MBP should be estimated as brachial DBP along with 40% of brachial artery pulse pressure [25]. We therefore conducted a sensitivity analysis by using a brachial form factor of either 0.33 or 0.4 to evaluate the impact of different calibrating methods on the prognostic significance of these mechanical biomarkers (Appendix I). Because the prognostic value (hazard ratio and Wald Chi²) were comparable between these two methods, we therefore only present the results of a traditional calibrating method with a form factor of 0.33. In the CVDFACTS study, central aortic pressure waveforms were obtained with a SphygmoCor device (AtCor Medical, Sydney, Australia) using radial arterial pressure waveforms and a validated generalized transfer function according to the manufacturer’s instructions [26]. Radial arterial pressure waveforms, obtained by applanation tonometry using a solid-state high-fidelity external Millar transducer, were calibrated with cuff SBP and DBP values, and then mathematically transformed by the validated transfer function [26] into corresponding central aortic pressure waveforms. Central PP was calculated as central SBP – central DBP.

Digitized central arterial pressure waveform signals were analyzed using custom-designed software on a commercial software package (Matlab^®, version 4.2 and 7.0, The MathWorks, Inc.). To avoid inter- and intra-observer variations, we performed a fully automatic batch analysis for all processed individual signals. The inflection point was recognized on the calibrated central arterial pressure waveform using the zero-crossing timings of the fourth derivative of the pressure wave [27]. Augmentation index (AI) was calculated accordingly [23]. Forward and reflected components of the central arterial pressure waveform were separated using the triangulation method based on the following equations [28]:

$$\mathrm{P}\mathrm{f}\left(\mathrm{t}\right)=\left[\mathrm{P}\mathrm{m}\right(\mathrm{t})+\mathrm{Z}\mathrm{c}\times \mathrm{F}(\mathrm{t}\left)\right]/2$$

$$\mathrm{P}\mathrm{b}\left(\mathrm{t}\right)=\left[\mathrm{P}\mathrm{m}\right(\mathrm{t})-\mathrm{Z}\mathrm{c}\times \mathrm{F}(\mathrm{t}\left)\right]/2$$

where $\mathrm{P}\mathrm{m}\left(\mathrm{t}\right)$ is the central arterial pressure wave, $\mathrm{F}\left(\mathrm{t}\right)$ is the approximated triangular-shaped flow wave, $\mathrm{Z}\mathrm{c}$ is the characteristic impedance, $\mathrm{P}\mathrm{f}\left(\mathrm{t}\right)$ is the forward pressure wave, and $\mathrm{P}\mathrm{b}\left(\mathrm{t}\right)$ is the backward pressure component. $\mathrm{P}\mathrm{f}$ and $\mathrm{P}\mathrm{b}$ are the pressure amplitudes of $\mathrm{P}\mathrm{f}\left(\mathrm{t}\right)$ and $\mathrm{P}\mathrm{b}\left(\mathrm{t}\right)$, respectively. The accuracy of the triangulation method in obtaining the magnitude of $\mathrm{P}\mathrm{f}$ and $\mathrm{P}\mathrm{b}$ has been validated previously [11,28]. Reflection magnitude (RM) was calculated as $\mathrm{P}\mathrm{b}/\mathrm{P}\mathrm{f}$.

2.4. Reservoir-wave analysis

Reservoir-wave analysis is a time-domain analysis, based on the understanding that the arterial system not only functions as a conduit system, conducting waves to periphery, but also as a hydraulic integrator, by which the arterial system charges during systole as inflow exceeding outflow, and discharges during diastole, through recoiling of the elastic arteries, continuously supplying blood to vital organs, notably the brain, heart and kidney [13]. In reservoir-wave analysis, central arterial pressure is presumed to be the instantaneous sum of a pressure associated with aortic blood volume and compliance (Preservoir), and a pressure due to propagation characteristics (Pexcess) [13]. Reservoir pressure was calculated from the ensemble averaged central arterial tonometric waveforms. Reservoir pressure is assumed to vary temporally in the same way throughout the aorta and large elastic arteries, but with a time lag that depends on the location and wave propagation characteristics of the arteries [29,30]. Details of the calculations of parameters of reservoir pressure are presented in Appendix II. In the present study, we also provided (in Table S2) the reference value of these parameters obtained previously [13,29,30] with invasive central aortic pressure and flow waveforms (Appendix III). Parameters calculated by using the pressure alone reservoir-wave analysis have been compared with those calculated by using the true aortic flow wave derived from Doppler echocardiography in another randomly selected 202 subjects in our laboratory (Appendix IV). Calculations of these parameters were in good agreement between these two approaches, with intra-class correlation coefficients ranging from 0.77 to 0.99 (Table S4 and Fig. S1: Bland–Altman plots in Appendix IV). The systolic rate constant (SC) of the reservoir pressure waveform represents the rate constant for reservoir filling and is inversely related to the product of aortic characteristic impedance and total arterial compliance. In contrast, diastolic rate constant (DC) represents reservoir emptying and is inversely related to the product of systemic arterial resistance and total arterial compliance. As shown in Fig. 2, an increase in SC would increase the magnitude and the rate of upstroke of reservoir pressure. Here we demonstrated that when an aortic reservoir is charged to a level (arbitrary), a decrease in DC would decrease the rate of discharging of aortic reservoir.

Fig. 2.

Effect of changes in systolic rate constant (SC) and diastolic rate constant (DC) on reservoir pressure waveform. The reservoir pressure waveform is represented by the bold solid line. (A) An increase in SC results in an upward shift of the ascending portion of the pressure waveform. (B) In contrast, a decrease in DC results in upward shift of the descending portion of the pressure waveform.

2.5. Follow-up

By linking our databases with the National Death Registry, we retrieved the dates and causes of death of all participants in our cohort from the National Death Registry database based on the certified death certificates coded according to the International Classification of Diseases, Ninth Revision (ICD-9). Cardiovascular deaths were identified with the ICD-9 codes 390–459. The accuracy of cause-of-death coding in Taiwan’s National Death Registry database has been validated. Individuals that did not appear in the National Death Registry on censoring dates were considered to be survivors.

2.6. Statistical analysis

Data are presented as percent or mean ± standard deviation. To evaluate the prognostic value of all PWA-derived biomarkers, we first categorized all subjects into quartiles by the biomarker values and calculated their corresponding cardiovascular mortalities. Then we used Cox proportional hazard regression analyses to identify predictors of cardiovascular mortality in these two population-based cohorts, respectively. A baseline multivariate model was then constructed by including conventional cardiovascular risk factors: age, sex, systolic BP, body mass index, fasting glucose, triglycerides, low-density-lipoprotein cholesterol, high-density-lipoprotein cholesterol, smoking, and alcohol. Based on the baseline multivariate model, parameters of all PWA-derived biomarkers were then examined for their adjusted hazard ratios and 95% confidence intervals per one-standard deviation increment in the specific cohort and in combination. We employed net reclassification index (NRI) to evaluate the reclassification effects of the independent prognostic biomarkers identified from the combined analyses for predicting future cardiovascular events [31–33]. Statistical analyses were performed using SAS 9.1 software. A two-tailed p < 0.05 was considered statistically significant.

3. Results

A total of 3483 subjects with 1272 subjects in the Kinmen study and 2211 subjects in the CVDFACTS study were included. The baseline characteristics of the participants and results of the central pressure waveform analysis are presented in Table 1. Cardiovascular mortality was higher in Kinmen due to a higher hypertension prevalence and more male participants in the Kinmen cohort than in the CVDFACTS cohort (4.0/1000 person-year vs. 1.6/1000 person-year). In the Kinmen study, during a median follow-up of 19.8 years, 315 (26.9%) deaths occurred (84 of cardiovascular origin). In the CVDFACTS study, a total of 171 deaths occurred (34 of cardiovascular origin) during a median follow-up of 10 years.

Table 1.

Characteristics of study population.

	Kinmen study (N = 1272)	CVDFACTS study (N = 2211)
Age, years	52.3 ± 12.8	53.4 ± 12.0
Men, %	54%	45.8%
Brachial SBP, mm Hg	139.0 ± 23.3	122.2 ± 17.0
Brachial DBP, mm Hg	88.2 ± 15.1	68.1 ± 10.3
Central SBP, mm Hg	126.4 ± 23.9	111.6 ± 16.2
Central PP, mm Hg	42.0 ± 16.1	41.5 ± 10.9
BMI, kg/m2	24.7 ± 3.6	24.2 ±3.2
FBS, mg/dl	99.8 ± 22.6	96.5 ± 9.4
LDL-C, mg/dl	122.4 ± 34.1	122.0 ± 37.2
HDL-C, mg/dl	50.8 ± 12.8	47.9 ± 16.8
Triglyceride, md/dl	125.3 ± 96.5	117.7 ± 77.4
Pf, mm Hg	36.3 ± 12.7	25.5 ± 6.1
Pb, mm Hg	18.6 ± 8.7	16.4 ± 5.1
Augmented pressure, mm Hg	14.9 ± 10.4	12.4 ± 6.6
AI	0.25 ± 0.21	0.40 ± 0.16
RM	0.51 ± 0.14	0.64 ± 0.11
Peak reservoir pressure	115.6 ± 21.8	104.9 ± 14.9
Amplitude of reservoir pressure	31.2 ± 13.1	33.0 ± 9.3
XSPI, mmHg-s−1	3.4 ± 1.6	2.1 ± 0.8
PRI, mm Hg-s−1	11.2± 4.8	12.6 ± 4.0
Systolic rate constant, s−1	18.6 ± 9.4	16.2 ± 5.8
Diastolic rate constant, s−1	4.3 ± 0.50	2.8 ± 1.1

BMI = body mass index; FBS = fasting blood sugar; LDL-C = low density lipoprotein cholesterol; HDL-C = high density lipoprotein cholesterol; Pf = forward pressure amplitude; Pb = backward wave amplitudes; SBP = systolic blood pressure; PP = pulse pressure; XSPI = excess pressure integral; PRI = reservoir pressure integral; AI = augmentation index; RM = reflection magnitude.

Increased brachial systolic BP, pulse pressure, Pb, and AI were significantly associated with increased cardiovascular mortality in both studies. Similarly, biomarkers derived from reservoir pressure-wave analysis were also positively associated with cardiovascular mortality. In Table 2, the age and gender adjusted hazard ratios for predicting cardiovascular mortality were examined firstly in Kinmen and then in CVDFACTS studies (Table 2). In the age and gender adjusted multivariate model, all parameters derived from the central pulse wave analysis in the Kinmen study had significant associations with cardiovascular mortality. In the CVDFACTS study, only peak of reservoir pressure and DC remained significant in predicting cardiovascular mortality (Table 2).

Table 2.

The hazard ratios of various hemodynamic parameters for predicting cardiovascular mortality in age and gender adjusted model.

	Kinmen study				CVDFACTS study
	Hazard ratio, per SD	95% confidence intervals		X2	p-Value	Hazard ratio, per SD	95% confidence intervals		X2	p-Value
Brachial PP	1.46	1.31	1.62	49.39	<0.0001	1.359	0.951	1.941	2.8322	0.0924
Pf	1.322	1.091	1.601	8.1115	0.0044	1.092	0.788	1.514	0.2795	0.5970
Pb	1.655	1.362	2.011	25.685	<0.0001	1.152	0.831	1.597	0.7162	0.3974
Augmented pressure	1.684	1.373	2.065	25.0715	<0.0001	0.956	0.726	1.260	0.1006	0.7511
AI	1.701	1.278	2.264	13.241	0.0003	1.288	0.783	2.120	0.9915	0.3194
RM	1.455	1.194	1.774	13.7741	0.0002	1.298	0.807	2.087	1.1596	0.2815
Peak of reservoir pressure	1.953	1.599	2.386	42.9522	<0.0001	1.502	1.055	2.139	5.0819	0.0242
Amplitude of reservoir pressure	1.450	1.224	1.718	18.4218	<0.0001	1.184	0.857	1.637	1.0503	0.3054
PRI	1.273	1.093	1.482	9.6269	0.0019	1.011	0.751	1.362	0.0053	0.9418
XSPI	1.147	1.017	1.293	5.0300	0.0249	0.949	0.687	1.312	0.0996	0.7523
Systolic rate constant	1.573	1.325	1.866	26.8654	<0.0001	1.298	0.973	1.733	3.1403	0.0764
Diastolic rate constant	1.762	1.475	2.104	38.9680	<0.0001	1.363	1.001	1.856	3.8751	0.0490

SD = standard deviation; Pf = forward pressure amplitude; Pb = backward wave amplitudes; SBP = systolic blood pressure; PP = pulse pressure; XSPI = excess pressure integral; PRI = reservoir pressure integral; AI = augmentation index; RM = reflection magnitude.

The associations between all mechanical biomarkers derived from pulse wave analysis and cardiovascular mortality was then examined in the multivariate Cox proportional hazards models that took into account cardiovascular risk factors including age, sex, systolic BP, body mass index, fasting glucose, triglycerides, low-density-lipoprotein cholesterol, and high-density-lipoprotein cholesterol, and smoking. As illustrated in Figs. 3 and 4, Pb, augmented pressure, AI, RM, peak of reservoir pressure, amplitude of reservoir pressure, SC, and DC were significant prognostic biomarkers in the Kinmen study when each of these parameters was incorporated into the above multivariate model. In the CVDFACTS cohort, only SC and DC remained significant in predicting cardiovascular mortality after accounting for the above covariates. In the combined cohort, AI, RM, SC, and DC were independent predictors, even after adjusting for the widely used cardiovascular risk factors.

Fig. 3.

The hazard ratio (per standard deviation) of each pulse-wave-analysis derived mechanical biomarker by the wave transmission model in the multivariate Cox proportional hazards model accounting for age, gender, body mass index, systolic blood pressure, triglycerides, HDL-cholesterol, glucose, and smoking. AI = augmentation index; Pf = forward pressure amplitude; Pb = backward wave amplitude; RM = reflection magnitude.

Fig. 4.

The hazard ratio (per standard deviation) of each pulse-wave-analysis derived mechanical biomarker by the reservoir pressure model in the multivariate Cox proportional hazards model accounting for age, gender, body mass index, systolic blood pressure, triglycerides, HDL-cholesterol, glucose, and smoking. PRI = reservoir pressure integral; XSPI = excess pressure integral.

The clinical utility of the PWA-derived biomarkers was then evaluated with NRI in the combined cohort. In the risk assessment for cardiovascular mortality, adding peak of reservoir pressure, SC, and DC into the above multivariate model resulted in significant net incremental improvement (Table 3).

Table 3.

The reclassification and discrimination of the risk prediction models by incorporating individual pulse-wave-analysis derived mechanical biomarkers in the combined cohort.

	Event NRI	No event NRI	Overall NRI	95% confidence interval	p-Value
Pf	0.000	− 0.010	−0.010	−0.047–0.028	0.6164
Pb	0.009	− 0.003	0.007	−0.025–0.039	0.6848
Augmented pressure	0.000	− 0.002	−0.002	− 0.006–0.001	0.1615
AI	−0.019	− 0.007	−0.026	− 0.053–0.001	0.0589
RM	− 0.019	− 0.007	−0.026	− 0.053–0.001	0.0621
Peak of reservoir pressure	0.037	0.011	0.049	0.003–0.094	0.0361
Amplitude of reservoir pressure	0.019	− 0.002	0.017	− 0.035–0.069	0.5249
PRI	0.000	− 0.004	−0.004	− 0.063–0.054	0.8825
XSPI	0.028	0.004	0.032	−0.017–0.081	0.2033
Systolic rate constant	0.037	0.005	0.043	0.006–0.080	0.0236
Diastolic rate constant	0.037	0.016	0.054	0.001–0.107	0.0469

NRI= net reclassification index.

Models were adjusted for age, sex, systolic blood pressure, body mass index, fasting glucose, triglycerides, low density-lipoprotein cholesterol, and high density-lipoprotein, alcohol, and smoking.

AI=augmentation index; Pf=forward pressure amplitude; Pb=backward wave amplitudes; PRI = reservoir pressure integral; RI = reflection index; RM = reflection magnitude; SBP = systolic blood pressure; SD = standard deviation; XSPI = excess pressure integral.

The subjects were classified by risk of less 2% for low, between 2% and 6% for middle and more than 6% for high CVD death risk group. Net reclassification index was test by 95% confidence intervals with two tails.

4. Discussion

In these two large prospective cohorts, the prognostic value and clinical utility of the mechanical biomarkers derived from pulse wave analysis based on wave transmission theory and arterial reservoir pressure model were systematically and comprehensively evaluated in an objective manner. These two studies are inception cohorts as subjects of the present study were free from any cardiovascular diseases and were not taking any cardiovascular medications at enrollment into the studies. Moreover, without the knowledge of the outcomes, all parameters for evaluation were obtained with a batch pulse wave analysis. In the combined cohort after accounting from the widely used cardiovascular risk factors, AI, RM, SC and DC of the reservoir pressure remained significantly for predicting cardiovascular mortality. Of all the PWA-derived biomarkers, only SC and DC could significantly predict cardiovascular mortality in both cohorts, and demonstrated significant reclassification clinical utility in the combined cohort.

4.1. Wave transmission theory

Although the prognostic significance of parameters derived from wave transmission theory has been evaluated in many studies, few studies addressed the clinical utility of these parameters. A previous systematic review has demonstrated that, despite the high heterogeneity, a 10% increase of central AI was associated with a RR of 1.264 for cardiovascular events and 1.384 for total mortality [34]. Our study findings agree with this review by showing that AI was associated with a hazard ratio of 1.34 (95% CI 1.01–1.77) in Fig. 3.

Conventional pulse wave analysis, based on wave transmission theory, involves investigation of pressure waves alone from the carotid artery, or radial and brachial artery with or without the use of a transfer function. In a recent position statement from the European Society of Cardiology Working group on peripheral circulation [35], using the surrogate endpoint criteria with 7 items, wave reflection by classic pulse wave analysis or wave separation analysis has been graded as Recommendation: IIb and Level of evidence: B. It was noted that the incremental prognostic value was less consistent in broader populations; some studies presented positive promising results [11,36], while there were still some negative associations being reported [37]. Our findings are in line with the above speculations: all wave reflection parameters were positively associated with adverse cardiovascular outcomes in the Kinmen study, but failed to reach statistical significance in the multivariate model in the CVDFACTS study. This position statement, however, had not addressed the usefulness of the arterial reservoir model, probably because it is presently a newly emerging and highly debated concept [38–42].

4.2. Arterial reservoir pressure

The arterial reservoir pressure was initially proposed to resolve the waveform differences between the pressure and flow waves of the human vascular system [13,38]. It is a time-domain approach based on the rationale that the volume-related pressure change (reservoir) is the major contributing factor to such waveform differences. Subsequently, this assertion was supported in a clinical study which invasively measured the arterial pressure waveform and Doppler flow velocity of the central aorta [43].

Two previous trials on hypertensive patients, ANBP2 [15] and ASCOT-CAFÉ studies [16], had examined the prognostic utility of the reservoir function. These studies demonstrated that the SC and XSPI could provide incremental prognostic value to conventional cardiovascular risk factors. In addition, in a heterogeneous high risk patient population, the amplitude of the reservoir pressure was shown to be a significant predictor for cardiovascular events [44]. Our study extended the prognostic utility of the arterial reservoir function to two community based population with long-term follow-up. Besides the very long-term follow-up duration, the strength of our study is that the independent prognostic value of the reservoir-wave model was not only noted in the Kinmen study, but also could be reproduced in another independent CVDFACTS cohort. These findings are physiologically plausible because deterioration of the arterial reservoir function is expected to produce a larger reservoir pressure wave with accelerated reservoir filling rate (increased SC) and faster reservoir emptying rate (increased DC) as presented in Fig. 2. In other words, with a constant inflow and outflow, the reduced arterial compliance should be characterized by a faster upslope and downslope change of the reservoir pressure wave. Moreover, the incremental prognostic value demonstrated by the reclassification analysis lends solid support to its clinical application of the arterial reservoir model. These findings suggest that the arterial reservoir function, as assessed by the reservoir pressure wave from the arterial reservoir model, contains important prognostic information that should be considered in arterial risk factors for CV mortality.

Although the arterial reservoir model provides a physiological interpretation for the central aortic pressure waveforms, it was argued in recent computer modeling studies [42,45], that the arterial reservoir model or the three-element Windkessel model is too simplified a model to be useful in complex clinical setting. In spite of the data presented above, the opponents of this concept also suggested that the discrepancy noted in the simulation studies does not necessarily preclude the application of the Windkessel or reservoir-wave paradigm from a clinical perspective [42]. Being inspired by the famous quote “Remember that all models are wrong; but some are useful” by Prof. George E.P. Box, we conducted the present study based on the concept of pragmatism. Indeed, through an objective, neutral, and rigorous analytic framework for all the PWA-derived biomarkers, we have demonstrated the incremental prognostic value and independent clinical utility of SC and DC in the two large prospective cohorts. Therefore, although refinement for the modeling to fit the complex clinical conditions might be warranted, we suggest that the reservoir-wave paradigm can be complimentary to the conventional wave transmission theory to refine the description of vascular conditions.

4.3. Clinical utility of PWA-derived mechanical biomarkers

Tremendous efforts have been made to introduce hemodynamic parameters beyond SBP and DBP into clinical practice [35]. With the acquisition of central pressure waveforms alone, our study supported the clinical application of the PWA technique to derive biomarkers that contain clinical imperative information. Moreover, as shown in Table 3, peak of reservoir pressure, SC and DC were useful to reclassify or discriminate the subjects with intermediate risk, predicted by the conventional multivariate model, into correct risk categorizations.

4.4. Disagreement with previous study findings

Disagreement between our results and that of previous studies is noted but may be apparent rather than real. In the ASCOT-CAFÉ study [16], XSPI was identified as the only independent predictor for cardiovascular events with a median 3.4 years of follow-up. The study, however, did not report the analysis on SC and DC. In addition, the arterial reservoir model in the ASCOT-CAFÉ study used radial pressure waveform for analysis, of which the XSPI is different from that of central aortic or central pressure curves. In the ANBP2 study, SC was found to be independently predictive of clinical outcome [15], but in a protective manner (hazard ratio, 0.33, 95% confidence interval 0.13–0.82). Moreover, the lack of association between DC and cardiovascular outcomes was also noted in the ANBP2 study. Such a disagreement is intriguing and hard to explicate. Theoretically, SC is inversely related to the product of aortic characteristic impedance and total arterial compliance, and DC is inversely related to the product of systemic arterial resistance and total arterial compliance (Appendix II). With vascular aging, the arterial system stiffens and is associated with decreased total arterial compliance and increased aortic characteristic impedance. Similarly, systematic arterial resistance and total arterial compliance may also have opposite changes with increasing blood pressure. Consequently, it is difficult to separate individual contributions of each constituent parameter, and SC and DC may reflect and represent the integrated characteristics of the vascular system, the arterial reservoir function. With a worsening reservoir function, faster filling and emptying of the arterial reservoir may result in poorer accommodation ability of the vascular system to physiological stress. In this regard, it is a reasonable speculation that the worsening reservoir function should be associated with a higher SC and the resultant higher peak and amplitude of reservoir pressure (Figs. 2 and 4), as well as a higher DC and the resultant lower DBP and poorer perfusion pressure to end organs (Fig. 2). Further studies should be conducted to confirm the above speculations.

4.5. Potential limitations of the present study

We calibrated the carotid pressure waveforms by using MBP and DBP obtained from the seated brachial BP measured with a standardized protocol. We have previously demonstrated that the measured brachial SBP and PP and the derived central SBP and PP were predictive of all-cause and cardiovascular mortality and were independent of traditional cardiovascular risk factors, pulse wave velocity, left ventricular mass index, carotid intima–media thickness, and estimated glomerular filtration rate [17], suggesting the validity of the blood pressure measurements. In addition, we also conducted a sensitivity analysis to evaluate the impact of different calibrating methods on the prognostic significance of these PWA-derived biomarkers, in which comparable results are shown (see Table S1).

4.6. Conclusion

Our study demonstrated that, in the combined cohorts, AI, RM, SC and DC were independent predictors of cardiovascular mortality even accounting for the widely used cardiovascular risk factors. However, of all PWA-derived mechanical biomarkers, the reservoir function related SC and DC were consistently identified as useful parameters for cardiovascular risk prediction in two independent prospective cohorts. These findings suggested that mechanical biomarkers derived from pulse wave analysis could not only independently predict the long-term cardiovascular risks beyond the traditional risk factors, but also provide a more accurate risk stratification by incorporating these mechanical biomarkers into the risk prediction models. The utilization of pulse wave analysis could therefore be a reasonable procedure in the daily clinical practice in future.