Peripheral or central augmentation index: an esoteric question or a non-invasive clue to central haemodynamics?

Historical perspective and definitions related to the augmentation index In the early 1980, when studying aorta input impedance in humans, Murgo et al. [1] introduced the concept of A- and C-type central pressure waveforms. In the C-type waveform (usually seen in the young and healthy adult), the amplitude of the reflected wave is small, arrives in late systole and the addition of the reflected wave onto the forward wave does not increase systolic pressure. On the other end of the spectrum, in A-type waveforms (usually seen in elderly subjects), the amplitude of the reflected wave is much higher and the addition of the reflected wave onto the forward wave in early systole leads to a pronounced ‘pressure augmentation’ (AP). The associated augmentation index (AI = AP/PP) is the ratio of the ‘augmented pressure’ (augmentation of pressure after a characteristic point; AP) to the total pressure amplitude (pulse pressure; PP). AI is mostly expressed as a percentage and obviously is dimensionless. Physiological and mathematical principles underlying pressure augmentation are detailed in McDonald's Blood Flow in Arteries[2]. In studies with sufficient numbers of subjects and variation of baseline characteristics, it has been clearly demonstrated that AI increases with age and peripheral blood pressure and decreases with increasing height and heart rate [3,4]. The interested reader will find greater detail on these factors affecting AI in a review article by Davies and Struthers [5]. Nevertheless, it has been shown that AI reaches a plateau from the age of 55–60 years [6]. The magnitude and timing of the reflected wave and pulse wave velocity are also well-accepted determinants of AI. Physiologically speaking, AI appears to depend on the back-and-forth travel time of the pressure wave to the reflection sites relative to the timing of cardiac ejection. Because AI is a relative index, it is possible to calculate AI without the need of scaling or calibration, offering advantages and broad opportunities in clinical settings. Unfortunately, several facts delayed the use of AI in clinical practice: (i) non-invasive alternatives transferring waveform information from more accessible peripheral sites to the ascending aorta had to be developed because invasive data are not always available, except in catheterization facilities, surgery wards or intensive care units; (ii) opposing concepts, conflicting data and misinterpretations between several research groups on methodological and purely technical aspects confused many clinicians, postponing its clinical implementation; and (iii) the principle itself that AI is an ideal parameter ‘reflecting’ wave reflection has been criticized (see below). Nonetheless, today, several automated easy-to-handle, ready-to-use devices (Sphygmocor from Atcor Medical, VP-2000 from Omron Healthcare; Bannockburn, Illinois, USA) provide clinicians with nearly on-line readings of ‘central’ AI values. The number of cross-sectional clinical studies linking AI with cardiovascular risk patterns is dramatically increasing, as is the number of studies devoted to method comparisons and the physiological interpretation of AI. Importantly, however, prospective validation data of either invasive or non-invasive AI against hard cardiovascular endpoints are still scarce. All of this will be briefly discussed in the context of a study published in this issue of the journal. Sugawara et al. [7] describe a reasonable strong association (r = 0.86) between radial and carotid AI (surrogate for aortic AI?) obtained by an applanation tonometer (holder-fixed at both locations) and a commercial automated device (VP-2000). Data were obtained in apparently healthy subjects in a broad age range. Is it an ‘esoteric’ study question or does comparison of AI between peripheral and central sites add to the physiological understanding of the mechanism of central pressure augmentation? Methodological difficulties in the calculation of peripheral and central AIs Next to the general definition of AI = AP/PP, several working definitions of AI are available in literature. They are all mathematical permutations (some requiring scaling) of the ratio between augmented and pulse pressure; some of them are written as a fraction, others as a percentage, but all require the identification of a ‘characteristic landmark point’ indicative of the contribution of the reflected pressure waves. A popular formulation of the AI formula is: where P2 is the systolic pressure and P1 is the pressure at the characteristic point for A-type waves, yielding positive AI values. Vice-versa for C-type waves, yielding negative AI values and (P2 – P1) being the ‘augmented pressure’ and (SBP – DBP) the pulse pressure. An alternative expression leads to AI > 100% in A-type waves and AI < 100% in C-type waves. Successful introduction of AI as an easy-to-handle but reliable clinical tool will critically depend upon accurate and (automated) identification of these characteristic landmark points. Apart from elementary eyeballing (i.e. by visual inspection, as performed by Murgo et al. [1]), several strategies (some of them unpublished) have been worked out: defining an inflection point as the middle of two consecutive second derivative cross-overs and defining a shoulder pressure on the fourth derivative of the pressure signal [8]. Segers et al. [9] applied a fourth-order derivative approach and found an excellent correlation (> 0.98) between visually and automatically identified characteristic points. Notwithstanding mathematical difficulties in identifying the characteristic points, reproducibility data are nearly always within acceptable limits even in Bland–Altman plots when an automated device (Sphygmocor) was validated [10]. Perhaps the most physiology-based methodology to identify the moment in time where the reflected wave adds to the forward wave is the method of linear wave separation analysis (see below), defining a time span of constructive interference between forward and backward waves. As mentioned above, it is not evident to obtain appropriate central aortic pressure waveforms in cases where direct invasive access is not available. Therefore, with variable degrees of success, central aortic waveforms were ‘synthesized’ from recorded waveforms or distension curves at more easily accessible vascular territories: the radial and carotid arteries. A commonly used methodology to assess central AI starts from the radial artery waveforms obtained by an applanation tonometer, which are fed to dedicated software programs, synthesizing central waveforms making use of so-called ‘transfer functions’ (TF). TFs are complex functions and are usually expressed in the frequency domain by plotting the amplifications of the moduli and the phase angles of Fourier harmonic series between a peripheral and a more central waveform. Half a century ago, eminent physiologists (among them McDonald and Womersley) championed the underlying principles of TFs, but Michael O'Rourke was the first to use the concept in medical applications; for historical details, see [11]. Of interest, the concept of TFs existed already before Murgo's classification in A- and C-type waves became common knowledge. The transfer functions in use are generated from selected data validated in small populations (generalized transfer functions). Attempts to calculate TFs on an individual basis (individualized transfer functions) have been made, although without much success [12]. The generalized function of Karamanoglu et al. [13] has been implemented in the commercial Sphygmocor device. Although central wave morphology and ‘synthesized’ central systolic and pulse pressure mimic quite adequately the invasively recorded data in validation studies [11], direct head-to-head comparisons of AIs calculated on invasive versus ‘synthesized’ waveforms or ‘native’ radial artery AI are scarce. Hope et al. [14] made simultaneous recordings of invasive central aortic versus non-invasive radial waveforms and calculated central AI by a transfer function. Invasive and transferred non-invasive AI values differed significantly and, at best, a weak correlation was evident (see below for a critical discussion by O'Rourke's group). Of note, other waveform parameters, such as the systolic and diastolic pressure time integrals, were closely correlated [15]. The largest validation study was reported by Gallagher et al. [11] and retrospectively summarized the 10-year-old invasive unpublished validation data of the late Ray Kelly. Radial and carotid tonometry signals were transformed to ‘synthesized’ central aortic waveforms and compared with the invasive data in 61 subjects during catheterization and acute dosing of nitroglycerin. Differences were within acceptable limits. To allow maximal information transfer from a radial to a central ‘synthesized’ waveform and to reconstruct details of the wave morphology, it is a prerequisite that also the higher harmonics (up to ten and even more) are appropriately transformed [16]. Part of the good correlation for systolic and pulse pressure between peripheral and central readings might be the consequence of nearly exclusive low-frequency content of these parameters. Additionally, most transfer functions used in the literature show limited dissimilarity in their spectrum of the TF modulus: most of the power (energy) is concentrated in the first ten harmonics with maximum amplitudes ranging between 1.5 and 3 for harmonics 3–6 [11]. Alternatively, the group of Van Bortel et al. [17] and Segers et al. [18] used non-invasive applanation and distension curves of the carotid artery obtained by the echo-tracking principle as useful surrogates for central aortic curves. The authors obtained excellent results assessing local pressures by a dedicated calibration procedure [19,20]. Relationship between peripheral and central augmentation indices Sugawara et al. [7] are not the first to address the correlation between a peripheral and a more central (carotid or ‘synthesized’ aortic) AI. There is a striking discrepancy among published results of the correlations between peripheral and central AIs and for invasive versus non-invasive methods, ranging from unexpectedly poor associations to unexpectedly high correlations largely exceeding r-values of 0.90 [14,15,21]. The basic assumption is that there should be some kind of association between peripheral and central AI, and not a one-to-one perfect mapping. Convincing data pointing in the direction of a physiological link (unbiased by TF and devices) between radial artery and ascending aorta AI were obtained more than a decade ago by Takazawa et al. [8], who measured invasive aortic curves. After SL nitroglycerin administration, central AI fell proportionally (r = 0.74) with radial AI, strongly indicating a physiological link between both. Studies reporting blatantly weak associations (r < 0.50) between ‘synthesized’ and, respectively, invasive or peripheral AI most likely are underpowered, badly designed or biased by methodological shortcomings, such as inadequate frequency response of long (fluid-filled) catheters [22]. On the other hand, it is also likely that some methodological or physiological reason underlies the nearly perfect correlations. The question may even arise whether either a generalized or an individualized transfer function will offer additional value in calculating central AI in the case of excellent correlations between an (easily measurable) peripheral and central AI. The first incidental evidence of a better than expected correlation came from Rietzschel et al. [23] when comparing the Sphygmocor's radial and ‘synthesized’ AI in the framework of a method comparison study of commercial devices. Later, Millasseau et al. [21], also using the algorithms of the Sphygmocor, reported a value of r > 0.96 between the radial and ‘synthesized’ aortic AI. The authors claimed that the AI could be directly obtained from the radial pulse without resorting to the use of a TF. Because the radial waveform theoretically contains all information to synthesize the central waveform, a good correlation falls within the expectations. Arguments trying to explain the very high correlations comprised the inclusion of repeated individual data before and after intervention strengthening any existing correlation and selection of easy to handle curves from younger subjects versus curves from older subjects with a single late systolic shoulder [24]. In general, ‘synthesized’ central aortic curves may be better suited for the calculation of wave reflection parameters because the falling systolic limb of the radial artery curve is much more difficult to handle. Hope et al. [25], using another transfer function (Fetics’), largely confirmed the excellent agreement (r = 0.89) between central and peripheral AI. Correlation dramatically decreased to r = 0.65 when the TF-derived AI was correlated with directly measured aortic AI. More generally, Hope et al. [25] concluded that, when only non-invasive pressures are accessible, analysis of untransformed radial waveforms produced smaller errors in the estimation of central pressure and waveform parameters than using a generalized TF. Segers et al. [9] recently presented data for carotid tonometry AI (after calibration and applying Chen's TF). Although radial and ‘synthesized’ aortic AI correlated surprisingly well (r > 0.90), correlation with carotid AI was moderate (r = 0.72). A word of caution might be needed: theoretically, an additional carotid-aortic transfer could influence and improve the observed moderate correlation. However, during an earlier study [26], they observed a correlation of 0.64 between invasive and ‘synthesized’ aortic AI. Other confirmatory evidence for a good radial (Omron HEM-9010AI device) to aortic (Sphygmocor) correlation of AI (r > 0.80) was provided by Kohara et al. [27]. How should we interpret and explain these remarkable findings? The issue has been subject to a myriad of correspondence in the literature, each replying with vigour to the opponent's opinions and reinforcing the strength of their own arguments. What can we learn from the still ongoing debate? Given the good correlation between peripheral and ‘synthesized’ central AI and the potential loss of information during transformation, are there arguments to prefer ‘synthesized’ aortic curves instead of ‘native’ peripheral curves? A ‘synthesized’ central aortic waveform does not contain more information than its radial untransformed input counterpart, but the radial artery and brachial artery are ready-accessible territories, respectively, for applanation tonometry and for calibration and transformed curves may be technically better suited to submit to automated proprietary algorithms available in high through-put commercial devices. Unfortunately, published evidence sustaining why transformed curves prove superior over untransformed curves to calculate central AI remains scarce. Augmentation index: limitations and future directions No doubt, there is now some documented evidence that more centrally recorded AI is a good indicator of increased cardiovascular risk. Nürnberger et al. [28] demonstrated that increased carotid AI was associated (in the expected direction) with all common major risk factors and indicators (age, gender, smoking, cholesterol, height, body mass index, heart rate), as well as with three different risk scores (ESC, SMART and EPOZ). Weber et al. [29] enrolled 465 symptomatic men and, after controlling for all relevant confounders, they observed an odds ratio > 6 (Q4 versus Q1 of ‘synthesized’ central AI) for premature coronary artery disease [29]. An initial but important limitation preventing the more widespread clinical use of AI or central pressure is that validation against hard endpoints is still limited. The issue was convincingly demonstrated in very high-risk patients, such as end-stage renal disease/haemodialysis patients [30]. Recently, the interest of the broader cardiovascular community for non-invasive central aortic waveforms was boosted by the publication of the CAFE study, a sub-study of ASCOT on 2199 participants [31]. In baseline risk factor adjusted Cox proportional hazards models, central pulse pressure and central AI [although only borderline for AI 1.18; 95% confidence interval (CI) = 0.98–1.43] predicted the composite endpoints independent of peripheral pulse pressure. The CAFE study clearly suggests a clinical role for non-invasive central waveforms and their derived indices. However, not all of the outcome studies favoured central over peripheral waveform indices. For example, Dart et al. [32] studied the prognostic value of non-invasive central pressure and carotid AI (from direct applanation tonometry) in elderly hypertensive women from the ANBP2 trial. After adjustments for classical risk factors, including brachial blood pressure, neither central systolic blood pressure nor carotid AI predicted cardiovascular events: with hazard ratios of 1.26 (95% CI = 0.72–2.23) and 0.82 (95% CI = 0.45–1.49), respectively. This may be due to the fact that, in older subjects, central AI tends to a plateau-value of approximately 50% at age 50–60 years [6]. Because the backward pressure wave cannot become more important than the forward pressure wave, AI per definition is limited to 50%. As such, AI may lose its sensitivity and specificity in older populations. A few cautionary notes are required when relying on carotid artery waveforms and when vascular territory of calibration differs from site of measurement. Carotid waveforms do not perfectly match central aortic waveforms and brachial to radial augmentation should not be neglected. Thus, direct extrapolation of carotid artery to central aortic AI and direct substitution of radial for brachial blood pressures may have some limitations. From a theoretical physiological point of view, such extrapolations cannot be recommended but, in clinical practice, carotid artery waveform parameters sometimes may be more-easy-to-access parameters and provide acceptable substitutes for their central aortic counterparts. For example, at first glance, carotid tonometry-derived and aorta waveform-derived AI values correlate reasonably well (r = 0.78) but carotid AI values are lower [33] and could not be called true equivalents. Calibrations executed on different vascular territories or with different principles may induce devastating consequences for the accuracy of estimated central pressures. Brachial artery is standard for calibration. Diastolic and mean blood pressures are considered as almost constant (physiology handbooks do document small deviations) all over the large arteries. More central systolic blood pressure (carotid, ascending aortic) readings may follow application of this principle in as much that the rule of thumb of mean arterial pressure (MAP) (diastolic blood pressure plus one-third of the pulse pressure) is replaced by numeric integration algorithms on the curve itself to AI, being a should not much from alternative or pressure However, in most aortic AI, central pressures or other parameters requiring calibration are A but important limitation of AI is its physiological link with wave reflection definitions of AI (see only require of pressure waves. AI on the magnitude and timing of the reflected wave, but the magnitude itself cannot be from a single AI analysis of wave starts from a of recorded pressure and waveforms that the impedance has been demonstrated that the wave reflection the concept of AI and may be by the pressure wave its forward and backward = and = – with being an index better pressure wave In these is aortic characteristic is the aortic and is the aortic pressure To the it should be that Murgo's [1] A waves was a of a study both pressure and signals to study input The of pressure and signals limited to aortic may allow for a more detailed and physiology-based of wave reflection and its role in pressure studies will whether transfer functions and indices pressure augmentation will add to a better risk or at their best, they are only poor ‘esoteric’ surrogates for the central consequences of wave and will in be replaced by the between pressure and