Aortic Waveform Analysis to Individualize Treatment in Heart Failure

Subjects aged >18 years with chronic HFrEF, NYHA class ≥II, on stable doses of angiotensin-converting enzyme inhibitor/angiotensin II receptor blocker and β-blockers for at least 3 months were enrolled. Subjects with LV ejection fraction <25% or ≥50%, brachial systolic blood pressure <110 mm Hg at most recent clinical assessment, AIx<15%, inability to exercise, irregular heart rhythm, pregnancy, myocardial infarction within 30 days, cardiac surgery within 60 days, significant valvular heart disease (>mild regurgitation or any stenosis), myocarditis, thyroid disease, severe renal disease (creatinine >2.0 mg/dL), or significant competing cause of exercise intolerance (eg, obstructive pulmonary disease and peripheral arterial disease) were excluded.

The study protocol was previously described in detail.⁵ In this trial, subjects were randomized to AIx-guided therapy versus sham (central blood pressure and AIx data acquired, but not shared with investigator). Because angiotensin-converting enzyme inhibitor/angiotensin II receptor blockers and specific β-blockers (metoprolol, carvedilol, and bisoprolol) were standard guideline-directed medical therapy in all patients with HFrEF at the time of trial design, medication titrations during the trial were required to be made after these therapies were increased to the maximal tolerated goal doses according to guidelines.⁴ The active treatment goal was to reduce central aortic wave reflections, assessed by AIx, to 0% using vasoactive drugs. Additional vasodilator medicines were added as tolerated to maximal guideline-directed medical therapy in the following order: spironolactone, nitrates, hydralazine, and amlodipine. Additional vasodilator therapy was added/dose escalated as long as central blood pressure was maintained in an acceptable range (systolic BP >80 mm Hg with no symptoms of hypotension).

Brachial blood pressure, radial pulse wave recording, and 6-minute walk test were performed at baseline and monthly for a total of 6 months. HF medication treatments were adjusted by investigators based on brachial (sham) or aortic pressure data (active treatment) at each of the 7 study visits to optimize brachial BP (in sham controls) or reduce AIx (in active treatment).

To detect changes in the aortic waveform associated with functional improvement on vasoactive HF therapy, subjects were separated by the improvement in 6-minute walk distance during the study. To decrease variability, to provide a longer-term assessment of functional capacity, and to achieve greater precision, the mean value of all 6-minute walk test distances after randomization (on treatment) was compared with the distance before randomization at the baseline visit. Improvement in submaximal exercise capacity was defined by a positive difference between mean postbaseline and baseline walking distance, whereas negative differences marked decrease in walking distance. Subjects with improvement in 6-minute walk distance on treatment were then compared with those with no improvement or deterioration in 6-minute walk distance.

To compare vasoactive medication usage at baseline and on treatment during the trial, ACE inhibitor doses were converted to lisinopril units, angiotensin II receptor blocker to losartan units, and β-blockers to carvedilol units.

Subjects were instructed to refrain from coffee and smoking in the morning of each study visit. After a 5-minute rest in the seated position, brachial blood pressure was measured using an oscillometric device. Arterial waveforms were then measured using a commercially available noninvasive, high-fidelity, hand-held applanation tonometer (SphygmoCor, AtCor Medical Ltd, Sydney, Australia). Built-in, custom software (SphygmoCor Cardiovascular Management Suite) applying a validated general transfer function¹¹ was used to convert the radial pressure waveform to the central aortic waveform. AIx, augmentation pressure, reflection time, amplitude of the first (P1) and the second (P2) systolic peaks were then calculated using pulse wave analysis by the built-in SphygmoCor software (Figure 1A). Previous studies have shown good to excellent reproducibility of central arterial measures derived from the radial pressure wave using the applanation tonometry.^12–14

Wave separation analysis is a frequency domain method based on the propagation model, which considers both aortic pressure and flow waves. In this analysis, the aortic waveform is decomposed into forward (Pf) and backward (Pb) travelling pressure waves (Figure 1B). The forward pressure wave is explained by ejection of blood from the LV to the aorta, whereas the reflected pressure wave originates from reflection sites of impedance mismatch along the peripheral circulation where part of the incident pressure wave is reflected backward.

Wave separation analysis has been described in detail elsewhere.¹⁵ Briefly, central aortic pressure curves were exported from the SphygmoCor software, and corresponding flow curves were estimated from measured pressure curves based on a 3-element Windkessel model where LV outflow is modeled as a dynamic system of the second order. Windkessel equations were formulated as an isoperimetric problem with a constraint to minimize hydraulic work. Pressure waveform area fitting and a second-order linear delay element were used to estimate the final flow shape. Wave separation was performed in the frequency domain from measured pressure (Pm), modeled aortic flow (Qm), and aortic characteristic impedance (Zc). Zc was calculated using the modulus of the complex input impedance, which was calculated as the ratio of pressure and flow. The forward pressure wave was calculated using the equation Pf=0.5×(Pm+Zc×Qm), whereas the reflected pressure wave as Pb=0.5×(Pm−Zc×Qm). Analyses were processed in Matlab (Mathworks, Natrick, MA) using algorithms from a commercially available software package validated in non-HF patients^15,16 and in patients with HFrEF.¹⁷

Arterial reservoir analysis was examined as another model to independently characterize the central aortic waveform. This method combines the 2-element Windkessel and wave propagation models. In this model, the aortic waveform is decomposed into the reservoir and excess pressure (Figure 1C). Reservoir pressure accounts for the overall compliance of the arterial system, whereas the excess pressure is determined by local conditions and is the difference between measured pressure wave and reservoir pressure. Custom written Matlab algorithms were applied to calculate reservoir and excess pressures as previously reported.¹⁸

The test was performed as recommended¹⁹ indoors, along a long, flat, straight, enclosed corridor with a hard surface. The walking course was 30 m in length. The length of the corridor was marked every 3 m, and the turnaround points were marked with cones. Subjects were instructed to walk as far as possible for 6 minutes. The total distance walked was calculated, rounding to the nearest meter.

Continuous data with normal distribution are presented as mean±SD, non-normally distributed variables as median (quartile 1 to quartile 3). Categorical data are shown as frequencies and percentages. Differences between groups with and without functional improvement at baseline were compared using t test, Mann–Whitney U test, χ² test, or Fisher exact test, as appropriate. To account for correlation on the same patient, generalized mixed models with a random intercept were used to compare differences in hemodynamic parameter changes derived by pulse wave, wave separation and arterial reservoir analyses between patients with and without functional improvement. In this model, group, time and interaction between group and time were analyzed as fixed effects. Gamma regression was used for right skewed data. The model-derived estimated marginal means with 95% confidence interval are reported. Calculations were done using SPSS version 21 (IBM SPSS Statistics, IBM Corporation, Armonk, New York, NY). A 2-sided P value of <0.05 was considered statistically significant.

At baseline, subjects with functional improvement had lower heart rate but no difference in brachial cuff-derived blood pressure measurements compared with those with no improvement (Table 1). In contrast to peripheral pressures, aortic waveform characteristics differed significantly between groups, with greater pressure wave pulsatility in responders compared with nonresponders, assessed by higher central PP, augmented pressure, greater forward and reflected pressure wave amplitudes, and higher reservoir pressure (all P<0.05, Table 2). Each of these differences remained significant after adjusting for heart rate (not shown).

Central PP was strongly correlated with the reflected pressure wave (r=0.976, P<0.001) and reservoir pressure (r=0.978, P<0.001), whereas there was also a strong association between the reflected pressure wave and the reservoir pressure (r=0.985, P<0.001). In the receive operating characteristic (ROC) analysis to predict functional improvement on therapy, baseline central aortic PP >36 mm Hg had 71% sensitivity and 62% specificity (ROC area under the curve=0.763, P<0.001), reflected pressure wave (Pb) >15 mm Hg had 75% and 69% specificity (ROC area under the curve=0.779, P<0.001), and reservoir pressure >39.0 mm Hg had 88% and 62% specificity (ROC area under the curve=0.760, P<0.001). There was no difference in ROC between these parameters (all P>0.05).

There was no difference in traditional brachial cuff pressure measures in subjects with and without functional improvement (Table 3). However, decreases in central aortic pressure wave pulsatility on treatment (central PP derived by pulse wave analysis, Pf and Pb calculated by wave separation analysis, and reservoir pressure by arterial reservoir analysis) were greater in subjects with functional improvement when compared with subjects without functional improvement (interaction P values all <0.05, Table 3; Figure 2). In addition, the reduction in Pb and reservoir pressure on treatment was directly correlated with greater increases in 6-minute walk distance during the trial (Figure 3). Although there was no difference in reflected wave timing (P for interaction=0.9), there was a greater reduction in pulse wave velocity in subjects with functional improvement (P for interaction=0.04), indicating greater reduction in aortic stiffness on treatment in responders, again despite similar brachial BP.

Medication changes (additions or titrations) were performed during the trial in all 38 participants. New medicines were added in 27 subjects. The average number of medication adjustments was 3.2. Despite the significant differences in aortic pressure waveforms in responders and nonresponders, there was no difference in vasodilator therapy at baseline or vasodilator therapies added (medicine type or dosage) during the trial between groups (Table 4). This suggests that it is the individual participant’s hemodynamic response to vasodilator therapy that is more important in explaining the relation to exercise improvement, rather than the absolute dose of vasodilator medicine provided. It also indicates that individual patients respond differently to vasoactive medicines in terms of central aortic waveforms, and that these differences are not evident from analysis of conventional cuff arterial pressures. These points are illustrated by examining the individual subject response to addition of hydralazine during the trial (Figure 4). Subjects with improved exercise capacity had higher aortic pulsatility at baseline and greater reduction on treatment when compared with nonresponders. These differences were not apparent from examination of baseline or changes in brachial BP.

The sample size of this trial is small and the split between responders and nonresponders is uneven. Therefore, precision of estimates and power to detect group differences are reduced, and there may have been other (small to moderate magnitude) differences between groups that were not detected. However, assessing variables at several time points allowed us to decrease variability and to achieve greater precision of point estimates. The aortic pressure waveforms used were transformed from radial recording using the generalized transfer function. This reconstructed aortic pressure curve may lack characteristics of directly measured aortic tracings, but this method has been used extensively in previous studies and clinical trials. Aortic flow was not directly measured but derived from pressure waveforms as previously described. While this method requires a greater number of assumptions, it shows similar ability to characterize arterial load.¹⁷ Furthermore, the consistency of results across the 3 models strongly supports the veracity of our conclusions. We used the 6-minute walk test that assess the submaximal exercise performance, rather than peak oxygen consumption because this test was performed more frequently during the trial and because it provides a better evaluation of capacity in the activities of daily life than true maximal aerobic capacity which is rarely achieved in HFrEF. Peak exercise capacity tended to improve more in responders, but this difference was not statistically significant, possibly related to greater variability in the assessment of peak oxygen consumption. There might have been a “ceiling effect” that affected the results, whereby further improvement in 6-minute walk distance might not have been possible, and we cannot discount this possibility. Because the assignment to groups was not random, we cannot exclude the effect of residual confounding. Because of large variability in the individual response to vasoactive therapy, our data do not allow to identify a single specific drug responsible for functional capacity improvement. Patients with severe LV dysfunction and hypotension were excluded from this trial; therefore, these results may not apply to patients with more advanced HFrEF, particularly patients with lower arterial pressures. There was a trend toward higher dose of β-blockers in responders at study entry, and although the differences in pressure pulsatility between responders and nonresponders remained significant after adjusting for heart rate, we cannot discount the possibility that greater tolerance of β-blocker dosage at study entry might have also helped identify a group more likely to respond to more aggressive vasodilator therapy. As the study population consisted predominantly from older men, future studies evaluating the effect of vasoactive therapy in younger patients and women with HFrEF are needed. While radial tonometry has good reproducibility,^12–14 it is not well described how aortic indices change during the course of the day in patients with HF, which would be important to understand to properly apply this technique in practice. This trial was not powered to examine harder end points, such as hospitalization burden or survival, and future study is needed in this regard.

Elevated aortic pulsatile load and greater decrease in pulsatile load using vasodilator therapy is associated with improvement in submaximal exercise capacity in patients with chronic HFrEF, even when guideline directed medical therapy has been maximized. Patients with increased pulsatile load that may benefit from aggressive therapy are not identifiable from conventional brachial blood pressure assessments, and changes in aortic pulsatile load on treatment are similarly not evident from cuff pressure assessment. These data suggest that there may be a role for individualized, precision medicine in the tailoring of vasoactive therapies in HFrEF, and prospective trials testing this concept are warranted.