Arterial Wave Reflections and Incident Cardiovascular Events and Heart Failure

MESA (Multiethnic Study of Atherosclerosis)

Study population

MESA enrolled 6,814 men and women ages 45 to 84 years who identified themselves as white, African American, Hispanic, or Chinese and who were free of clinically apparent cardiovascular disease, from 6 U.S. communities between 2000 and 2002 (16). The study was approved by the institutional review boards of participating centers and participants gave informed consent.

Data collection

Standardized questionnaires were used to obtain information about cardiovascular risk factors and medication use. Resting blood pressure was measured in triplicate using a Dinamap-Pro100 oscillometric sphygmomanometer (GE Medical Systems, Waukesha, Wisconsin). The average of the last 2 measurements was used. Hypertension was defined as systolic blood pressure ≥140 mm Hg, diastolic blood pressure ≥90 mm Hg, and/or antihypertensive medication use (17). Serum total cholesterol, high-density lipoprotein (HDL) cholesterol, and glucose were measured after a 12-h fast. Diabetes mellitus was defined as fasting glucose ≥126 mg/dl and/or hypoglycemic medication use (18).

Hemodynamic measurements

Radial arterial waveform 30-s recordings were obtained at baseline in the supine position using the HDI/PulseWave-CR2000 tonometry device (Hypertension Diagnostics, Eagan, Minnesota), digitized at 200 Hz and exported for offline processing using custom-designed software written in Matlab (The Mathworks, Natick, Massachusetts).

A generalized transfer function (19) was applied to the radial pressure waveform to obtain a central pressure waveform. Aortic-radial pulse pressure amplification (PPA) was computed as: (radial pulse pressure/aortic pulse pressure) × 100 (Fig. 1). This computation does not depend on calibration of the radial pressure waveform.

The first and second systolic peaks were identified in the aortic pressure waveform as previously described (20). AIx was computed as (Second/First systolic peak) × 100 (Fig. 1). A physiologic flow waveform (11) was used for separation of the pressure wave into forward and reflected waves (21). RM was calculated as shown in Equation 1 (Fig. 1).backward wave amplitudeforward wave amplitude×100Further details about this analysis methods can be found in the Online Appendix.

Event adjudication

In addition to undergoing 3 on-site examinations, participants were contacted by a telephone interviewer every 9 to 12 months to inquire about incident CVEs. Two physicians independently reviewed copies of medical records and death certificates for hospitalizations and outpatient cardiovascular diagnoses, for blinded endpoint classification using pre-specified criteria (22), as summarized in the Online Appendix. The following endpoints were defined a priori for this study: 1) hard CVEs, defined as a composite prevalence of myocardial infarction, resuscitated cardiac arrest, cardiovascular disease–related death, stroke, or stroke-related death; 2) all CVEs, defined as a composite prevalence of any hard CVE, angina, other atherosclerosis-related death, or other cardiovascular disease–related death; and 3) CHF, which required clinical symptoms (e.g., dyspnea) or signs (e.g., edema), a physician's diagnosis of CHF, and medical treatment for CHF, in addition to: a) pulmonary edema/congestion by chest radiography; and/or b) dilated ventricle or poor LV function by echocardiography or ventriculography, or evidence of LV diastolic dysfunction. Whereas the 2 former are composite endpoints commonly used in cardiovascular risk studies, CHF was defined as a specific endpoint a priori based on previous experimental data.

Statistical analysis

A more detailed description of the statistical methods can be found in the Online Appendix. The association between hemodynamic measures and time to CVE or CHF was examined using the Kaplan-Meier method and Cox regression. Model goodness-of-fit was assessed with the Akaike's information criterion and Bayesian information criterion (23,24). Model discrimination was assessed with the Harrel's c-index (which is analogous to the area under the receiver operator characteristic curve) (23,25). Calibration was assessed with the Hosmer-Lemeshow test. Improvements in subject reclassification was further assessed using the net reclassification improvement (NRI) (23,25), which depends on the increased probability that a new model will categorize case subjects as higher risk and on the decreased probability that it will categorize control subjects as lower-risk, compared to a base model, as explained in more detail in the Online Appendix. We computed 2 versions of the NRI: a category-based NRI based on usual categories for 10-year cardiovascular disease risk (adapted at 5 years as <2.5%, 2.5 to <5%, 5 to <10%, and ≥10%). Because no established categories exist that guide clinical decisions for CHF risk, we computed category-free reclassification measures, which are independent on arbitrarily defined risk thresholds (26). These include the category-free NRI and the relative integrated discrimination improvement (rIDI), which expresses the relative improvement in discrimination slopes (difference in mean predicted probabilities between case and control participants) between the base model and new model (23,25–27). Various indices of model performance were used to: 1) assess the added predictive value of central pressure indices; and 2) compare the predictive value of aortic hemodynamic indices to that of well-established risk factors, as previously described (23).

All tests were 2-sided with alpha = 0.05. Analyses were performed using SPSS version 19.0 (SPSS Inc., Chicago, Illinois).

Cardiovascular events

During a median follow-up of 7.61 (interquartile range: 7.34 to 7.78) years, 407 subjects experienced a first CVE, 281 subjects experienced a first hard CVE, and 117 experienced a first episode of CHF.Hazard ratios (HRs) for incident CVE and CHF associated with a 10-point increase in AIx, RM, or PPA are shown in Table 2.Table 2 also shows standardized HRs (those associated with a 1-SD increase in the predictors). In adjusted analyses, AIx was independently associated with hard CVEs (model 3, Table 2) (HR per 10% increase: 1.08; 95% confidence interval [CI]: 1.01 to 1.14; p = 0.016). The category-free NRI, rIDI and increase in c-index achieved by adding PPA to the other variables in this model were 0.036, 0.002, and 0.004, respectively. PPA was independently associated with all CVEs (model 3, Table 2) (HR per 10% increase: 0.82; 95% CI: 0.70 to 0.96; p = 0.012). The category-free NRI, rIDI, and increase in c-index achieved by adding PPA to the other variables in this model were 0.10, 0.02, and 0.002, respectively. PPA and AIx were not independently associated with incident CHF.

RM was independently associated with incident CVEs. After adjustment for age, sex, ethnicity, systolic blood pressure, diastolic blood pressure, antihypertensive medication use, height, weight, diabetes mellitus, total cholesterol, HDL cholesterol, current smoking, heart rate, and glomerular filtration rate, a 10% increase in RM was predictive of a higher risk for CVEs (HR: 1.34; 95% CI: 1.08 to 1.67; p = 0.009) and hard CVEs (HR per 10% increase: 1.46; 95% CI: 1.12 to 1.90; p = 0.006). The category-free NRI, rIDI, and increase in c-index for the prediction of all CVEs achieved by adding RM to the other variables shown in model 3 (Table 2) were 0.15, 0.05, and 0.002, respectively. The category-free NRI, rIDI, and increase in c-index for the prediction of hard CVEs when RM was added to the other variables shown in model 3 (Table 2) were 0.13, 0.08, and 0.002, respectively.

RM was strongly predictive of incident CHF (HR per 10% increase: 2.69; 95% CI: 1.79 to 4.04; p < 0.0001; Table 2). Table 3 shows independent predictors of incident CHF in a fully adjusted model, together with standardized HRs for each term, in order to allow for easier comparison between various predictors. The full model is similar to the one used in Table 2, with the exceptions of height and weight, which were replaced by body mass index to more intuitively assess the independent contribution of obesity to CHF risk prediction in the model. Table 3 also shows improvements in model performance observed when individual predictors were added to a model containing all other variables except the predictor in question. RM was associated with a standardized HR of 1.61 (95% CI: 1.32 to 1.96), the largest Wald statistic and the greatest decreases in Akaike's information criterion and Bayesian information criterion (indicating improvement in model fit) and the greatest increases in IDI and rIDI. With the exception of age, a nonmodifiable risk factor, RM was associated with the greatest NRI. Of note, these improvements in model performance were also superior to changes induced by adding systolic blood pressure and diastolic blood pressure together.

The addition of RM to a model containing all other variables shown in Table 3 resulted in a category-free NRI of 0.38 and a rIDI of 0.48, indicating a 48% relative increase in the discrimination slope achieved by all variables in the base model. The addition of RM to the model resulted in a category-based NRI of 0.17. A category-dependent reclassification table can be found in Online Table 1. The full model containing RM demonstrated adequate calibration (Hosmer-Lemeshow chi square: 8.50; p = 0.49).

Finally, in order to more directly compare the value of brachial pulse pressure versus RM as predictors of CHF, RM or brachial pulse was separately added to a base model that included age, sex, body mass index, diabetes mellitus, ethnicity, antihypertensive medication use, total cholesterol, HDL cholesterol, current smoking, heart rate, and estimated glomerular filtration rate. When RM was added to this base model, the standardized HR for RM was 1.54 (95% CI: 1.26 to 1.88; p < 0.0001), with an achieved NRI, rIDI, and increase in c-index of 0.32, 0.42, and 0.011, respectively. When pulse pressure was added to this base model, the standardized HR for RM was 1.42 (95% CI: 1.18 to 1.70; p < 0.0001), with an achieved NRI, rIDI and increase in c-index of 0.27, 0.20, and 0.011, respectively.

Figure 2A shows cumulative hazard curves for incident CHF among subjects, stratified by RM tertiles. There was a progressive increase in CHF risk with increasing RM tertile (p < 0.0001). For further comparison of the risk for CHF associated with hypertension versus increased RM, the population was stratified according to the presence (prevalence: 45%) or absence of hypertension. The population was also stratified into those above or below the 55th RM percentile to obtain a prevalence of ”high” RM identical to the prevalence of hypertension (45%), allowing for a more direct comparison of the contribution of each factor to CHF risk in the population. Figures 2B and 2C show adjusted cumulative hazard functions for each substratum according to the presence or absence of hypertension and high RM. As shown, nonhypertensive subjects with low RM were at lowest risk, hypertensive subjects with high RM were at highest risk, and hypertensive subjects with low RM and nonhypertensive subjects with high RM were at intermediate risk for incident CHF.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Reflection Magnitude as a Predictor of Incident Heart Failure

(A) Cumulative hazard for heart failure among subjects stratified according to tertiles of reflection magnitude (RM). (B) Hazard function for incident of heart failure in subjects stratified according to the presence or absence of hypertension (prevalence: 45%) or reflection magnitude above (“high” RM) or below (“low” RM) the 55th percentile (prevalence of “high” RM: 45%), adjusted for other significant heart failure predictors shown in Table 3. (C) Analogous hazard functions after further adjustment for ethnicity, antihypertensive medication use, total-cholesterol, HDL-cholesterol, current smoking, glomerular filtration rate, and heart rate.