High Coronary Shear Stress in Patients With Coronary Artery Disease Predicts Myocardial InfarctionFree Access

Study group and study design

The design of the FAME II (Fractional Flow Reserve Versus Angiography for Multivessel Evaluation II) trial (15) and its 3-year results (3) have been previously published. Briefly, in the FAME II trial, 1,220 patients from 28 sites in Europe and North America with stable angina and angiographically documented CAD involving up to 3 vessels were randomized and assigned, when at least 1 vessel had FFR ≤0.80, to receive medical therapy only (n = 441) or to undergo FFR-guided percutaneous coronary intervention in addition to medical therapy (n = 447). Patients with FFR >0.80 across all lesions were not randomized, and 50% of these patients were followed in a registry. For this post hoc analysis, only the medical therapy group was used (n = 441).

Outcomes

The primary outcome of this study was vessel-related myocardial infarction (VR-MI). Follow-up was censored at 3 years. All events and culprit vessels were adjudicated by an independent clinical event committee, blinded to FFR values, which went through the detailed narrative of each event and lesion assigned as VR-MI. In patients with >1 designated culprit vessel, the vessel with lower FFR was studied.

Angiographic reconstruction of target vessels

All baseline angiographic reconstruction, computational fluid dynamics (CFD), and WSS computations were done at the Emory University Cardiovascular Imaging and Biomechanical core laboratory in Atlanta, Georgia by independent analysts who were blinded to baseline FFR values, clinical data, and patients’ outcomes. QAngio XA 3D RE (Medis Medical Imaging Systems, Leiden, the Netherlands) was used to create 3-dimensional (3D) geometric reconstructions of each patient’s target vessel by using end-diastolic angiographic projections at least 25° apart. All visible branching vessels were added as cylindrical extensions perpendicular to the vessel centerline with the branch location, diameter, and orientation determined from the angiograms (Online Appendix). Validity and interobserver and intraobserver variability of 3D quantitative coronary angiography by QAngio XA 3D RE has been previously reported (16–18). The resulting 3D vessel point cloud was wrapped to form a triangulated surface (Geomagic Studio 12, Geomagic, Research Triangle Park, North Carolina). Extensions were added to each inlet (2 diameters) and outlet (8 diameters) to ensure a smooth transition of flow at the boundaries. The geometry was then meshed using ICEM CFD (Ansys ICEM, Ansys 17, Ansys, Canonsburg, Pennsylvania).

Boundary conditions and computational fluid dynamics

Patient-specific velocities were calculated from baseline angiograms by using a novel 3D contrast velocity method (Online Figures 1 and 2, Online Appendix). The calculated velocity was applied as an inlet boundary condition with a flat profile. The distribution of flow to each outlet boundary was derived from Murray’s law (19). Blood was assumed to be an incompressible Newtonian fluid with a density of 1,050 kg/m³ and a viscosity of 0.0035 kg/m-s, and the no-slip boundary condition (null velocity) was applied at the vessel wall. Steady flow simulations were performed using the commercial CFD solver Ansys Fluent (Ansys 17) (Online Appendix).

Hemodynamic analysis

For each target vessel, an experienced angiographer identified lesion start points and endpoints by using the QAngio XA 3D RE program, thus allowing automatic quantification of 3D lesion length, 3D percentage diameter stenosis (DS%), and minimum lumen diameter (MLD) location. The vessel WSS was determined from the flow field solution output by the CFD simulation and was imported into MATLAB (MATLAB R2013b, MathWorks, Natick, Massachusetts). Five segments of interest, along with the lesion itself, were identified for further analysis. These were the proximal, middle, and distal thirds of the lesion and 5-mm segments proximal and distal to the lesion (Central Illustration). In each segment, the mean WSS was calculated, and these values were used to determine the association between WSS and patients’ outcome.

Statistical analysis

Normally distributed continuous variables were summarized using data expressed as mean ± SD, and non-normally distributed continuous variables were summarized as median (interquartile range). Categorical data are presented as frequency counts and percentages. Comparisons between groups were performed using 2-sample t-test, Wilcoxon rank sum test, chi-square test, and Fisher exact test, as appropriate. Correlations between continuous variables were determined using Spearman’s rank correlation coefficient.

To account for confounding factors, a 1:1 propensity score matching approach was adopted to balance any baseline demographic or clinical differences between patients who had VR-MI within 3 years and a control group consisting of patients who did not reach any of the pre-defined clinical outcomes of FAME II (all-cause death, unplanned hospitalization leading to urgent revascularization, and MI) in 3 years. The logistic regression model for propensity score calculation included relevant demographic (age, sex) and clinical (body mass index, smoking status, hypertension, diabetes, hypercholesterolemia, peripheral vascular disease, family history of CAD, history MI, history of percutaneous coronary interventions in target vessels, history of stroke/transient ischemic attack, left ventricular ejection fraction by echocardiogram <50%, and baseline angina class) variables. Each patient in the MI group was matched with a patient from the non-MI control group with the same vessel type (left anterior descending artery, left circumflex artery, or right coronary artery) by the smallest difference in the propensity score. To determine the variables associated with VR-MI, the marginal Cox model approach was used to adjust for the correlated failure times with matched pairs. Specifically, the regression parameters in the Cox model were estimated by the maximum partial likelihood estimates under an independent working assumption (20). A robust sandwich covariance matrix estimate was used to account for the within-pair dependence (21). To evaluate the discrimination of WSS measured in each segment of the lesion, Harrell’s C-index was calculated from univariate marginal Cox models.

The likelihood ratio test was performed to examine the significance of addition of WSS_prox to a model with FFR. Incremental prognostic value of WSS was defined as the presence of both a statistically significant increase in global chi-square value and an increase in continuous net reclassification improvement and integrated discrimination improvement. Because there were 29 outcome observations, 5 separate marginal Cox models were constructed to assess the value of WSS_prox in the prediction of VR-MI after adjusting for distance of lesion MLD from vessel ostium, distance of proximal segment of lesion from MLD, DS%, lesion length, and lesion FFR. The optimal cutoff value for detecting MI was chosen as the value maximizing sensitivity and specificity on a receiver operating characteristic curve. We compared the risk of VR-MI between the 2 WSS groups (>4.71 Pa vs. ≤4.71 Pa) by using stratified log-rank test and constructing nonparametric (Kaplan-Meier) survival curves that accounted for vessel type-specific sampling fractions (22). A 2-sided p value <0.05 was considered statistically significant. All statistical analyses were performed with SPSS 24.0 software (SPSS, Chicago, Illinois), and R software version 3.4.3 (R Foundation for Statistical Computing, Vienna, Austria).

Patients’ characteristics

Among 441 patients in FAME II with FFR ≤0.80 who were randomized to medical therapy alone, 34 (8%) had subsequent MI by 3-year follow-up. One patient was found to have MI related to supply-demand mismatch with no specific designated culprit vessel and was therefore excluded from the final analysis. Four patients with target VR-MI had insufficient angiographic projections of the target vessel, thus rendering adequate 3D reconstruction impossible, and they were also excluded from the study.

Among the remaining 29 culprit vessels from 29 patients with VR-MI and 29 vessels from 29 matched non-MI control patients from the cohort, median age was 62 (interquartile range [IQR]: 57 to 69) years, and 46 (79%) were male. Median lesion length was 19.4 (IQR: 13.3 to 30.8) mm, mean DS% was 49 (38, 58), and median FFR was 0.70 (IQR: 0.62 to 0.77). VR-MI and non-MI control patients were similar in terms of baseline demographic and clinical factors (Table 1). The median follow-up period was 1,021 (IQR: 245 to 1,080) days.

Wall shear stress calculations and vessel-related myocardial infarction

Median values of WSS_prox, WSS_{total_lesion}, and DS% were significantly higher among patients who had a VR-MI compared with the non-MI control patients (p < 0.05 for all) (Table 2). Similarly, median baseline FFR was lower among patients who had VR-MI compared with non-MI control patients (p = 0.012). The median time to MI was 636 (IQR: 211 to 1,080) days.

In univariate marginal Cox analysis, a 3-Pa increase in WSS_prox and WSS_{total_lesion} was associated with a 23% (95% CI: 8% to 41%) and 20% (95% CI: 0.3% to 43%) increase in the hazard of VR-MI in 3 years, respectively (Table 3). However, WSS calculated in the upstream, middle, distal, and downstream segments of the lesion did not show significant association with MI.

The optimal threshold of WSS_prox for predicting MI was 4.71 Pa (stratified log rank p = 0.012) (Central Illustration). Given that WSS_prox showed the highest C-index (Harrell’s C-index = 0.65) and likelihood ratio test statistics among different areas of WSS measurement (Table 3), additional covariate adjustment was performed for WSS_prox. To validate the association of WSS_prox with VR-MI further, 5 separate multivariate marginal Cox models were constructed using anatomic and hemodynamic variables that might be associated with VR-MI. These included the following: 1) distance of lesion MLD to vessel ostium; 2) distance from proximal segment of lesion to MLD (this distance reflects the proximity of measured WSS_prox to the MLD); 3) DS%; 4) lesion length; and 5) lesion FFR. In final multivariable models, WSS_prox remained independently associated with MI (Table 4). Interestingly, in another multivariable model, WSS_{total_lesion} showed a trend but was not significantly associated with VR-MI (hazard ratio: 1.197; 95% CI: 0.998 to 1.435; p = 0.052) when adjusted for FFR.

Incremental prognostic value of proximal segment WSS over FFR for predicting myocardial infarction

Adding WSS_prox to a model containing FFR resulted in a significant increase in the global chi-square value for the model predicting VR-MI (p = 0.045 for WSS_prox) (Figure 1). Adding WSS_prox to FFR also resulted in a significantly improved prediction model with a net reclassification improvement of 0.69 (95% CI: 0.207 to 1.173; p = 0.005) and an integrated discrimination improvement of 0.11 (95% CI: 0.0256 to 0.187; p = 0.010). Figure 2A shows a representative case with baseline FFR of 0.71. After 3D reconstruction and CFD modeling of the baseline target vessel angiogram, WSS calculated in the proximal segment of the lesion was found to be high (WSS_prox = 7.12 Pa). This patient had a left anterior descending artery-related MI 498 days after this baseline angiogram. In contrast, Figure 2B demonstrates a WSS map from a non-MI control patient who did not have subsequent cardiac event at the end of 3-year follow-up. This patient also had an FFR of 0.71, but WSS at the proximal segment of the lesion in the left anterior descending artery was lower (WSS_prox = 1.43 Pa).

Clinical relevance of wall shear stress in hemodynamically significant lesions

In more advanced atherosclerotic plaques associated with hemodynamically significant lesions, a wide array of WSS values ranging from low to physiological to high are observed. In addition, as plaque progresses and regresses and the vessel remodels over time, the WSS evolves in response to the resulting anatomic changes (5,9,23–25). Within this complex environment, segments with low WSS tend to reside in the inner curvature of vessels, outer hips of bifurcations, and distal to lesions (26–28). These segments are associated with an ongoing inflammatory stimulus and subsequent plaque progression (23). In contrast, high-WSS segments tend to reside in proximal and middle segments of lesions and are also associated with proinflammatory and proatherosclerotic pathways that tend to transform into a more vulnerable plaque phenotype (11,24,29). Furthermore, segments with high WSS are often adjacent to areas of low WSS distal to a stenosis, thereby making it even more challenging to tease out the relative contributions of low versus high WSS in the ongoing pathogenesis of atherosclerosis. In the present study, higher WSS in proximal segments of lesions, as well as across entire lesions, but not at other locations were associated with a greater risk of MI at 3 years. Interestingly, we did not observe a significant relationship between lower WSS at any location and a greater risk of MI.

Several other vascular and plaque biomechanical parameters may be relevant in predicting future MI. In addition to WSS, axial plaque stress and plaque structural stress have been proposed as biomechanical indices of plaque vulnerability. Axial plaque stress is the fluid stress projected along the vessel centerline and is of a higher magnitude than WSS in the presence of an epicardial vessel obstruction (30). A recent retrospective report investigated the relationship between baseline hemodynamic parameters calculated using computed tomography angiography and acute coronary syndromes. The investigators found that delta FFR, high WSS_{total_lesion}, and axial plaque stress had incremental value over adverse plaque characteristics such as low-attenuation plaque, positive remodeling, napkin-ring sign, and spotty calcification for predicting acute coronary syndromes (31). Plaque structural stress has also been found to be higher in culprit lesions and ruptured plaques of patients with acute coronary syndromes (14,32,33). It is likely that a complex interplay among these local biomechanical forces leads to an increased accumulation of inflammatory, nonoxidized material within the plaque that decreases tensile strength and increases stress within the plaque, ultimately leading to rupture and MI (11,33). Taken together, data from this study suggest that focal biomechanical forces likely play a role in the development of MI.

Location of abnormal wall shear stress: Are proximal segments of lesions at higher risk?

Histological data suggest that macrophages and plaque rupture are more prevalent in proximal segments of coronary lesions, in contrast to distal shoulders of the plaque, where smooth muscle cells dominate (34). As blood accelerates through the stenosis, proximal and middle segments of coronary plaques are exposed to higher WSS (14,26). Supraphysiological or high WSS enhances the phosphorylation of p38, c-Jun, and ATF2 and has been associated with up-regulation of macrophages and apoptosis of smooth muscle cells that lead to matrix breakdown and plaque instability (6,7,29,35,36). Low WSS observed upstream and downstream from lesions also activates numerous inflammatory pathways, including inflammatory adhesion molecules, nuclear factor κB, and JNK-cAMP-dependent transcription factor ATF-2; however, phenotypically it is associated with stable plaque progression, rather than transformation to a vulnerable phenotype (37). To investigate these lesion-dependent differences in biological and fluid dynamics characteristics, we divided the index lesions into 5 segments corresponding to the fluid mechanical concepts outline earlier. These included proximal, middle, and distal thirds of the lesion, as well as the adjacent 5-mm upstream and downstream segments. It is also important that higher proximal WSS was observed in patients with subsequent MI regardless of lesion length (ranging from 8 to 55 mm). Although geometry is an important variable in the derivation of WSS, and 3D quantitative coronary angiography was used for angiographic reconstructions, other important variables in the derivation of WSS include blood flow boundary conditions (contrast frame count velocity and flow splitting) and viscosity. Indeed, we demonstrate that WSS_prox was associated with MI independent of DS% (ranging from 24% to 77%). We also demonstrate that higher WSS in the proximal segments of lesions was predictive of MI, independent of the baseline FFR or the location of the lesion within the vessel.

Clinical implications

There continues to be debate regarding the benefit of revascularization in patients with stable CAD (38,39). In the medical therapy arm of patients with stable CAD and hemodynamically significant lesions in FAME II, although a larger proportion of patients underwent target vessel revascularization over 3 years, only 8% of patients developed MI. The significantly improved prediction of VR-MI by FFR plus WSS_prox over FFR alone suggests that focal hemodynamic assessment may be useful for further risk stratification of patients with stable CAD and positive FFR. Future prospective studies are warranted to confirm our findings and to investigate whether intensified systemic or interventional therapies would improve outcomes for patients with stable CAD with lesions with FFR <0.80 and high WSS.

We also show that WSS_prox has an incremental prognostic value over FFR (a pressure-derived hemodynamic index). Although WSS is tangential, pressure is perpendicular to the lumen wall. Even though the geometry of the plaque may eventually couple or combine their action, these 2 forces are in fact linearly independent. This may explain why we find a poor correlation between WSS_prox and FFR and ultimately why adding WSS_prox to the information provided by FFR may add incremental clinical value to predict future MI.

Study limitations and future directions

Although the present study is limited by angiography as its imaging modality, it is notable that even angiographically derived WSS was predictive of future MI. Although angiograms are 2-dimensional representations of the vessel and have spatial resolution inferior to that of intravascular imaging, we made every effort to optimize our WSS calculations by using significantly different angiographic projections for target vessel reconstruction, adding side branches for accurate geometry, and then adding patient-specific boundary conditions. It has been shown that exclusion of side branches significantly alters blood flow estimation and hence WSS values (40). Significant effort was also made to ensure that boundary conditions were carefully defined and as patient-specific as possible through the 3D contrast velocity method, as well as the use of a well-known physiological model coupled with optimization procedures to assess flow splitting to branch outlets (19).

Whether CFD simulations using vessel geometries derived from intravascular ultrasound or optical coherence tomography combined with angiography could enhance the predictive accuracy of WSS for MI requires investigation. Furthermore, additional mechanistic insight into the precise location of subsequent MI could be gained by introducing intravascular optical coherence tomography or ultrasound during the follow-up angiography at the time of the event. Although we studied the association between mean circumferential WSS calculated across various lesion-specific segments and MI, future studies looking at the circumferential heterogeneity of WSS by dividing the circumference of the vessel into sectors could add to our findings.

In this study, steady-state WSS derived from angiograms reflecting daily clinical practice was predictive of future MI. Although from an engineering standpoint, the ideal patient-specific model is time-dependent and compliant and considers the 3D movement of the heart over the cardiac cycle, the results of this study suggest that such level of detail may not be necessary to predict clinical outcomes. Finally, the propensity score matching on the likelihood of VR-MI resulted in non-MI control subjects who were well matched in terms of cardiovascular risk factors but may decrease the generalizability of our study results to the general pool of patients with stable CAD and FFR <0.80. Hence, large prospective studies are warranted to: 1) establish definitively whether angiographically derived high WSS is sufficiently predictive of subsequent MI to serve as a clinical tool in risk stratification of patients with stable CAD; and 2) investigate whether intensified systemic antiatherosclerotic therapies or focal therapies such as percutaneous coronary interventions in lesions with high proximal WSS could affect outcomes.