Potential relationship between high wall shear stress and plaque rupture causing acute coronary syndrome

The relationship between high wall shear stress (WSS) and plaque rupture (PR) in longitudinal and circumferential locations remains uncertain. Overall, 100 acute coronary syndrome patients whose culprit lesions had PR, documented by optical coherence tomography (OCT), were enrolled. Lesion-specific three-dimensional coronary artery models were created using OCT data. WSS was computed with computational fluid dynamics analysis. PR was classified into upstream-PR, minimum lumen area-PR, and downstream-PR according to the PR’s longitudinal location, and into central-PR and lateral-PR according to the disrupted fibrous cap circumferential location. In the longitudinal 3-mm segmental analysis, multivariate analysis demonstrated that higher WSS in the upstream segment was independently associated with upstream-PR, and thinner fibrous cap was independently associated with downstream-PR. In the PR cross-sections, the PR region had a significantly higher average WSS than non-PR region. In the cross-sectional analysis, the in-lesion peak WSS was frequently observed in the lateral (66.7%) and central regions (70%) in lateral-PR and central-PR, respectively. Multivariate analysis demonstrated that the presence of in-lesion peak WSS at the lateral region, thinner broken fibrous cap, and larger lumen area were independently associated with lateral-PR, while the presence of in-lesion peak WSS at the central region and thicker broken fibrous cap were independently associated with central-PR. In conclusion, OCT-based WSS simulation revealed that high WSS might be related to the longitudinal and circumferential locations of PR.


Introduction
Coronary plaque rupture (PR) is the most common mechanism leading to acute coronary syndrome (ACS) [1]. Several morphological characteristics, including a thin fibrous cap, large necrotic core, and foamy macrophage infiltration, are considered potential features of rupture-prone vulnerable plaques [2]. Recent studies have demonstrated that high wall shear stress (WSS) is associated with plaque vulnerability progression and PR onset [3][4][5]. These studies speculated the potential relationship between high WSS and atherosclerotic progression by averaging the WSS in the longitudinal and circumferential directions in vessel segments; however, WSS often varies significantly in the longitudinal and circumferential directions within a luminal narrowing region. Thus, the direct relationship between regional WSS and local plaque findings remains uncertain.
Optical coherence tomography (OCT) is an intracoronary imaging modality that can provide in vivo high-resolution coronary plaque images and clearly visualize the exact location of the fibrous cap disruption in PR lesions [6]. We developed a novel WSS assessment method using OCT-derived lumen profiles. This method employs a computational fluid dynamics (CFD) analysis over the entire patient-specific three-dimensional coronary model generated from the twodimensional OCT data and preserves the exact correspondence between the CFD solution and OCT data. Therefore, it has a high potential to assess the relationship between PR and WSS more accurately and directly than conventional techniques. However, because the novel OCT-based WSS simulation cannot evaluate vessel tortuosity, the accuracy of the OCT-based WSS simulation in curved stenotic lesions remains uncertain. Therefore, we first conducted a computational experimental study to clarify how the WSS value changes according to the vessel tortuosity using several curved stenotic models mimicking real coronary arteries. Then, we conducted a retrospective clinical study to investigate the detailed relationship between PR and WSS using the novel OCT-based WSS simulation in patients who underwent OCT for ACS culprit lesions with PR.

Model creation and CFD simulation
Straight stenosis models (straight-models) and curved stenosis models (curved-models) were created using Image J 1.51 (National Institutes of Health, Bethesda, MD, USA) and SolidWorks 2021 (Dassault Systèmes SolidWorks Corporation, Waltham, MA, USA) [7]. First, we created a straightmodel with a reference lumen diameter of 3.5 mm and a length of 20 mm with a minimum lumen area (MLA) site in the center of the longitudinal axis. The diameter stenosis severity rates were set as 50% and 65% (i.e., percent area stenosis was set as 75% and 87.75%), respectively. At the MLA site, the lumen center was set away from the center of the vessel to mimic eccentric plaque distribution (Supplementary Fig. 1a). To create the curved-models, we first extracted the central axis from the bidirectional angiographic images of normal left anterior descending arteries from seven patients ( Supplementary Fig. 1b). Then, the curvedmodel was created by bending the straight-model according to the extracted vascular central axis ( Supplementary  Fig. 1a-2). Finally, CFD analysis was performed to evaluate the WSS distribution in a total of 16 vascular models (one 50% straight-model, seven 50% curved-models, one 65% straight-model, and seven 65% curved-models). Each model had one peak WSS region (in-lesion peak WSS) where WSS value was the highest in the model. The location of the inlesion peak WSS region was evaluated three-dimensionally using the arc and distance ( Supplementary Fig. 1c). Details of the WSS assessment are described in Supplementary Materials and Methods.

Study design and patient population
The Kobe OCT ACS registry is a multicenter, retrospective observational registry that can be used to evaluate the morphological characteristics of ACS culprit lesions in patients undergoing primary percutaneous coronary intervention for ACS [8]. In total, 436 patients from four Japanese institutions (Kobe University (Kobe, Japan), Osaka Saiseikai Nakatsu Hospital (Osaka, Japan), Hyogo Prefectural Awaji Medical Center (Sumoto, Japan), and Hyogo Prefectural Himeji Cardiovascular Center (Himeji, Japan)) were enrolled in the registry. Among them, 145 consecutive ACS patients whose culprit lesions had PR documented by OCT were screened. ACS was defined as unstable angina, non-ST elevation myocardial infarction, and ST elevation myocardial infarction.
The exclusion criteria were in-stent restenosis, coronary artery bypass grafting, insufficient OCT image quality, culprit lesion predilated with a balloon before OCT imaging, and ruptured plaques that were insufficient to reconstruct three-dimensional models.
Lesion-specific three-dimensional coronary artery models were created using the OCT data. WSS was computed using CFD analysis at a single core laboratory (Ehime University, Toon, Japan). We retrospectively collected all data from patient records.
The requirement of written informed consent was waived because of the retrospective nature of the study. The study protocol complied with the principles of Declaration of Helsinki and was approved by the ethics committee of each institution.

OCT image acquisition and analysis
After thrombus aspiration, OCT imaging and off-line OCT analysis were performed as previously described [6]. OCT images were analyzed by two independent investigators (YF, YT), blinded to the clinical presentation, using validated plaque characterization criteria. PR was defined as the presence of a fibrous cap discontinuity with a clear cavity formed inside the plaque. At the ruptured portion, the broken fibrous cap thickness was measured at the site where the fibrotic cap remnant was thinnest (Fig. 1), as the broken fibrous cap thickness. In addition, the thinnest fibrous cap thickness of non-PR region was measured at the same cross-section. A detailed OCT analysis was described in Supplementary Materials and Methods.

Three-dimensional geometry reconstruction and CFD simulations
Lesion-specific three-dimensional coronary artery models were created using cross-sectional OCT images by tracing the luminal contour. Specifically, at the ruptured portion, the tracing of the luminal edge of the residual fibrous cap was smoothly extrapolated to reconstruct the luminal contour before PR ( Supplementary Fig. 2). We selected the ruptured plaques with a significant amount of residual fibrous cap to accurately reconstruct the original vessel. Supplementary Fig. 3 showed the representative crosssections with irregular flaps protruding too much into the lumen or with excessive deformation, thereby preventing accurate reconstruction of the original vessel. If such a cross-section existed, the original vessel was reconstructed based on the relationship with the previous and the following cross-sections, and the cases with insufficient reconstruction of the original vessels were excluded. CFD analysis was performed using ANSYS Fluent (ANSYS, Inc., Canonsburg, PA, USA). Detailed three-dimensional geometry reconstruction and CFD simulations were described in Supplementary Materials and Methods and Supplementary  Fig. 2.

Relationship between WSS and PR locations
Relationships between WSS and PR locations were assessed using the following two methods. Two independent investigators (KS, KK) blinded to the clinical presentation performed the analyses; discordance was resolved by consensus.

Longitudinal 3-mm segmental analysis
Each culprit lesion was subdivided into five 3-mm segments with respect to the MLA location at the central segment. To investigate the regional impact of luminal narrowing on the relationship between WSS and PR, we focused on the segment with the MLA and two consecutive segments: proximal (upstream: UP1, UP2) and distal (downstream: DN1, DN2) segments (Fig. 2a). The average WSS was calculated for each 3-mm segment. All lesions were classified into upstream-PR, MLA-PR, and downstream-PR according to the PR's longitudinal location (Supplementary Materials and Methods); the 3-mm segments where the PR was located were subdivided into cross-sections with and without PR, and the average WSS was compared among them (Fig. 2a).
Circumferential analysis On each PR cross-section, the vessel wall was divided into two circumferential regions: PR and non-PR. In the PR region, the ruptured plaque cavity could be clearly detected (Fig. 2b, c). We measured the maximum, average, and minimum WSS values in each region (PR and non-PR regions) on every PR cross-section (Fig. 2c).

Relationship between the in-lesion peak WSS location and PR types
In each PR lesion, the cross-sections with the in-lesion peak WSS were selected (Fig. 2a). On such cross-sections, the vessel wall was circumferentially divided into subregions. The PR regions were divided into three equally separate subregions: central and two lateral regions; the non-PR region was divided into subregions: two semi-lateral and other regions (Fig. 2b). The in-lesion peak WSS location was determined according to these regions (central, lateral, semi-lateral, or other). In the same cross-section, each PR was classified into 3 types according to the disrupted fibrous cap location. Central-PR (C-PR) and lateral-PR (L-PR) were defined as PRs whose disrupted fibrous caps were located at the central and lateral regions, respectively. Other-PR was defined as a non-classifiable PR due to the lack of a detectable fibrous cap disruption site because of imaging artifacts or massive thrombus presence (Figs. 1, 2) [9]. When there were different PR types in different cross-sections within the same lesions, the aforementioned classification of the in-lesion peak WSS cross-section was employed.

Statistical analysis
Statistical analyses (Supplementary Materials and Methods) were performed using SPSS for Windows version 26 (IBM SPSS Inc., Chicago, IL, USA). In the longitudinal 3-mm segmental analysis, univariate and multivariate analyses were implemented to identify the parameters associated with each longitudinal PR type. In the circumferential analysis, univariate and multivariate analyses were implemented to identify the parameters associated with L-PR or C-PR. In each statistical test other than the post hoc analysis of categorical data, the significance level was set at p < 0.05 and 95% confidence intervals (CIs) were reported, unless otherwise stated.

Computational experiment
Two straight-models with 202 cross-sections and 14 curvedmodels with 1414 cross-sections were analyzed. The maximum, average, and minimum WSS values in each crosssection of the straight-models were compared with those of the corresponding cross-sections of the curved-model. Overall, the 65% model had higher WSS values than the 50% model. There were no significant differences between the straight-and curved-models in terms of the maximum, average, and minimum WSS values in both 50% and 65% stenosis models (Supplementary Table 1). The in-lesion peak WSS location was almost identical between the straight-and curved-models in both 50% and 65% stenosis models. The average differences in the arc and distance of the in-lesion peak WSS from those of the straight-model were 5.2 ± 3.3 degrees and − 0.08 ± 0.15 mm, and 0.4 ± 4.6° and 0.01 ± 0.06 mm in the 50% and 65% stenosis model, respectively. The average difference in the in-lesion peak WSS value between the straight-and curved-models was 0.4 ± 0.81 Pa in the 50% stenosis model, and 1.4 ± 2.8 Pa in the 65% stenosis model (in-lesion peak WSS value of 50% stenosis straight-model, 21.2 Pa; in-lesion peak WSS value of 65% stenosis straight-model, 67.8 Pa; Supplementary Table 2).

Baseline characteristics
Among 145 consecutive ACS patients, 45 were excluded because of in-stent restenosis (n = 6), insufficient OCT image quality (n = 15), culprit lesions predilated with a balloon before OCT imaging (n = 9), and ruptured plaques that were insufficient to reconstruct the three-dimensional models (n = 15). Finally, 100 ACS patients with PR detected by OCT were enrolled. The baseline characteristics of patients were summarized in Supplementary Table 3.

Longitudinal 3-mm segmental analysis
The incidences of upstream-PR, MLA-PR, and downstream-PR were 45%, 40%, and 15%, respectively. The highest average WSS in the upstream-PR was in UP1 and that in the MLA-PR was in the MLA segment (Table 1). In the downstream-PR, the highest WSS was in the MLA segment, followed by DN1. Regarding OCT findings, the broken fibrous cap was significantly thinner and the in-lesion peak WSS was significantly lower in the downstream-PR (

Circumferential analysis of the relationship between WSS and PR regions
A total of 1470 PR cross-sections and 1830 non-PR crosssections were analyzed. In the PR cross-sections, irregular protruding flaps ( Supplementary Fig. 3a) were observed in 98 cross-sections (0.07%) from 39 patients, and excessive deformation flaps (Supplementary Fig. 3b) were observed in 91 cross-sections (0.06%) from 41 patients. The average WSS of the PR cross-sections was significantly higher than that of the non-PR cross-sections (

Circumferential analysis of the relationship between WSS and PR types
Among the 100 lesions with PR, 51 (51%) were classified as L-PR, 42 (42%) as C-PR, and 7 (7%) as other-PR. The in-lesion peak WSS was frequently observed in the lateral region (66.7%), followed by the semi-lateral region (13.7%) in the L-PR and central region (70%) in the C-PR (Fig. 3). Table 3 compared the OCT and WSS-related parameters between the L-PR and C-PR. Compared with the C-PR, the L-PR had a significantly larger lumen area at the inlesion peak WSS cross-sections (1.  Table 3).
Multivariate analysis demonstrated that the presence of the in-lesion peak WSS at the lateral region, a thinner broken fibrous cap, and a larger lumen area at the in-lesion peak WSS cross-sections were independent parameters associated with L-PR. Moreover, the presence of the in-lesion peak WSS at the central region and a thicker broken fibrous cap were independent parameters associated with C-PR (Table 4).

Discussion
In this study, we first conducted a computational experimental study to clarify how the OCT-derived WSS value changes according to vessel tortuosity. We found that in both 50% and 65% stenosis models, the maximum, average, and minimum WSS values in each cross-section of the straight-models were not statistically different from those of the corresponding cross-sections of the curved-models. The in-lesion peak WSS value and location were almost identical between the straight-and curved-models in both 50% and 65% stenosis. Subsequently, we investigated the potential relationships between WSS values and the PR location in ACS patients. By conducting CFD analysis on a three-dimensional coronary model augmented with OCT geometry, we evaluated the WSS values at the exact measurement location where the  local OCT features were measured. Besides the traditional longitudinal 3-mm segmental comparison, we conducted a circumferential regional analysis to compare the WSS values between the regions with PR and those without PR within the same longitudinal location. Thus, we identified a direct relationship between PR and local WSS distribution in ACS patients from a global lesion level to a local level.

OCT-derived WSS and vessel tortuosity
In the initial computational experiment, we obtained two main findings. First, by comparing the WSS values in each corresponding cross-section between the straight-and curved-models, we found that maximum, average, and minimum WSS values were not statistically different between these two models. Second, the in-lesion peak WSS locations of each model were almost identical, and there was no difference in the in-lesion peak WSS value. Indeed, the average arc and distance difference of the in-lesion peak WSS region between the straight-and curved-models were 5.2 ± 3.3° and − 0.08 ± 0.15 mm in the 50% stenosis model, and 0.4 ± 4.6° and 0.01 ± 0.06 mm in the 65% stenosis model, respectively. In both severity models, the z-axis difference was below the OCT spatial resolution (− 0.08 ± 0.15 mm in the 50% stenosis model; 0.01 ± 0.06 mm in the 65% stenosis model < 0.1 mm). The difference in the in-lesion peak WSS location decreased as the stenosis decreased (5.2 ± 3.3 degrees in the 50% stenosis model, 0.4 ± 4.6° in the 65% stenosis model). Importantly, the target lesions enrolled in the subsequent clinical study were all ACS culprit lesions, whose average percent area stenosis was 88.2%. This value was almost identical to the percent area stenosis of the 65% stenosis model (87.75%). Thus, based on the results of the experiment, we consider that the impact of vessel tortuosity on high WSS distribution in the severely stenotic lesion could be minimal, and that evaluating the relationship between WSS and PR using OCT-derived WSS in the subsequent clinical study simulations is feasible.

High WSS is associated with plaque destabilization and rupture
In non-diseased arteries, high WSS is considered atheroprotective [10], whereas in diseased arteries with luminal narrowing, higher WSS can promote further vulnerable changes leading to thin-cap fibroatheroma [11][12][13], which could result in PR. We enrolled ACS patients with PR and demonstrated that higher WSS was associated with PR. Several studies have reported that high WSS can induce thinning of the fibrous caps by promoting plasmin-induced metalloproteinase activity, smooth muscle cell apoptosis, decreased matrix synthesis, accelerated angiogenesis, and transformation to a vulnerable phenotype [14][15][16]. Studies have demonstrated the potential contribution of high WSS to plaque vulnerability using various imaging modalities [11][12][13]. However, data regarding the relationship between high WSS and PR are limited. A CFD study with intravascular ultrasonography of ruptured plaques in 20 ACS patients demonstrated a strong correlation between focal elevation in WSS and PR location (k = 0.79) [3]. Moreover, Kumar  [4] demonstrated that higher WSS in the proximal segments of atherosclerotic lesions is a predictive factor of subsequent myocardial infarction occurrence within 3 years.
These results indicate that high WSS may play a pivotal role in PR initiation in patients with high-risk vulnerable plaque.
In the longitudinal segmental analysis, we demonstrated that the average WSS of the PR cross-sections was significantly higher than that of the non-PR cross-sections within the same longitudinal segment. We further demonstrated that the location of the highest average WSS corresponded to the PR location in the upstream-PR and MLA-PR. In line with our results, Kumar et al. [4] demonstrated that higher WSS in the proximal segments of atherosclerotic lesions is predictive of subsequent myocardial infarction occurrences. Considering that higher proximal WSS was observed in patients with subsequent myocardial infarction regardless of lesion location and morphology, it might play an important role in PR, leading to subsequent myocardial infarction. During this study, 15% of the PR cases were observed in the downstream segment. Interestingly, in the downstream-PR, the location of the highest WSS did not correspond to the PR location, but rather that of the second-highest WSS. These results indicate that PR may be determined by balancing several factors. In the downstream-PR, the broken fibrous cap was significantly thinner and the in-lesion peak WSS was significantly lower than that in the other-PR types. Therefore, we speculate that downstream-PR could be induced with relatively lower WSS if there is a highly vulnerable plaque in the downstream segment. Generally, lower WSS is observed in the downstream segment rather than in the MLA and upstream segments [17]. Downstream-PR might occur only if a plaque with a higher vulnerability Table 4 Univariate and multivariate logistic regression analyses of lateral-PR and central-PR CI confidence interval, OR odds ratio, PR plaque rupture, WSS wall shear stress a The central, semi-lateral, and other regions were coded 0, and the lateral region was coded 1 b The lateral, semi-lateral, and other regions were coded 0, and the central region was coded 1 can be broken with a relatively low WSS. This characteristic of downstream-PR was consistent with that of L-PR in circumferential direction and might be relevant to the large proportion of L-PR in downstream-PR (80%). Further studies with larger populations are warranted for confirmation.

Direct relationship among WSS, plaque vulnerability, and PR
Hemodynamic parameters are significantly affected by several geometric characteristics including luminal narrowing, lesion location, and lesion length [11]. Although previous studies demonstrated a potential relationship between high WSS and PR, they employed longitudinal and circumferential averaging of WSS values in a certain range. To the best of our knowledge, no study has clarified the direct relationship between WSS and vulnerable plaque features by eliminating the potential influence of geometric conditions. By comparing WSS between regions with and without specific OCT findings within the same longitudinal ranges, we could investigate the relationship more specifically and directly by eliminating the influence of geometric factors on WSS. Accordingly, we found that WSS and plaque vulnerability were significantly higher in regions with PR than in those without features for the same longitudinal regions with the same degree of stenosis. These data suggest the importance of high WSS on PR regardless of geometric factors, including lesion location and luminal narrowing. A necropsy study [9] reported that the circumferential PR incidence rates at the lateral, central, and other regions were 59%, 35%, and 6%, respectively, which corresponds with our results (L-PR, 51%; C-PR, 42%; other-PR, 7%). Interestingly, we demonstrated colocalization of in-lesion peak WSS and the site of fibrous cap disruption, which led to ACS events. In L-PR, the in-lesion peak WSS was observed in the lateral or semilateral region in > 80% of the ruptured plaques; however, in C-PR, it was most frequently observed in the central region (70%) (Fig. 3). Although spatial colocalization is not mechanistic evidence of causality, these results suggest that identifying focal areas of high WSS over atheroma harboring other advanced morphologic features may improve detection of plaques before symptomatic rupture.

Potential factors associated with the prediction of circumferential PR type
In the multivariate analysis, the presence of in-lesion peak WSS at the lateral region, a thinner broken fibrous cap, and larger lumen area at the in-lesion peak WSS site were independently associated with L-PR. These data suggest that L-PR was induced by relatively lower WSS at a highly vulnerable plaque with a thinner fibrous cap, whereas C-PR was induced by higher WSS at a plaque with a thicker fibrous cap. During a previous autopsy study including sudden death cases, Burke et al. [18] demonstrated that among 16 PRs occurring on exertion, 75% (12 PRs) were C-PR; however, among 20 PRs occurring without exertion (rest PRs), 65% (13 PRs) occurred in the shoulder region, suggesting that physical exertion tends to shift the PR site from the shoulder to the midcap where the WSS is expected to be the highest. Although no direct assessment was performed on the local WSS in this study, these data might support our findings that the C-PR had significantly higher in-lesion peak and average WSS than the L-PR. Interestingly, C-PR had a significantly smaller lumen area at the cross-section with in-lesion peak WSS and a significantly thicker broken fibrous cap than L-PR. These data suggest that C-PR might occur at a plaque with a relatively thicker fibrous cap triggered by relatively high WSS, whereas L-PR can be induced by relatively low WSS because of advanced plaque vulnerability. Integrating morphological features, plaque vulnerability, and adverse hemodynamic conditions might improve the accuracy of identifying high-risk plaques that can cause future ACS.

Limitations
This study has some limitations. First, the sample size of the computational experiment was small, and only the left anterior descending artery was mimicked. The distribution of low WSS may change further if we create a model with a strong bending, such as the left circumflex artery or right coronary artery. However, through the computational experiment, we have confirmed that the distribution of high WSS could not be affected by vessel tortuosity in lesions with severe stenosis. Second, this was a retrospective and observational clinical study with a small sample size, which might increase the possibility of selection bias. The present study analyzed the relationship between high WSS and PR in the longitudinal and circumferential directions separately, but the sample size was too small for a combined analysis of characteristics in both directions. Nonetheless, our results could provide important mechanistic insight regarding the relationship between PR and WSS. Third, there are limitations of three-dimensional geometry reconstruction using cross-sectional OCT imaging. Lumen tracing was performed as an implicit assumption to reconstruct the luminal contour before PR. The geometry of the remaining flap after PR, such as protruding flap or deformation flap, may prevent accurate reconstruction of the original vessel. Although the percentage of cross-sections with such flaps in this study was negligible, there could be some differences in lumen area and lumen morphology between the original and reconstructed vessels, which might affect WSS-related variables. In addition, none of the patients had a history of coronary spastic angina as a baseline clinical characteristic, and no cases of suspected coronary artery spasm were recorded. However, if the coronary spasm triggered ACS, the lumen area of the reconstructed vessel could be different from that at the time of ACS, which might also affect WSS-related variables. Finally, there are limitations of the OCT-based CFD simulation. In this study, CFD simulation was performed based on the non-pulsatile and steady flow model, whereas in vivo coronary flow was pulsatile. The WSS value increased as coronary flow velocity increased, and varied on cardiac cycle timing. However, in our additional simulations, we found that there was no difference in high WSS distribution between steady and pulsate flow simulations, although low WSS distribution varied in both value and distribution. In this study, we focused on the distribution of high WSS to observe the relationship between the location of PR and WSS. When low WSS would be evaluated, pulsatile flow simulation with a transient model is necessary because the magnitude and distribution of the low WSS varies with the flow velocity. Moreover, the external compression of the coronary artery, such as bending motion or squeezing, has an impact on flow conditions. It was impossible to specify the time phase (systole or diastole) using OCT automatic pullback; therefore, we cannot evaluate the influence caused by these external forces. Coronary CT data might help us evaluate the impact of coronary structural changes on WSS distribution and future cardiac events.

Conclusion
The OCT-based CFD simulation revealed that high WSS might be related to the longitudinal and circumferential location of PR. Integrating morphological features, plaque vulnerability, and adverse hemodynamic conditions might improve the accuracy of identifying high-risk plaques that can cause future ACS.
Funding This research received no grant from any funding agency in the public, commercial, or not-for-profit sectors.

Data availability
The authors confirm that the data supporting the findings of this study are available within the article and/or its supplementary materials. Consent to participate Written informed consent was waived due to the retrospective nature of the study.