Clinical value of aortic arch morphology in transfemoral TAVR: artificial intelligence evaluation

Abstract

Background:

The impact of aortic arch (AA) morphology on the management of the procedural details and the clinical outcomes of the transfemoral artery (TF)-transcatheter aortic valve replacement (TAVR) has not been evaluated. The goal of this study was to evaluate the AA morphology of patients who had TF-TAVR using an artificial intelligence algorithm and then to evaluate its predictive value for clinical outcomes.

Materials and methods:

A total of 1480 consecutive patients undergoing TF-TAVR using a new-generation transcatheter heart valve at 12 institutes were included in this retrospective study. The AA measurements were evaluated by deep learning, and then the approach index (I_A) was determined. The machine learning algorithm was used to construct the predictive model and was validated externally.

Results:

The area under the curve of the I_A model using random forest and logistic regression was 0.675 [95% confidence interval (CI): 0.586–0.764] and 0.757 (95% CI: 0.665–0.849), respectively. The I_A model was validated externally, and consistent distinctions were obtained. After we used a generalized propensity score matching method for continuous exposure, the I_A was the strongest correlation factor for major procedural events (odds ratio: 3.87; 95% CI: 2.13–7.59, P < 0.001). When leaflet morphology or transcatheter heart valve type was an interactive item with I_A, neither of them was statistically significant in terms of clinical outcomes.

Conclusion:

I_A may be used to identify the impact of AA morphology on procedural and clinical outcomes in patients having TF-TAVR and to help to predict the procedural complications.

Introduction

The safety and efficacy of transcatheter aortic valve replacement (TAVR) have been validated; it has gradually become an alternative treatment for surgical aortic valve replacement; and its indications have been expanded to low-risk patients^[1,2]. The transfemoral artery (TF) approach is currently the most used approach for TAVR. According to the registration study, results of transcatheter valve therapy, the proportion of TF-TAVR in high-, median-, and low-risk populations was 93.6%, 96.0%, and 98.0%, respectively^[3]. However, due to the need to deliver the complex vascular access, the procedural difficulty and the risk of vascular complications will undoubtedly increase^[4].

The morphology of the aortic arch (AA) in TF-TAVR is still a hot topic of debate due to the contradictory study results^[5-8]. Although, with the subsequent iterations of transcatheter heart valve (THV) design and the accumulation of TAVR experience, it is possible to achieve more accurate size selection and positioning, thus reducing the incidence of periprocedural complications. However, the unique anatomical and technical challenges brought by AA morphology have not been thoroughly explored and elaborated thus far^[8]. The accurate evaluation of peripheral vascular morphology is particularly important for successful delivery and THV implantation. Meanwhile, multidetector computed tomography, as the main approach for pre-TAVR planning, is not yet able to provide operators with a comprehensive evaluation of the TF approach due to its inherent two-dimensional image limitation. In a previous study, we speculated that a preprocedural prediction based on a deep learning (DL) algorithm is effective^[9]. In this context, this study is the largest to date, its goal being to comprehensively evaluate the AA morphology of patients undergoing TF-TAVR using DL algorithms and to further explore its predictive value for clinical outcomes.

Materials and methods

Study population

This retrospective observational study enrolled 1941 patients with symptomatic severe aortic stenosis who underwent TAVR using a new generation THV at 12 institutes from January 2021 to May 2023. In addition, an external validation cohort (n = 454) of equally treated patients was provided by three institutes (enrolled from January 2022 to May 2023). The inclusion criteria included: (1) In patients with symptomatic severe aortic stenosis, the effective orifice area (EOA) was <1.0 cm², and the mean pressure gradient was >40 mm Hg (1 mm Hg = 0.133 kPa); (2) New York Heart Association (NYHA) functional class ≥II; (3) Society of Thoracic Surgeons (STS) score <4%. The exclusion criteria included: (1) preprocedural multi-slice computed tomography (MSCT) showed that the aortic root diameter of the patient was ≥40 mm; (2) life expectancy of patients <1 year; (3) hypertrophic obstructive cardiomyopathy; (4) left ventricular thrombus or infective endocarditis^[10,11]. When identifying the study population for the derivation and external validation cohort, the following patients were excluded: (1) those without evaluable MSCT scans (n = 137); (2) those with images of suboptimal quality (n = 62); (3) those who received non-TF-TAVR (n = 344); or (4) those who received valve-in-valve TAVR (n = 59). The self-expandable devices in this study include Evolut Pro (Medtronic Inc., Minneapolis, MN, USA), Venus-A Plus (Qiming Medical Equipment Co., Ltd., Hangzhou, China), and VitaFlow (Microport Medical Technology Co., Ltd, Shanghai, China). The balloon-expandable devices were Sapien 3 (Edwards Lifesciences Inc., Irvine, CA, USA) and Prizvalve (Newmed Medical Technology Co., Ltd, Shanghai China), respectively^[12]. This study complied with the Declaration of Helsinki and was approved by the local ethics commissions. All patients provided written informed consent for the procedures and for subsequent data collection. In addition, the work has been reported in line with the STROCSS criteria^[13].

Data collection and definition

Data collection for patients undergoing TAVR was done at the institute level. Each patient had a dedicated case report form detailing baseline, computed tomography angiography (CTA), echocardiography, procedural, and in-hospital characteristics for the overall cohort.

All patients received cardiac CTA as prescribed. The scan image required clear anatomical structures with no artifacts. The left ventricle, aorta, and coronary arteries were well perfused. All CTA scans were in the Digital Imaging and Communications in Medicine format and were acquired from Siemens CT scanners (Siemens Medical Solutions, Forchheim, Germany). Section thickness was set as 1.0 mm or 0.75 mm.

Automatic segmentation and measurements were performed using the certified software (CVPILOT Software, TAVIMercy, Inc., Nanjing, China). Based on acquiring the three-dimensional AA model, the self-defined anatomical parameters were measured. First, the center line and the curvature of the entire AA were extracted. For the determination of a point on the curve, we used its parameter representation, that is r(t) = [x(t), y(t), z(t)] (t represents the spatial variable). Therefore, the curvature was calculated as follows:

$$C=\frac{\left|r^{\prime }\times r^{″}\right|}{{\left|r^{\prime }\right|}^3}=\frac{\sqrt{{\left(z^{″}{y}^{\prime }-y^{″}{z}^{\prime }\right)}^2+{\left(x^{″}{z}^{\prime }-z^{″}{x}^{\prime }\right)}^2+{\left(y^{″}{x}^{\prime }-x^{″}{y}^{\prime }\right)}^2}}{{\left({x^{\prime }}^2+{y^{\prime }}^2+{z^{\prime }}^2\right)}^{3/2}}$$

Among the above, r' and r” represent the r of XY and the XZ plane of the CTA scan, respectively.

Subsequently, the maximum curvatures of the descending aorta and ascending aorta were obtained successively and denoted as C₁ and C₂, respectively. Finally, the diameters of each layer from the maximum curvature plane of the ascending aorta to the annulus plane were measured: The maximum diameter of the ascending aorta and the annulus were denoted as L₁ and L₂, respectively (Fig. 1). After C₁, C₂ and L₁/L₂ of the derivation cohort were measured, a logistic regression model was constructed according to the procedural outcomes of all patients, and the model results are shown in Table S1 (available at: http://links.lww.com/JS9/D755). According to the preceding results, the approach index (I_A) was defined as:

Figure 1.

Automatic segmentation was conducted in the certified software (CVPILOT Software, TAVIMercy, Inc., Nanjing, China) and self-defined anatomical parameter measurements were performed. (A) Multidetector computed tomography scan of the aortic arch. (B, C) The maximum curvature point of the descending aorta (C₁) and ascending aorta (C₂) were acquired successively, and the maximum diameters of the ascending aorta (L₁) and the annular diameter (L₂), respectively, were measured.

I_A = 40% * (C₁ * 100) + 20% * (L₁/L₂*100) + 40% * (C₂ * 100)

Furthermore, none of the engineers was aware of the type of implanted THV until the I_As were measured.

‌ In addition, we defined the aortic leaflet phenotype. Tricuspid aortic valve usually consists of three leaflets, which are located on the left, right, and posterior sides of the aortic root^[14]. Bicuspid aortic valve (BAV) refers to dysplasia of the aortic valve, which leads to only 2 leaflets with ≤3 antagonistic borders between leaflets. In this study, the classification method used is the Sievers’ classification, which is divided into type 0 (no raphe), type 1 (one raphe), and type 2 (2 raphes) according to the number of raphes^[15].

End points

After calculating the I_A for all patients, the derivation cohort was divided into two groups: the low I_A (I_A < 33.8) group and the high I_A group (I_A ≥ 33.8), based on the results of the restricted cubic spline (RCS) (Fig. 2). The device success and procedural outcomes were defined according to the Valve Academic Research Consortium 3^[14]. Primary end point was the device success, and secondary end points included periprocedural major events (all-cause death, stroke, cardiovascular adverse events and major vascular complications), postprocedural major complications [≥mild paravalvular leakage (PVL), new permanent pacemaker implant, valve-in-valve implant, and aortic dissection, AA hematoma, and pseudoaneurysm], and procedural performance (procedural duration, fluoroscopy time, and contrast medium).

Figure 2.

The cutoff value of the approach index. The cutoff value of the approach index was 33.8 based on the results of the restricted cubic spline.

Statistical analysis

Categorical variables are reported as quantities (percentages), and continuous variables are reported as median and a range (25–75 percentile). Categorical variables were determined using the χ² test or the Fisher exact test, if applicable. The independent sample Student t-test was used to analyze continuous data.

The RCS was used to truncate the I_A, and the population was divided into the high I_A group and the low I_A group according to the results. The comparison between the groups was then carried out.

In the derivation cohort, potential predictors of procedural success were selected from 44 baseline variables using the least absolute shrink and selection operator (LASSO). This method obtained a refined model by constructing a penalty function that compressed the regression coefficients and removed the weaker features until the optimal number of features was reached (n = 9). Tenfold cross-validation was used to prevent overfitting.

After variable selection, the prediction model was established by random forest and logistic regression, respectively. (1) The variables selected by LASSO were incorporated into the prediction model: Random forest was constructed and denoted as RF; the logistic model was denoted as LR. (2) Only I_A was included as a predictor: Random forest was constructed and denoted as RF'; the logistic model was denoted as LR'.

In the external validation cohort, the area under curve (AUC), net reclassification index, and improved discriminant index were used to evaluate the differentiation of the prediction models.

Correlation study of I_A and postprocedural death/stroke included the following: (1) univariate and multivariate logistic regression analysis; (2) the generalized propensity score matching method for continuous exposures was used to further analyze the correlation between I_A and postprocedural death/stroke under the framework of causal inference; (3) subgroup analysis was used to further explain the correlation between I_A and postprocedural death/stroke.

For standard analyses, a bilateral P <0.05 was considered to indicate statistical significance. R-programming language (Version 4.2.2, R Foundation for Statistical Computing, Vienna, Austria) and The RStudio software (Version 2022.7.2.576, RStudio, PBC, Boston, MA) were used for analysis.

Results

Baseline characteristics of derivation cohort and validation cohort

The derivation cohort and validation cohort included 1480 and 313 patients (baseline characteristics are displayed in Table S2, available at: http://links.lww.com/JS9/D755). The derivation cohort baseline characteristics of the low I_A group and the high I_A group are shown in Table 1. Patients in the high I_A group had a higher proportion of the New York Heart Association ≥functional class III classification (82.0% vs. 75.5%, P < 0.001), a higher proportion of hyperlipidemia (26.9% vs. 18.8%, P < 0.001), previous MI (13.6% vs. 7.6%, P < 0.001), and stroke (15.2% vs. 7.8%, P < 0.001) compared with patients in the low I_A group. Interestingly, patients in the high I_A group had a higher proportion of atrial fibrillation (28.5% vs. 19.2%, P < 0.001). Baseline characteristics grouped by AV morphology and THV type are shown in Table S3 (available at: http://links.lww.com/JS9/D755) and Table S4 (available at: http://links.lww.com/JS9/D755).

Table 1.

Baseline characteristics of derivation cohort based on approach index

	Low IA group (n = 754)	High IA group (n = 726)	P value
Age, years	70.0 (64.0–77.0)	71.0 (66.0–78.0)	0.14
Male, % (n)	52.4 (395)	55.2 (401)	0.30
BMI, kg/m2	22.8 (20.0–25.5)	23.4 (20.6–26.2)	0.002
BSA, m2	1.7 (1.5–1.8)	1.7 (1.5–1.9)	<0.001
NYHA functional class III or IV, % (n)	75.5 (569)	82.0 (595)	<0.001
Diabetes, % (n)	20.8 (157)	28.2 (205)	0.001
Hypertension, % (n)	77.7 (586)	80.9 (587)	0.16
Hypercholesterolemia, % (n)	18.8 (142)	26.9 (195)	<0.001
Previous CAD, % (n)	26.0 (196)	29.5 (214)	0.15
Previous PCI, % (n)	14.9 (112)	18.3 (133)	0.085
Previous CABG, % (n)	6.5 (49)	8.0 (58)	0.31
Previous MI, % (n)	7.6 (57)	13.6 (99)	<0.001
Previous stroke, % (n)	7.8 (59)	15.2 (110)	<0.001
Peripheral artery disease, % (n)	12.6 (95)	13.5 (98)	0.66
STS score, %	5.4 (3.4–7.2)	5.7 (3.6–8.2)	0.003
COPD, % (n)	15.1 (114)	17.2 (125)	0.31
Chronic kidney disease, % (n)	14.9 (112)	13.1 (95)	0.37
Atrial fibrillation, % (n)	19.2 (145)	28.5 (207)	<0.001
Permanent pacemaker implant, % (n)	6.4 (48)	7.6 (55)	0.42

BMI, body mass index; BSA, body surface area; NYHA, New York Heart Association; CAD, coronary artery disease; PCI, percutaneous coronary intervention; CABG, coronary artery bypass grafting; MI, myocardial infarction; STS, Society of Thoracic Surgeons; COPD, chronic obstructive pulmonary disease.

The details of preprocedural echocardiography and CTA are shown in Table 2 and Table S5 (available at: http://links.lww.com/JS9/D755). Notably, patients in the high I_A group had a higher proportion of BAV morphology (32.5% vs. 22.5%, P < 0.001), and EOA index, left ventricle, and other major aortic root measurements (annulus, left ventricular outflow tract, sinuses, and ascending aortic diameter) were smaller than those in the low I_A group. Baseline characteristics grouped by AV morphology and THV type are shown in Table S6 (available at: http://links.lww.com/JS9/D755) and Table S7 (available at: http://links.lww.com/JS9/D755).

Table 2.

Procedural imaging assessment characteristics of derivation cohort based on approach index

	Low IA group (n = 754)	High IA group (n = 726)	P value
Transthoracic echocardiography
Bicuspid aortic valve,  % (n)	22.5 (170)	32.5 (236)	<0.001
Left ventricular ejection  fraction, %	50.0 (42.0–57.0)	51.5 (45.0–59.0)	0.014
Mean aortic valve  gradient, mm Hg	64.2 (54.4–75.0)	66.8 (55.0–77.0)	0.060
Peak aortic valve  velocity, m/s	5.4 (4.1–6.2)	5.7 (4.5–6.3)	0.003
≥Moderate mitral  regurgitation, % (n)	40.3 (304)	32.4 (235)	0.002
EOA, cm2	0.9 (0.7–1.1)	0.8 (0.6–1.1)	<0.001
Indexed EOA, cm2/m2	0.8 (0.6–1.0)	0.7 (0.5–0.9)	<0.001
LVEDV, mL	167.5 (115.0–214.0)	134.0 (88.0–192.0)	<0.001
LVESV, mL	89.5 (67.0–115.0)	77.0 (54.0–94.0)	<0.001
Computed tomography angiography
Aortic valve area, mm2	526.7 (424.6–658.4)	596.3 (488.2–696.9)	<0.001
Minimum annular  diameter, mm	22.0 (19.8–24.9)	25.1 (22.5–27.4)	<0.001
Maximum annular  diameter, mm	27.4 (25.0–30.4)	30.2 (27.5–32.7)	<0.001
Annular perimeter, mm	75.6 (69.1–84.1)	85.5 (77.1–93.2)	<0.001
Annular ellipticity	1.3 (1.2–1.4)	1.3 (1.2–1.4)	0.010
Perimeter-derived  annular  diameter, mm	24.4 (22.3–27.3)	27.6 (24.9–30.0)	<0.001
Area-derived annular  diameter, mm	24.9 (22.7–27.6)	28.0 (25.2–30.4)	<0.001
LVOT diameter, mm	25.2 (23.4–27.7)	27.4 (24.5–30.0)	<0.001
STJ diameter, mm	28.0 (26.2–30.9)	30.4 (27.3–33.2)	<0.001
Sinus of Valsalva  diameter, mm	30.6 (28.4–33.0)	32.5 (29.6–35.1)	<0.001
Ascending aorta  diameter, mm	35.4 (32.8–38.1)	36.7 (34.2–39.2)	<0.001
Horizontal aorta, % (n)	19.0 (143)	27.0 (196)	<0.001
Porcelain aorta, % (n)	4.4 (33)	5.2 (38)	0.52
AV calcification, mm3 a	1752.5 (1046.0–2551.0)	1889.5 (1047.0–2587.0)	0.30
C1	0.0300 (0.0250–0.0350)	0.0711 (0.0311–0.0753)	<0.001
L1/L2	1.23 (1.19–1.32)	2.04 (1.73–3.08)	<0.001
C2	0.0496 (0.0447–0.0549)	0.1016 (0.0518–0.1102)	<0.001
IA	28.1 (27.1–31.4)	45.6 (36.8–68.4)	<0.001

AV, aortic valve; EOA, effective orifice area; LVEDV, left ventricular end-diastolic volume; LVESV, left ventricular end-systolic volume; LVOT, left ventricular outflow tract; STJ, sinotubular junction; C₁, the maximum curvature of the descending aorta; C₂, the maximum curvature of the ascending aorta; L₁, the maximum diameter of the ascending aorta; L₂, the maximum diameter of the annulus.

^aAV calcification includes 4-mm superannular, annular, and 4-mm left ventricular outflow tract calcification.

Clinical outcomes of derivation cohort

The clinical outcomes of the derivation cohort are shown in Table 3. As expected, compared with the low I_A group, the operation duration in the high I_A group [113.0 (88.0–142.0) min vs. 108.0 (85.0–134.0) min (P < 0.001)] and the fluoroscopy time [14.9 (10.7–20.7) min vs. 13.4 (9.9–17.6) min, P < 0.001] were longer. The radiation amount [682.0 (574.0–786.2) mGy vs. 554.4 (461.8–674.6) mGy, P < 0.001] was larger. In addition to the incidence of tamponade and coronary artery obstruction, the incidences of conversion to surgical aortic valve replacement (1.9% vs. 0.4%, P = 0.012), malpositioning (6.7% vs. 1.9%, P < 0.001), ≥mild PVL (30.0% vs. 2.7%, P < 0.001), new permanent pacemaker implant (14.5% vs. 6.1%, P < 0.001), valve-in-valve implant (8.5% vs. 0%, P < 0.001), aortic root injury (2.3% vs. 0.1%, P < 0.001), and THV displacement (5.8% vs. 0.5%, P < 0.001) were significantly higher in the high I_A group than in the low I_A group. Furthermore, the incidence of life-threatening/major bleeding (4.5% vs. 1.5%, P < 0.001), AA hematoma (3.4% vs. 0%, P < 0.001), aortic dissection (3.9% vs. 0.3%, P < 0.001), and AA pseudoaneurysm (1.1% vs. 0%, P = 0.003) was much higher than that in low I_A group. Importantly, clinical outcomes during hospitalization in the high I_A group were poorer than those in the low I_A group (all P < 0.01), except for postprocedural acute kidney injury ≥stage 3.

Table 3.

Procedural details and in-hospital outcomes

	Low IA group (n = 754)	High IA group (n = 726)	P value
Procedural details
Device success, % (n)	98.3 (741)	85.3 (619)	<0.001
Device type
Evolut Pro, % (n)	6.5 (49)	7.3 (53)	0.58
Venus A, % (n)	45.4 (342)	44.1 (320)	0.56
VitaFlow, % (n)	29.8 (225)	31.7 (230)	0.36
Sapien 3, % (n)	11.5 (87)	12.0 (87)	0.82
Prizvalve, % (n)	6.8 (51)	5.0 (36)	0.21
Oversizing perimeter ≥15%, % (n)	47.5 (358)	26.0 (189)	<0.001
Predilation, % (n)	92.0 (694)	83.6 (607)	<0.001
Postdilation, % (n)	24.0 (181)	18.0 (131)	0.006
Sheath diameter, mm	18.0 (16.0–20.0)	16.0 (16.0–18.0)	<0.001
Procedure  duration, min	108.0 (85.0–134.0)	113.0 (88.0–142.0)	<0.001
Fluoroscopy time, min	13.4 (9.9–17.6)	14.9 (10.7–20.7)	<0.001
Contrast medium, mGy	554.4 (461.8–674.6)	682.0 (574.0–786.2)	<0.001
Conversion to SAVR, % (n)	0.4 (3)	1.9 (14)	0.012
Malpositioning, % (n)	1.9 (14)	6.7 (49)	<0.001
Valve-in-valve implant,  % (n)	0 (0)	8.5 (62)	<0.001
Annulus root injury, % (n)	0.1 (1)	2.3 (17)	<0.001
Tamponade, % (n)	1.6 (12)	2.6 (19)	0.23
Coronary artery obstruction, % (n)	0.3 (2)	0.8 (6)	0.17
Device displacement,  % (n)	0.5 (4)	5.8 (42)	<0.001
In-hospital outcomes
All-cause death, % (n)	0 (0)	2.2 (16)	<0.001
Cardiovascular adverse  events, % (n) a	0.7 (5)	2.6 (19)	0.006
Stroke, % (n)b	0 (0)	2.6 (19)	<0.001
Life-threatening/major  bleeding, % (n)	1.5 (11)	4.5 (33)	<0.001
Major vascular complications
AA hematoma, % (n)	0 (0)	3.4 (25)	<0.001
Aortic dissection, % (n)	0.3 (2)	3.9 (28)	<0.001
AA pseudoaneurysm, % (n)	0 (0)	1.1 (8)	0.003
Acute kidney injury stage  ≥3, % (n)C	5.4 (41)	5.8 (42)	0.86
New permanent pacemaker implant, % (n)	6.1 (46)	14.5 (105)	<0.001
Mild PVL at discharge,  % (n)	2.7 (20)	21.2 (154)	<0.001
Moderate PVL at  discharge, % (n)	0 (0)	8.8 (64)	<0.001
AV pressure gradient  ≥20 mmHg, % (n)	6.6 (50)	15.6 (113)	<0.001
Intensive care unit stay,  days	1.0 (1.0–2.0)	3.0 (2.0–3.0)	<0.001
In-hospital stay, days	7.0 (5.0–9.0)	11.0 (7.0–13.0)	<0.001

AA, aortic arch; AV, aortic valve; I_A, approach index; SAVR, surgical aortic valve replacement; PVL, paravalvular leakage.

^aCardiovascular adverse events included: myocardial infarction, angina pectoris, stroke, and heart failure rehospitalization.

^bStroke is defined as significant injury to the cerebral nervous system, which includes ischemic stroke, hemorrhagic stroke, stroke (no imaging/pathological classification), and symptomatic hypoxic-ischemic injury.

^CAcute kidney injury is defined as creatinine increase of ≥44.2 mmol/L within 24 hours (or an absolute creatinine increase of ≥26.5 umol/L within 48 hours according to specific criteria, or a known or presumed creatinine increase of ≥50% above baseline in the previous 7 days), and may be accompanied by symptoms of hypuria or anuria.

Notably, logistic analysis of the two subgroups (AV morphology and THV type) showed that when AV morphology or THV type and I_A were interactive items, neither had a statistically significant effect on clinical outcomes [tricuspid AV vs. tricuspid AV: 72.6% vs. 27.4%, P = 0.411; self-expandable valves (SEV) vs. balloon-expandable valves (BEV): 82.9% vs. 17.1%, P = 0.615] (procedural details are shown in Table S8, available at: http://links.lww.com/JS9/D755 and Table S9, available at: http://links.lww.com/JS9/D755).

Model development and clinical applicability

Baseline and procedural variables that best predicted clinical outcomes and complications in the derivation cohort were included in the model (included variables are shown in Table S10, available at: http://links.lww.com/JS9/D755). The variables were screened using the LASSO algorithm. The results of screening and the classification contributions are shown in Figure S1 (available at: http://links.lww.com/JS9/D755) and Figure S2 (available at: http://links.lww.com/JS9/D755), respectively. The results acquired above were validated by the tenfold cross-verification (Figure S3, available at: http://links.lww.com/JS9/D755). Then, the variables screened by LASSO were incorporated into the prediction model, and the model was constructed by random forest and logistic regression, respectively. The AUC was evaluated to determine the applicability of the I_A model in clinical practice. As shown in Fig. 3, the AUC of RF, LR, RF', and LR' were, respectively, 0.793 [95% confidence interval (CI): 0.712–0.873], 0.766 (95% CI: 0.676–0.857) and 0.675 (95% CI: 0.586–0.764) and 0.757 (95% CI: 0.665–0.849). This result confirms that the results of the I_A model in the derivation cohort are similar and stable in relation to those of the full-variable risk model constructed by the LASSO algorithm.

Figure 3.

Model development and performance of derivation cohort. Black, red, yellow, and green lines represent full-variable and approach index models using random forest (RF) methods and logistic regression (LR) analysis, respectively.

A generalized propensity score matching method for continuous exposures was used to further analyze the association between I_A and periprocedural major events. The average absolute correlation before matching was 0.18, and the correlation after matching was minimal to 0.07 (Figure S4, available at: http://links.lww.com/JS9/D755). The univariate and multivariate results before matching are shown in Table 4. Furthermore, the matched analysis found that I_A was the strongest correlation factor for periprocedural major events (odds ratio: 4.26; 95% CI: 2.55–8.63, P < 0.001).

Table 4.

Univariate and multivariate logistic regression analysis of periprocedural major events

Characteristics	OR (95% CI)	P value	OR (95% CI)	P value
BMI	1.46 (1.21–1.67)	<0.001
Diabetes	3.05 (1.93–4.30)	<0.001	2.16 (1.49–2.98)	0.001
Hypercholesterolemia	1.81 (1.35–2.73)	<0.001
Previous CAD	2.60 (1.44–5.42)	<0.001
Previous MI	1.69 (1.33–2.51)	<0.001
Peripheral artery disease	3.32 (1.57–6.26)	<0.001
COPD	1.85 (1.06–2.73)	0.047
Atrial fibrillation	1.95 (1.02–2.49)	0.036
Permanent pacemaker implant	1.32 (1.17–1.70)	0.003
STS score	2.92 (1.86–5.14)	<0.001	2.10 (1.42–3.65)	<0.001
IA	6.58 (3.75–12.48)	<0.001	4.62 (2.56–9.31)	<0.001

BMI, body mass index; CAD, coronary artery disease; CI, confidence interval; COPD, chronic obstructive pulmonary disease; I_A, approach index; MI, myocardial infarction; OR, odd ratio; STS, Society of Thoracic Surgeons.

External validation

The I_A model was externally validated, and consistent distinctions were obtained. The clinical outcomes of the validation cohort are shown in Table S11 (available at: http://links.lww.com/JS9/D755). In the validation cohort, AUC (Fig. 4), net reclassification index (RF as reference), and improved discriminant index (RF as reference) were used to evaluate the differentiation of the prediction models (Table S12, available at: http://links.lww.com/JS9/D755). There was no significant difference between prediction models that included only I_A and those that were constructed with LASSO-screened variables (all P > 0.05).

Figure 4.

The receiver operating characteristic curve of derivation cohort. The approach index (I_A) model achieved a similar net benefit in clinical outcomes compared to the least absolute shrink and selection operator (LASSO) model. Horizontal black lines indicate the assumption that no patient will experience the event. The solid gray line represents the assumption that all patients will experience this event.

Discussion

The I_A risk model is the first specifically developed and validated model to identify the impact of different AA morphology on clinical outcomes in patients who had TF-TAVR. The model incorporates the main clinical parameters for AA morphology in TF-TAVR (the maximum curvature of the descending and ascending aorta center lines and the ratio of the maximum ascending aorta diameter to the annular diameter). Before a procedure, operators can calculate the I_A in real time to evaluate the difficulty of procedures for the specific patient and to estimate the likelihood of major complications.

The primary consideration in TAVR is the approach, and the TF approach is the preferred approach for TAVR, and more than 95% of patients clinically choose it^[3]. However, TAVR with the TF approach is difficult for some patients due to challenging AA morphology or severe peripheral artery disease^[4]. Studies have shown that post-TAVR complications (including stroke, PVL, and conduction block) may increase the rate of postprocedural rehospitalization and all-cause mortality^[14]. It is worth noting that the previously reported incidence of vascular complications associated with TF-TAVR ranged from 1.9% to 30.7%^[15]. With the development of delivery systems and THVs, the incidence of vascular-related complications has decreased, but the overall incidence is still high^[1,2,16-19]. Therefore, preprocedural evaluation of the TF-TAVR approach is particularly important^[19].

On the one hand, aortic angulation has received more attention in previous studies. Extreme aortic angulation may result in poor THV localization and the need for recovery and/or redeployment, leading to increased incidence of postprocedural stroke, PVL, and PPM^[5]. However, the results of previous studies indicate that, with the subsequent iterations of THV design and the continuous improvement of implant technology, angulation is no longer effective for the new generation of BEV or SEV^[8]. Therefore, the impact of angulation on the results are still controversial and need to be further demonstrated. Considering AA as a three-dimensional anatomical structure, we propose to use 40% of C₂ and 20% of L₁/L₂ instead of angulation in this study, to more comprehensively evaluate the influence of AA morphology on the coaxiality and implanted depth of the THV. As expected, the coaxiality of the implanted THV decreased due to the increased procedural challenge associated with high I_A: The incidence of ≥mild PVL and new permanent pacemaker implant was obviously higher in the high I_A group (all P < 0.001); performance related to the implanted depth was also less desirable than in the low I_A group: A higher incidence of device displacement, malpositioning and valve-in-valve implantation was occurred in the high I_A group (all P < 0.001).

On the other hand, although the new generation of THV has a lower sheath and delivery system, operators need to pay careful attention to the entire length of the aorta before determining the best location for the implant. Panoulas et al emphasized the importance of AA, especially the descending aorta, during the preprocedural evaluation of TAVR^[20]. Alhafez et al found that the degree of AA distortion in the BAV population increased, which led to an increase in the incidence of thoracic aortic diseases in this population^[21]. Similarly, the proportion of BAV in the high I_A group in this study was significantly higher than that in the low I_A group (32.5 vs. 22.5%, P < 0.001). At present, the analysis of AA relevant parameters may be mainly studied by operators performing thoracic endovascular aortic repair^[22], and has not been systematically analyzed in the pre-TAVR evaluation. The study found that the incidence of life-threatening/major bleeding, AA hematoma, aortic dissection and AA pseudoaneurysm was much higher than that in low I_A group (all P < 0.001). Moreover, the weight of C₁ in I_A reached 40% after logistic regression analysis.

In view of the preceding problems of AA morphology, most of the previous studies were carried out to solve the problems from the aspect of procedural technique improvement^[7], device iteration^[23,24], and preprocedural evaluation improvement^[25]. Popma et al proposed best practice techniques for patients with extreme angulation (including the use of coplanar aortography, the use of stiffer guide wires, and the use of a delivery catheter/guide wire located at the apex of the left ventricle to apply a gentle forward force to stabilize the delivery catheter)^[7]. The new generation of SEV uses an improved nitinol annulus design to optimize radial force and a nitinol delivery catheter envelope, thereby improving the ability to treat patients with complex anatomical structures. The BEV delivery catheter has a bending property, which improves THV alignment and accurate deployment and significantly reduces the incidence of PVL^[23,24]. The results of this study showed no statistical significance of both AV morphology and THV type on clinical outcomes. Considering that C₂ and L₁/L₂ jointly influence the changes of aortic angulation, while C₁ is closely related to the occurrence of vascular complications, it is necessary to comprehensively evaluate the preceding anatomical parameters and then propose the I_A.

In recent years, DL has been widely used in medical image segmentation. In this study, the CVPILOT system based on the nnUNET algorithm was used to automatically segment the angulation and aortic root^[7,9,26]. From the results of this study, the I_A risk model obtained the variable screening and validation of the LASSO algorithm (including machine learning, standard univariate analysis, and final cross-validation). Importantly, to our knowledge, no specific risk assessment algorithm has been developed for patients having TF-TAVR. Our study is the largest to date to highlight the feasibility of I_A to predict TF-TAVR procedural and clinical outcomes. The weight assignment of I_A was generated from a large initial summary of baseline and procedural characteristics as the strongest factor in identifying the impact of different AA morphologies on clinical outcomes. When I_A is higher (I_A ≥ 33.8), patients are at increased risk of procedural complications and poor prognosis. In the ROC analysis of this study, the I_A risk model provided a consistent net benefit in predicting clinical outcomes, so the model could be implemented in the periprocedural period of TAVR. Notably, the guidelines were followed in the practice this study examined retrospectively^[10,11]. The rather young patients in this study and the relatively high proportion of BAV may challenge that simply referring to the guidelines was adequate. Future studies should confirm whether I_A and its parameters (C₁, C₂, and L₁/L₂) can improve TAVR outcomes, specifically reducing the incidence of procedural complications.

Study limitations

However, there are some limitations in this study. First, this study is retrospective, so selection bias is unavoidable. Second, in current studies, there is no unified understanding of the incidence of postprocedural complications using old and new THVs^[27-29], especially because the procedural performance of the different devices of the new generation THVs may be comparable^[29]. Notably, some of the devices used in this study have not been confirmed by large-scale randomized controlled trials. The preceding factors may influence the further expansion of the application of the I_A risk model in TF-TAVR. Third, this study lacked an independent core laboratory to analyze medical images (including curvature measurement based on DL) and independent clinical events, which leads to certain measurement errors. Fourth, this study did not conduct aggregate statistics on the hardness of guide wires used in the study; the stiffer guide wire may provide better control and transmission effect in the complex AA morphology. Finally, the use of different types of THV and corresponding randomized controlled trials in the future may shed further light on this topic.

Conclusions

In contemporary TAVR practice, we developed the first I_A risk model for evaluating TF approaches. The model is based on a comprehensive evaluation of AA, particularly on the ascending and descending aorta, and may help personalize risk stratification for patients undergoing TAVR. Meanwhile, the model is based on the automatic segmentation of the DL algorithm, which can quickly measure AA parameters and can be easily applied in routine practice. Additionally, we demonstrated that I_A ≥33.8 was associated with an increased risk of procedural complications and poor prognosis in TAVR patients.

Ethical approval

The protocol was approved by the Ethics Committee of Xijing Hospital (approval number: KY-20192138-C-1).

Consent

Written informed consent was obtained from patients for publication of this study and accompanying images. Copies of the written consent are available for review by the Editor-in-Chief of this journal on request.

Sources of funding

This work was supported by the National Key R&D Program of China (2020YFC2008100); the Development and Transformation of New Technology and Construction of Precision Diagnosis and Treatment System for Transcatheter Interventional Diagnosis and Treatment of Structural Heart Diseases (2022YFC2503400); Research on Key Techniques of Minimally Invasive Treatment for Valvular Heart Diseases (2023-YBSF-105).

Author’s contribution

Substantial contributions to the conception or design of the work: Y.M., Y.L. Acquisition, analysis, or interpretation of data for the work: M.Z., P.J., F.C., Y.Y., T.Y. Drafting the work or revising it critically for important intellectual content: Y.L., G.Z., T.N., R.L., X.P., Y.W., J.Y. Final approval of the version to be published: Y.G., H.Z., L.W., J.L., G.Z., K.X., X.S., Y.Z., B.N., H.L., M.T., Z.J., Y.Y., X.P., Y.W., J.Y. Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved: Y.G., H.Z., L.W., J.L., J.C.L., X.P., Y.W., J.Y.

Conflicts of interest disclosure

The authors have no conflicts of interest to declare.

Research registration unique identifying number (UIN)

ClinicalTrials.gov Protocol Registration System (NCT05044338).

Guarantor

Jian Yang.

Provenance and peer review

Not commissioned, externally peer-reviewed.

Data availability statement

Yes.

Assistance with the study

None.

Presentation

None.