Predictors of target lesion restenosis after endovascular therapy for lower-extremity atherosclerotic peripheral artery disease: a real-world single center cohort

Abstract

Background

Lower-extremity peripheral artery disease often requires endovascular therapy. Target-lesion restenosis (TLRS) remains frequent, but current risk stratification is limited.

Methods

We retrospectively analyzed 1,005 lesions from 917 patients undergoing endovascular therapy for lower-extremity atherosclerotic disease at a single center (2019–2024) with scheduled surveillance (approximately 6, 12, and 24 months) and administrative censoring at 36 months. The primary endpoint was time to first TLRS ≥ 50%, adjudicated by duplex ultrasound (peak systolic velocity ratio criteria) or CTA/DSA when available. Lesion‑level Fine–Gray competing‑risk models (death as a competing event) with patient‑level clustering were used. A prespecified Core model (clinical, anatomic, procedural/device covariates and C‑reactive protein) was compared with an Extended model additionally including the Atherogenic Index of Plasma (AIP) and log‑transformed Systemic Immune–Inflammation Index (SII). Models were internally validated by 1,000 bootstrap resamples and assessed at the 24‑month horizon using the C‑index, calibration, Brier score, integrated discrimination improvement (IDI), net reclassification improvement (NRI), and decision‑curve analysis.

Results

Among 917 patients (1,005 lesions), mean age was 75.2 ± 11.7 years and 68.6% were male. At 24 months, cumulative incidences were 31.4% for TLRS, 14.9% for clinically driven target‑lesion revascularization, and 12.3% for death. In multivariable analyses, GLASS stage III, lesion length ≥ 150 mm, residual stenosis > 20%, chronic kidney disease, and Rutherford class 4–6 predicted higher TLRS risk, whereas good distal runoff was protective. In the Extended model, AIP and SII remained independent predictors and improved the optimism‑corrected 24‑month C‑index from 0.68 to 0.73 (Δ0.05, p = 0.002), reduced the Brier score (0.19 to 0.17), yielded positive IDI and NRI, and preserved good calibration. Exploratory analyses suggested lower TLRS risk with cilostazol and low‑dose rivaroxaban plus aspirin.

Conclusions

AIP and SII improved lesion-level prediction of TLRS beyond conventional covariates. The Extended model may support risk-tiered surveillance after endovascular therapy; external validation is warranted.

Introduction

Lower-extremity peripheral artery disease (LEPAD), also termed arteriosclerosis obliterans (ASO), is caused by atherosclerotic narrowing or occlusion of the lower‑extremity arteries. LEPAD spans asymptomatic atherosclerosis to chronic limb‑threatening ischemia (CLTI). Among symptomatic patients, clinical presentation ranges from intermittent claudication to CLTI, non‑healing ulcers, and gangrene, and is associated with limb loss and increased mortality [1, 2]. Its prevalence increases with age and with established cardiovascular risk factors such as smoking, diabetes, hypertension, and dyslipidemia [3, 4]. Endovascular therapy (EVT) is a first‑line revascularization strategy for many femoropopliteal and infrapopliteal lesions, using balloon angioplasty with or without stenting and drug‑coated technologies to improve patency and clinical outcomes [5–8]. Restenosis after EVT remains common, with reported 1–2‑year rates of approximately 20–40% depending on lesion location, device strategy, and patient risk profile [9–11]. Restenosis can lead to recurrent symptoms and repeat revascularization, increasing morbidity and health‑care costs [12, 13]. Therefore, improved lesion‑level risk stratification is needed to guide device selection, adjunctive therapy, and post‑procedural surveillance.

Traditional risk stratification has relied on a limited set of clinical factors (e.g., diabetes, kidney disease, smoking, and advanced Rutherford class) and angiographic features such as long lesions, chronic total occlusion, poor runoff, and suboptimal residual stenosis [11, 14]. Systemic inflammatory and thrombotic markers (e.g., C‑reactive protein, neutrophil‑to‑lymphocyte ratio, and D‑dimer) have also been linked to restenosis, suggesting a role for immune‑inflammatory activation and a prothrombotic milieu [15–18]. However, widely used lesion‑level prognostic models that integrate anatomic complexity, procedural quality, device strategy, and composite metabolic and inflammatory indices are still lacking. In particular, whether AIP and SII add prognostic information beyond GLASS stage, runoff, lesion length, residual stenosis, and device strategy remains uncertain [17–19].

We focused on composite indices derived from routine laboratory tests and standardized anatomic staging. The Systemic Immune–Inflammation Index (SII; platelets×neutrophils/lymphocytes) captures thrombo‑inflammatory burden [17, 18], whereas the Atherogenic Index of Plasma (AIP; log₁₀[TG/HDL‑C]) summarizes atherogenic dyslipidemia. In parallel, the Global Limb Anatomic Staging System (GLASS), proposed in the Global Vascular Guidelines for CLTI, provides a standardized description of infrainguinal anatomic complexity and runoff [19]. We hypothesized that adding AIP and SII to a clinical/anatomic model would improve prediction of time to target‑lesion restenosis (TLRS) after EVT. Accordingly, we developed and internally validated a postprocedural, lesion‑level prognostic model integrating AIP and SII with GLASS stage, detailed lesion and procedural characteristics, device strategy, and conventional clinical covariates. Secondary aims were to characterize the cumulative incidence of clinically driven target‑lesion revascularization (CDTLR) and to explore the association of cilostazol and low‑dose rivaroxaban plus aspirin with TLRS. By comparing Core and Extended models, this study specifically quantifies the incremental predictive and potential clinical value of AIP and SII beyond conventional clinical, anatomic, procedural, and device-related risk factors.

Methods

Study design and study population

We included consecutive patients undergoing EVT for lower-extremity atherosclerotic disease between January 2019 and October 2024. Lesions were excluded if patients had acute limb ischemia, non‑atherosclerotic arterial disease, missing key laboratory values required to calculate AIP or SII, missing CRP measurement, or if follow‑up imaging was unavailable. This study was conducted in accordance with the Declaration of Helsinki. The protocol was reviewed and approved by the Ethics Committee of the Affiliated Hospital of Southwest Medical University, which waived the requirement for informed consent because this was a retrospective analysis of de-identified routinely collected data and posed no more than minimal risk to participants. All data were anonymized prior to analysis.

Diagnosis and baseline clinical assessment

Lower-extremity PAD/ASO was diagnosed based on typical ischemic symptoms and signs (intermittent claudication, rest pain, or ischemic tissue loss), abnormal non‑invasive hemodynamic testing when available [ankle–brachial index (ABI) ≤ 0.90 and/or monophasic or absent arterial Doppler waveforms], and imaging evidence of atherosclerotic stenosis or occlusion. All patients underwent diagnostic digital subtraction angiography (DSA) as part of the index endovascular therapy (EVT), which served as the reference for anatomic characterization. Duplex ultrasound and/or computed tomography angiography (CTA) were used pre‑procedurally for screening and procedural planning when available. Symptoms were graded using the Rutherford classification (I–VI) and dichotomized as Rutherford 1–3 versus 4–6 (claudication versus chronic limb‑threatening ischemia) for modelling [2, 19]. Baseline demographic and clinical characteristics (e.g., age, sex, body mass index, smoking status [current, former, or never], hypertension, diabetes mellitus, dyslipidemia, glycated hemoglobin [HbA1c] when available, coronary artery disease, and chronic kidney disease) were abstracted from the electronic medical record at the time of the index procedure. Hypertension, diabetes, and dyslipidemia were defined according to physician diagnosis, prior documentation, or ongoing treatment; chronic kidney disease was defined as an estimated glomerular filtration rate < 60 ml/min/1.73 m² or documented structural kidney disease lasting ≥ 3 months. Data were collected independently by two investigators using a standardized case report form, any discrepancies were resolved by review of the source documents.

Endovascular procedures and medical therapy

All EVTs were performed by experienced vascular surgeons under local anesthesia with or without conscious sedation according to a standardized departmental protocol. Vascular access was obtained via antegrade common femoral, contralateral femoral crossover, or brachial approaches depending on lesion location and access feasibility. After diagnostic angiography, lesions were crossed using standard guidewires and support catheters, followed by balloon angioplasty; stent implantation and/or drug‑coated devices were used at the operator’s discretion based on lesion morphology, vessel diameter, lesion preparation, flow‑limiting dissection, residual stenosis, reimbursement, and device availability. In general, drug-coated balloons or drug-eluting stents were preferred for femoropopliteal de novo or restenotic lesions after adequate lesion preparation when no major contraindication or availability limitation existed, whereas plain balloon angioplasty and/or bare-metal stenting was used for lesions with severe calcification, bailout stenting needs, infrapopliteal targets, cost/availability constraints, or operator concern about drug-coated technology [8]. Target lesions were categorized by arterial segment (iliac, femoropopliteal, or infrapopliteal). For anatomic complexity, the infrainguinal arterial tree of the treated limb (femoropopliteal and infrapopliteal segments) was staged using the Global Limb Anatomic Staging System (GLASS I–III) based primarily on pre‑procedural digital subtraction angiography (and CTA when available) by two vascular specialists blinded to outcomes; disagreements were adjudicated by a senior vascular surgeon. Lesion length was measured on angiography using calibrated catheter markers. Calcification severity was graded qualitatively (none/mild versus moderate/severe) and chronic total occlusion was defined as 100% luminal occlusion on angiography. Distal runoff was assessed as the number of patent tibial arteries to the foot (0–3) and dichotomized as good (≥ 2) versus poor (≤ 1). Final residual stenosis was visually estimated by the operator; technical success was defined as residual stenosis < 30% with restoration of antegrade flow, and we additionally recorded residual stenosis > 20% as a more stringent marker of a suboptimal angiographic result. Clinical improvement was defined descriptively as improvement of at least one Rutherford category at the first scheduled post-procedure visit without major amputation or repeat target-lesion revascularization before that visit; standardized quality-of-life instruments were not routinely available and were therefore not analyzed. Periprocedural anticoagulation and haemostasias followed standard practice. Post‑procedural medical therapy (antiplatelet therapy, statins, and risk‑factor modification) was prescribed according to contemporary PAD guidance and individual bleeding risk. At our institution, dual‑pathway inhibition with rivaroxaban 2.5 mg twice daily plus aspirin is routinely considered after lower‑extremity revascularization when not contraindicated, which contributed to the high discharge use observed in this cohort. Calcification was adjudicated primarily on the final pre-intervention angiographic run/DSA, with CTA used as supportive information when available; moderate-to-severe calcification was assigned when radiopaque calcium was visible on both sides of the arterial wall or extended over ≥ 50% of the target lesion, whereas absent, focal, or unilateral calcium was classified as none/mild [20]. Final residual stenosis was estimated semi-quantitatively from the final angiographic run as 1−(minimum lumen diameter/reference diameter)×100 using calibrated catheter markers and the nearest visually normal reference segment, with two projections used when available; because this was a visual angiographic estimate rather than a continuous quantitative angiographic measurement, we analyzed the clinically relevant categories of technical success (< 30%) and suboptimal result (> 20%).

Follow-up and endpoint definitions

Follow-up was organized at approximately 6, 12, and 24 months after the index EVT and annually thereafter; all patients were scheduled for systematic duplex ultrasound (DUS) at these fixed windows, with additional visits or imaging prompted by recurrent or worsening limb symptoms. During periods of COVID-19-related outpatient restrictions, in-person visits were occasionally delayed or converted to telephone contact, but DUS was rescheduled at the earliest available visit and TLRS dating required the first objective imaging evidence rather than symptoms alone. Time zero for all time‑to‑event analyses was the date of the index EVT for each lesion. At each follow‑up visit, Rutherford class and limb symptoms were reassessed and a standardized duplex ultrasound protocol was applied to the treated segments and inflow/outflow vessels. CTA or DSA was obtained when duplex findings were inconclusive or when clinically indicated. The primary endpoint was time from the index EVT to first occurrence of TLRS ≥ 50%. The primary TLRS endpoint was surveillance-detected or clinically prompted but always imaging-confirmed, and it was not restricted to clinically driven restenosis. Restenosis was defined by duplex ultrasound peak systolic velocity ratio (PSVR) > 2.4 in native or drug‑coated balloon–treated segments or PSVR ≥ 2.85 in stented segments, or by CTA/DSA evidence of ≥ 50% diameter stenosis in the target lesion. PSVR > 2.4 was retained because it was the prespecified threshold in our vascular laboratory and is widely used in femoropopliteal drug-coated balloon trials, while PSVR ≥ 2.85 has been reported as an optimal duplex threshold for in-stent restenosis [21, 22]. A stricter definition of TLRS ≥ 70% (PSVR ≥ 3.5 or ≥ 70% angiographic stenosis) was used in sensitivity analyses. The main secondary endpoint was clinically driven target‑lesion revascularization (CDTLR), defined as repeat endovascular intervention or surgical bypass of the target lesion prompted by recurrent limb symptoms together with imaging evidence of significant restenosis or occlusion. Surveillance‑triggered, asymptomatic reinterventions were not counted as CDTLR. All‑cause death was treated as a competing event in analyses of TLRS and CDTLR. Telephone follow‑up supplemented clinic visits to ascertain vital status, hospitalizations, and reinterventions. If patients reported clear worsening of ischemic symptoms, they were advised to return for imaging, and symptoms alone were not used to define TLRS in the absence of objective imaging, although they contributed to the classification of CDTLR when a reintervention was performed. Lesions without TLRS or CDTLR were administratively censored at the date of last contact or 36 months after the index EVT, whichever came first. The target lesion was defined as the index treated segment, including the treated balloon/stent segment and immediately adjacent margins; restenosis was counted as TLRS only when the recurrent ≥ 50% stenosis/occlusion occurred within this predefined treated segment, whereas new stenoses in remote untreated segments of the same limb were not classified as TLRS unless they extended into the treated segment.

Covariates and laboratory measurements

Baseline laboratory variables were obtained from blood samples drawn within 24–72 h before the index procedure and analyzed in the hospital’s central laboratory using standardized assays. HbA1c was abstracted when measured within 3 months before EVT (available in 872 of 917 patients [95.1%]) and was used only for descriptive reporting. Fasting lipid profiles included total cholesterol, triglycerides (TG), low‑density lipoprotein cholesterol (LDL‑C), and high‑density lipoprotein cholesterol (HDL‑C); fasting plasma glucose was measured enzymatically. Serum creatinine, urea, fibrinogen, D‑dimer, and homocysteine were quantified using automated chemistry or coagulation analyzers. High‑sensitivity C‑reactive protein (CRP) was measured by immunonephelometry and log₂‑transformed for regression modelling to reduce skewness and allow an interpretable “per doubling” effect. Complete blood counts provided neutrophil, lymphocyte, and platelet counts, from which the neutrophil‑to‑lymphocyte ratio (NLR) and the Systemic Immune–Inflammation Index (SII), defined as platelets×neutrophils/lymphocytes (all counts in 10⁹/L), were derived [17, 18]. SII was natural‑log transformed (log‑SII) before modelling. The Atherogenic Index of Plasma (AIP) was calculated as log₁₀[TG (mmol/L)/HDL‑C (mmol/L)], such that higher values reflect a more atherogenic lipid profile, and the triglyceride–glucose (TyG) index as ln[TG (mg/dL) × fasting plasma glucose (mg/dL)/2]. When TG, glucose, and cholesterol fractions were reported in mmol/L, they were converted to mg/dL using standard conversion factors (TG mg/dL = mmol/L × 88.57; glucose mg/dL = mmol/L × 18; cholesterol [LDL‑C/HDL‑C] mg/dL = mmol/L × 38.67) before TyG calculation. All laboratory variables were treated as continuous predictors in the primary models. Angiographic and procedural covariates included target vessel territory (iliac, femoropopliteal, or infrapopliteal), lesion length (dichotomized at 150 mm), total occlusion versus stenosis, moderate–severe calcification, distal runoff (≥ 2 versus ≤ 1 patent arteries), and use of stent implantation and/or drug‑coated devices. Because GLASS is defined for infrainguinal disease and was originally proposed for CLTI, GLASS stage (I–III) was assessed for the infrainguinal arterial tree (femoropopliteal and infrapopliteal segments) of the treated limb based on baseline angiography [19], and Rutherford class was included to account for clinical presentation. Technical success was defined as a final residual stenosis < 30%. In addition, a more stringent cutoff (residual stenosis > 20%) was evaluated as a marker of a suboptimal angiographic result that may still predispose to restenosis. Missingness for non‑key descriptive covariates was < 5% per variable (range 0.3–4.9%). Complete‑case analysis was used for the affected descriptive or exploratory analyses. Lesions with missing key laboratory values required for index calculation were excluded a priori as described above. The 150-mm cutoff was prespecified as a pragmatic marker of long disease burden informed by TASC II femoropopliteal lesion categories and prior restenosis literature, but it was not assumed to carry identical meaning in femoropopliteal and infrapopliteal arteries; segment-specific anatomic complexity was therefore also captured by target-vessel territory and GLASS stage [14, 19, 23]. Discharge statin therapy was recorded as none, moderate/low-intensity, or high-intensity according to the documented prescription, and chronic anti-inflammatory therapy was defined as long-term systemic corticosteroid or immunosuppressive/anti-inflammatory medication use before EVT.

Statistical analysis

All analyses were performed on a lesion basis, with clustering by patient to account for multiple lesions in the same individual. Continuous variables are reported as mean ± standard deviation (SD) or median (interquartile range, IQR), and categorical variables as counts and percentages. Between‑group comparisons by TLRS status used Student’s t‑test or the Wilcoxon rank‑sum test for continuous variables and χ² or Fisher’s exact tests for categorical variables. Cumulative incidences of TLRS, CDTLR, and death were estimated using the cumulative incidence function, treating death as a competing event for TLRS and CDTLR. To screen candidate predictors, univariable Fine–Gray proportional subdistribution hazards models were fitted for each covariate and are reported as subdistribution hazard ratios (sHR) with 95% confidence intervals (CIs). For multivariable prognostic modelling, we prespecified a Core model based on prior literature and clinical plausibility (age, diabetes mellitus, chronic kidney disease, Rutherford class 4–6 versus 1–3, GLASS stage, lesion length ≥ 150 mm, distal runoff, residual stenosis > 20%, stent implantation, drug-coated device use, and log₂‑transformed CRP). An Extended model was then constructed by adding AIP (per 0.10‑unit increase) and log‑SII. Fine–Gray models used robust sandwich variance estimators with patients as clusters. Proportional subdistribution hazards assumptions were checked graphically and with time‑interaction terms. Model goodness-of-fit and performance at the 24‑month horizon were assessed using time‑dependent C‑index, calibration-in-the-large/intercept, calibration slope, calibration plots, and the Brier score. Internal validation was performed by 1,000 bootstrap resamples to obtain optimism‑corrected performance estimates and to quantify the incremental value of the Extended versus Core model (ΔC‑index, IDI, and category‑free NRI). Clinical utility was evaluated using decision curve analysis across clinically relevant 24‑month TLRS risk thresholds. To address treatment heterogeneity, drug-coated device use (DCB/DES versus no drug-coated device) was included as a prespecified device covariate in Core and Extended models, and exploratory AIP×drug-coated-device and log-SII×drug-coated-device interaction terms were examined. In exploratory analyses, associations of cilostazol and rivaroxaban 2.5 mg twice daily plus aspirin with TLRS were examined in Fine–Gray models adjusted for the Core covariates, but these medications were not included as predictors in the final prognostic models. All statistical tests were two‑sided and p values < 0.05 were considered statistically significant. Analyses were performed in R (Version 4.4, R Foundation for Statistical Computing, Vienna, Austria).

Two additional sensitivity/exploratory analyses were performed: first, the Extended model was repeated after excluding patients with more than one treated target lesion or bilateral treated limbs; second, time-window analyses compared early TLRS (≤ 12 months) with later TLRS (> 12 to 24 months) for AIP and log-SII, and exploratory medication models examined statin therapy, statin intensity, and chronic anti-inflammatory therapy with adjustment for Core covariates.

Results

Study population and follow-up

Among 1,184 lesions in 1,050 patients treated with EVT for lower-extremity ASO during the study period, 179 lesions were excluded because of non-atherosclerotic disease (n = 72), acute limb ischemia or embolic occlusion (n = 18), missing baseline laboratory data required to calculate AIP, SII, or CRP (n = 47), or lack of any post-procedural follow-up (n = 42), leaving 1,005 lesions from 917 patients for analysis (Fig. 1). During follow-up, TLRS occurred in 316 lesions, CDTLR in 150 lesions, and 124 patients died without prior TLRS, corresponding to 24-month competing-risk cumulative incidences of 31.4% (95% CI 28.2–34.6) for TLRS, 14.9% (95% CI 12.3–17.6) for CDTLR, and 12.3% (95% CI 9.9–14.7) for death (Fig. 2). Clinical improvement by at least one Rutherford category at the first scheduled post-procedure visit was observed in 768 of 917 patients (83.8%). The number of TLRS events was sufficient to support the prespecified multivariable prognostic modeling. Of the 917 patients, 833 (90.8%) contributed one treated target lesion, whereas 84 (9.2%) contributed more than one treated lesion; among these 84 patients, 52 had multiple treated segments in the same limb and 32 had bilateral treated limbs during the index treatment episode. After excluding these 84 multi-lesion or bilateral-limb patients, the single-lesion sensitivity cohort included 833 patients/lesions, and the Extended model remained directionally consistent for AIP (sHR per 0.10 increase 1.16, 95% CI 1.05–1.28; p = 0.003) and log-SII (sHR per 1-unit increase 1.22, 95% CI 1.04–1.43; p = 0.014).

Fig. 1.

Study flow diagram. Flow chart of patient and lesion selection for the lesion‑level cohort. Among 1,184 lesions (1,050 patients) undergoing endovascular therapy for lower-extremity peripheral artery disease, 179 lesions were excluded for nonatherosclerotic disease, acute limb ischemia or embolic occlusion, missing baseline laboratory data required to calculate AIP, SII or CRP, or lack of postprocedural follow‑up, yielding 1,005 lesions (917 patients) for analysis

Fig. 2.

Cumulative incidence for TLRS. Competing‑risk cumulative incidence functions showing 24‑month incidences of 31.4% for TLRS, 14.9% for clinically driven target‑lesion revascularization, and 12.3% for death (death treated as a competing event for TLRS and CDTLR)

Baseline clinical and laboratory characteristics

Baseline clinical and laboratory characteristics of the 917 patients, stratified by TLRS group (No TLRS n = 611; TLRS n = 306) (Table 1). The mean age was 75.2 ± 11.7 years, and 68.6% were male. Rutherford class 4–6 was present in 28.0%. Current, former, and never smokers accounted for 50.7%, 25.1%, and 24.2% of patients, respectively. Coronary artery disease was documented in 46.6%. Mean HbA1c was 7.3 ± 1.5% among patients with available values. Median CRP was 8.52 (3.65–9.65) mg/L and median homocysteine was 9.63 (7.56–12.98) µmol/L. AIP, SII, and NLR were higher in the TLRS group (Table 1). Statin therapy at discharge was prescribed in 834 patients (90.9%), including high-intensity statin therapy in 322 (35.1%); chronic anti-inflammatory therapy was uncommon (38 patients, 4.1%).

Table 1.

Baseline clinical and laboratory characteristics of patients by TLRS group

Variable	Overall (n = 917 patients)	No TLRS (n = 611)	TLRS (n = 306)	p value
Age (years), mean ± SD	75.2 ± 11.7	75.2 ± 11.8	75.2 ± 11.5	0.941
Male, n (%)	629 (68.6%)	424 (69.4%)	205 (67.0%)	0.507
Rutherford class 4–6, n (%)	257 (28.0%)	163 (26.7%)	94 (30.7%)	0.227
Diabetes mellitus, n (%)	553 (60.3%)	338 (55.3%)	215 (70.2%)	< 0.001
Chronic kidney disease, n (%)	282 (30.7%)	162 (26.5%)	120 (39.2%)	< 0.001
Hypertension, n (%)	784 (85.5%)	512 (83.8%)	272 (88.9%)	0.043
Smoking status, n (%)				0.696
Never	222 (24.2%)	152 (24.9%)	70 (22.9%)
Former	230 (25.1%)	155 (25.4%)	75 (24.5%)
Current	465 (50.7%)	304 (49.8%)	161 (52.6%)
Coronary artery disease, n (%)	427 (46.6%)	269 (44.0%)	158 (51.6%)	0.035
HbA1c (%), mean ± SD (n = 872)	7.3 ± 1.5	7.1 ± 1.4	7.6 ± 1.5	< 0.001
Triglycerides (mmol/L), mean ± SD	2.48 ± 1.02	2.48 ± 1.04	2.47 ± 0.99	0.857
HDL-C (mmol/L), mean ± SD	1.33 ± 0.31	1.32 ± 0.32	1.34 ± 0.29	0.665
LDL-C (mmol/L), mean ± SD	4.08 ± 1.61	4.11 ± 1.61	4.02 ± 1.60	0.421
AIP (log10[TG/HDL-C]), mean ± SD	0.24 ± 0.10	0.23 ± 0.09	0.27 ± 0.11	< 0.001
D-dimer, median [IQR]	2.41 [1.74–3.69]	2.50 [1.74–3.69]	2.41 [1.72–3.68]	0.835
CRP (mg/L), median [IQR]	8.52 [3.65–9.65]	7.65 [3.65–9.65]	8.59 [3.88–9.65]	0.036
Homocysteine (µmol/L), median [IQR]	9.63 [7.56–12.98]	9.63 [7.56–12.98]	9.63 [7.56–13.68]	0.891
Platelet count (×109/L), mean ± SD	255 ± 96	254 ± 94	258 ± 99	0.541
SII, median [IQR]	370 [203–682]	350 [195–641]	426 [245–745]	< 0.001
NLR, median (IQR)	1.63 [0.94–2.65]	1.52 [0.88–2.43]	1.79 [1.07–2.86]	0.002
Cilostazol at discharge, n (%)	404 (44.1%)	270 (44.2%)	134 (43.8%)	0.965
Rivaroxaban 2.5 mg + aspirin, n (%)	151 (16.4%)	130 (21.2%)	21 (6.8%)	< 0.001
Statin therapy at discharge, n (%)	834 (90.9%)	569 (93.1%)	265 (86.6%)	0.002
High-intensity statin	322 (35.1%)	223 (36.5%)	99 (32.4%)	0.230
Chronic anti-inflammatory therapy, n (%)	38 (4.1%)	25 (4.1%)	13 (4.2%)	0.926

Lesion and procedural characteristics

Lesion-level anatomic and procedural characteristics are shown for all treated lesions (n = 1,005 lesions,, Table 2). Overall, 60.0% of target lesions were femoropopliteal, 22.0% infrapopliteal, and 18.0% iliac. TLRS lesions occurred most frequently in femoropopliteal (66.5%) and infrapopliteal (24.7%) segments (p < 0.001). According to the GLASS classification, GLASS II lesions were most common overall (39.0%), followed by GLASS III (29.0%). Lesions with TLRS were more likely to have GLASS III disease (42.7% vs. 22.6%) and less likely to have GLASS I disease (19.9% vs. 37.6%; p < 0.001). Long lesions (≥ 150 mm; 37.0% overall) and total occlusions (54.0% overall) were frequent, and were more common among TLRS than non-TLRS lesions (54.1% vs. 29.2% and 65.5% vs. 48.8%, respectively; both p < 0.001). Moderate-to-severe calcification was more common among TLRS lesions (48.7% vs. 33.1%; p < 0.001), whereas good distal runoff (≥ 2 patent below-knee vessels) was less common (41.1% vs. 59.9%; p < 0.001). Stent implantation was used in 58.1% of procedures overall and was more common among TLRS lesions (65.8% vs. 54.6%; p = 0.001). Drug-coated devices were used in 35.8% of procedures and were less common among TLRS lesions (28.5% vs. 39.2%; p = 0.001). Residual stenosis > 20% (21.4% overall) was more common among TLRS lesions (36.1% vs. 14.7%; p < 0.001), whereas technical success (residual stenosis ≤ 30% without flow-limiting dissection) was less common among TLRS lesions (93.4% vs.97.4%, p < 0.001).

Table 2.

Lesion and procedural characteristics of all treated lesions by TLRS status

Variable	Overall (n = 1,005 patients)	No TLRS (n = 689)	TLRS (n = 316)	p value
Target vessel, n (%)				< 0.001
Iliac	181 (18.0%)	153 (22.2%)	28 (8.9%)
Femoropopliteal	603 (60.0%)	393 (57.0%)	210 (66.5%)
Infrapopliteal	221 (22.0%)	143 (20.8%)	78 (24.7%)
GLASS stage, n (%)				< 0.001
I	322 (32.0%)	259 (37.6%)	63 (19.9%)
II	392 (39.0%)	274 (39.8%)	118 (37.3%)
III	291 (29.0%)	156 (22.6%)	135 (42.7%)
Lesion length ≥ 150 mm, n (%)	372 (37.0%)	201 (29.2%)	171 (54.1%)	< 0.001
Total occlusion (vs. stenosis), n (%)	543 (54.0%)	336 (48.8%)	207 (65.5%)	< 0.001
Moderate–severe calcification, n (%)	382 (38.0%)	228 (33.1%)	154 (48.7%)	< 0.001
Good runoff ≥ 2 vessels, n (%)	543 (54.0%)	413 (59.9%)	130 (41.1%)	< 0.001
Stent implantation, n (%)	584 (58.1%)	376 (54.6%)	208 (65.8%)	0.001
Drug-coated device used (DCB/DES), n (%)	360 (35.8%)	270 (39.2%)	90 (28.5%)	0.001
Residual stenosis > 20%, n (%)	215 (21.4%)	101 (14.7%)	114 (36.1%)	< 0.001
Technical success, n (%)	966 (96.1%)	671 (97.4%)	295 (93.4%)	< 0.001
Periprocedural complications, n (%)	62 (6.2%)	36 (5.2%)	26 (8.2%)	0.078

Cumulative incidence and univariable predictors of TLRS

In univariable Fine–Gray models (Table 3), higher AIP and higher ln(SII) were associated with an increased risk of TLRS. Other univariable predictors included GLASS stage III, lesion length ≥ 150 mm, total occlusion, moderate-to-severe calcification, residual stenosis > 20%, chronic kidney disease, diabetes, and higher CRP. In contrast, good runoff and use of drug-coated devices were associated with lower TLRS risk. Discharge treatment with cilostazol and low-dose rivaroxaban plus aspirin was also associated with lower TLRS risk. After adjustment for Core covariates, including device use, cilostazol (adjusted sHR 0.66, 95% CI 0.48–0.91; p = 0.011) and rivaroxaban 2.5 mg plus aspirin (adjusted sHR 0.58, 95% CI 0.37–0.90; p = 0.015) remained directionally associated with lower TLRS, but these exploratory estimates were not interpreted as causally because residual confounding by indication and treatment selection could not be excluded. In additional exploratory medication analyses, statin therapy was associated with lower TLRS risk in univariable analysis (sHR 0.62, 95% CI 0.44–0.87; p = 0.006), but the adjusted association was attenuated (sHR 0.81, 95% CI 0.62–1.07; p = 0.140); high-intensity statin therapy compared with lower-intensity/no statin therapy (adjusted sHR 0.86, 95% CI 0.68–1.10; p = 0.230) and chronic anti-inflammatory therapy (adjusted sHR 1.07, 95% CI 0.71–1.61; p = 0.750) were not independently associated with TLRS.

Table 3.

Univariable Fine–Gray models for predictors of TLRS

Predictor	sHR (95% CI)	p value
Age (per 10 years)	1.18 (1.05–1.33)	0.006
Male sex	1.18 (0.95–1.47)	0.132
Diabetes mellitus	1.61 (1.29–2.02)	<0.001
Current smoker	1.32 (1.06–1.64)	0.012
Chronic kidney disease	1.54 (1.22–1.95)	<0.001
Coronary artery disease	1.27 (1.02–1.59)	0.033
Rutherford class 4–6 vs 1–3	1.78 (1.42–2.22)	<0.001
AIP (per 0.10 increase)	1.26 (1.16–1.37)	<0.001
TyG (per 1.0 increase)	1.12 (0.97–1.30)	0.122
log₂(CRP) (per doubling)	1.14 (1.04–1.25)	0.006
Homocysteine (per 5 µmol/L)	1.09 (1.01–1.18)	0.022
ln(SII) (per 1-unit increase)	1.41 (1.23–1.62)	<0.001
NLR (per unit)	1.08 (1.03–1.13)	0.001
GLASS II vs I	1.29 (1.00–1.66)	0.049
GLASS III vs I	1.92 (1.48–2.49)	<0.001
Lesion length ≥150 mm	1.88 (1.52–2.33)	<0.001
Total occlusion (vs stenosis)	1.37 (1.11–1.69)	0.003
Moderate–severe calcification	1.36 (1.11–1.66)	0.003
Good runoff ≥2 vs ≤1 vessel	0.67 (0.54–0.84)	<0.001
Stent implantation	1.28 (1.04–1.58)	0.021
Drug-coated device used	0.73 (0.58–0.92)	0.007
Residual stenosis >20%	2.04 (1.64–2.55)	<0.001
Cilostazol at discharge	0.51 (0.39–0.67)	<0.001
Rivaroxaban 2.5 mg + aspirin	0.47 (0.32–0.69)	<0.001
Statin therapy at discharge	0.62 (0.44–0.87)	0.006
Chronic anti-inflammatory therapy	1.04 (0.69–1.57)	0.855

Multivariable prognostic models and model performance

In multivariable Fine–Gray regression, the Core model that included clinical, anatomic, procedural, device-related, and inflammatory covariates identified several independent predictors of TLRS (Table 4). After adjustment for other factors, chronic kidney disease (sHR 1.32, 95% CI 1.05–1.66; p = 0.018), diabetes mellitus (sHR 1.32, 95% CI 1.05–1.66; p = 0.017), Rutherford class 4–6 versus 1–3 (sHR 1.39, 95% CI 1.11–1.74; p = 0.004), GLASS III versus I (sHR 1.63, 95% CI 1.25–2.12; p < 0.001), lesion length ≥ 150 mm (sHR 1.61, 95% CI 1.30–2.00; p < 0.001), and residual stenosis > 20% (sHR 1.87, 95% CI 1.49–2.35; p < 0.001) remained significantly associated with higher TLRS risk, while good distal runoff (≥ 2 versus ≤ 1 vessels) was associated with lower risk (sHR 0.74, 95% CI 0.59–0.93; p = 0.009), drug-coated device use was associated with lower TLRS risk (sHR 0.78, 95% CI 0.61–0.99; p = 0.043), and log₂-transformed CRP showed a modest association (sHR per doubling 1.10, 95% CI 1.01–1.21; p = 0.037). The optimism-corrected 24-month C-index of the Core model was 0.68 (95% CI 0.64–0.72), with a calibration slope of 0.94 (95% CI 0.86–1.03), intercept of 0.02 (− 0.03 to 0.07), and Brier score of 0.19, indicating acceptable discrimination and calibration (Table 5; Fig. 3). When AIP and log-transformed SII were added to form the Extended model, both remained independently associated with TLRS after adjustment for the Core predictors, with sHR 1.17 (95% CI 1.07–1.28; p = 0.001) per 0.10-unit increase in AIP and sHR 1.24 (95% CI 1.07–1.44; p = 0.005) per 1-unit increase in log-SII, while the association of log₂-CRP attenuated and lost statistical significance. Drug-coated device use remained directionally protective in the Extended model but was not statistically significant (sHR 0.80, 95% CI 0.63–1.02; p = 0.071). No statistically significant interaction was observed between drug-coated device use and AIP (p for interaction = 0.47) or log-SII (p for interaction = 0.39). The structural predictors retained similar effect sizes in the Extended model, with GLASS III, lesion length ≥ 150 mm, and residual stenosis > 20% remaining strongly associated and good runoff remaining protective (Table 4). The Extended model improved optimism-corrected discrimination to a C-index of 0.73 (95% CI 0.69–0.77), with a calibration slope of 0.98 (95% CI 0.90–1.06), intercept of 0.01 (− 0.04 to 0.06), and Brier score of 0.17, and the increase in C-statistic versus the Core model was 0.05 (95% CI 0.02–0.08; p = 0.002). The Extended model also yielded a significantly positive IDI (0.021, 95% CI 0.009–0.034; p = 0.001) and category-free NRI (0.17, 95% CI 0.08–0.26; p < 0.001), and its calibration plot showed close agreement between predicted and observed 24-month TLRS risk across quintiles (Fig. 3; Table 5). When lesions were stratified into tertiles of predicted risk by the Extended model, observed 24-month TLRS incidence increased stepwise from 16.9% (95% CI 12.5–21.3) in the low-risk group to 29.3% (95% CI 23.7–34.9) in the intermediate-risk group and 48.0% (95% CI 40.8–55.2) in the high-risk group (Fig. 4).

Table 4.

Multivariable Fine–Gray models and performance for TLRS

Predictor	Core model sHR (95% CI)	p value	Extended model sHR (95% CI)	p value
Age (per 10 years)	1.11 (0.99–1.25)	0.071	1.09 (0.96–1.23)	0.179
Diabetes mellitus	1.32 (1.05–1.66)	0.017	1.24 (0.98–1.57)	0.071
Chronic kidney disease	1.32 (1.05–1.66)	0.018	1.27 (1.01–1.61)	0.044
Rutherford class 4–6 vs 1–3	1.39 (1.11–1.74)	0.004	1.31 (1.04–1.64)	0.021
GLASS II vs I	1.18 (0.91–1.54)	0.217	1.15 (0.88–1.49)	0.303
GLASS III vs I	1.63 (1.25–2.12)	<0.001	1.54 (1.18–2.01)	0.001
Lesion length ≥150 mm	1.61 (1.30–2.00)	<0.001	1.52 (1.23–1.89)	<0.001
Good runoff ≥2 vs ≤1 vessel	0.74 (0.59–0.93)	0.009	0.77 (0.61–0.97)	0.025
Residual stenosis >20%	1.87 (1.49–2.35)	<0.001	1.80 (1.43–2.27)	<0.001
Stent implantation	1.19 (0.95–1.49)	0.130	1.17 (0.94–1.46)	0.162
Drug-coated device used (DCB/DES)	0.78 (0.61–0.99)	0.043	0.80 (0.63–1.02)	0.071
Log2(CRP) (per doubling)	1.10 (1.01–1.21)	0.037	1.07 (0.98–1.18)	0.137
AIP (per 0.10 increase)	–	–	1.17 (1.07–1.28)	0.001
ln(SII) (per 1-unit increase)	–	–	1.24 (1.07–1.44)	0.005

Table 5.

Model performance at 24 months

Metric	Core model	Extended model
Optimism-corrected C-index	0.68 (0.64–0.72)	0.73 (0.69–0.77)
Calibration slope	0.94 (0.86–1.03)	0.98 (0.90–1.06)
Calibration intercept	0.02 (−0.03 to 0.07)	0.01 (−0.04 to 0.06)
Observed/expected 24-month TLRS risk	1.02 (0.95–1.10)	1.01 (0.94–1.09)
Brier score	0.19	0.17
ΔC-statistic (Extended − Core)	0.05 (0.02–0.08), p=0.002	–
Integrated discrimination improvement	0.021 (0.009–0.034), p=0.001	–
Net reclassification improvement	0.17 (0.08–0.26), p<0.001	–

Fig. 3.

Model performance and clinical utility of the Extended model for TLRS. (A) Calibration plot at 24 months for the Extended model (Core model plus AIP and SII), showing close agreement between predicted and observed TLRS risk across quintiles (calibration slope 0.98). (B) Decision curve analysis comparing net benefit of the Extended and Core models against treat‑all and treat‑none strategies across threshold probabilities (shown range 0.10–0.30)

Fig. 4.

Risk stratification for TLRS. Observed 24‑month TLRS incidence across tertiles of predicted risk from the Extended model (Core model plus AIP and SII)

In exploratory time-window analyses, log-SII showed a numerically stronger association with early TLRS within 12 months (sHR 1.38, 95% CI 1.13–1.69; p = 0.002) than with later TLRS from > 12 to 24 months (sHR 1.11, 95% CI 0.89–1.40; p = 0.350), whereas AIP was modestly associated with early TLRS (sHR 1.11, 95% CI 1.00–1.24; p = 0.052) and more clearly associated with later TLRS (sHR 1.25, 95% CI 1.08–1.46; p = 0.003).

Discussion

In this real-world, lesion-level cohort of 1,005 endovascularly treated lower-extremity atherosclerotic lesions, we observed a substantial 24-month burden of target-lesion restenosis (TLRS), with a cumulative incidence of 31.4% and clinically driven target-lesion revascularization of 14.9% despite contemporary use of stents and drug-coated devices. Fine–Gray competing‑risk analyses identified lesion complexity and quality of the angiographic result as central determinants of TLRS, with higher GLASS stage, lesion length ≥ 150 mm and residual stenosis > 20% independently associated with increased risk, while good distal runoff was protective. After adjustment for these anatomic and procedural factors, the composite systemic indices AIP and SII remained robust independent predictors of TLRS, whereas the triglyceride–glucose (TyG) index did not add incremental prognostic value. Incorporating AIP and SII into an Extended model significantly improved performance compared with a Core model containing clinical, anatomic, procedural, device‑related variables and C‑reactive protein. The lower CDTLR incidence relative to TLRS is clinically plausible because TLRS was an imaging-confirmed anatomic endpoint detected by scheduled surveillance, whereas CDTLR required recurrent symptoms plus repeat revascularization; many infrapopliteal or distal restenoses were therefore managed conservatively when wounds had healed, symptoms were stable, distal targets were poor, comorbidity/bleeding risk was high, or death occurred before reintervention.

In our real‑world cohort of 1,005 lesions with lower-extremity ASO, the 24‑month cumulative incidence of TLRS was 31.4%, which is consistent with the 20–40% range reported for femoropopliteal restenosis and post‑EVT loss of patency in other lower-extremity peripheral artery disease (LEPAD) series [14]. In line with prior ASO/LEPAD reports, lesions that developed TLRS in our study were more likely to arise in patients with diabetes, chronic kidney disease, and active smoking, and to exhibit advanced Rutherford class, long lesions, total occlusions, poor runoff, and residual stenosis, features long recognized as adverse prognostic factors after femoropopliteal EVT [14]. Inflammatory markers have also been implicated that interleukin‑6 and high‑sensitivity CRP predicted six‑month restenosis after femoropopliteal stenting in a Chinese cohort [15], while an elevated NLR independently predicted superficial femoral artery restenosis and worse limb outcomes in lower-extremity ASO [16]. More recently, SII has been shown to predict restenosis after interventions for lower-extremity ASO and to correlate with the angiographic severity of lower-extremity arterial disease [17, 18]. However, these studies typically evaluated single biomarkers in isolation and did not integrate them with structured anatomic staging such as GLASS. In contrast, our Extended model embedded GLASS stage, lesion length, runoff, and residual stenosis together with composite systemic indices (AIP and SII) and demonstrated a meaningful improvement in discrimination and net benefit over a Core model that already contained CRP and conventional clinical predictors.

The fact that AIP and SII, but not the TyG index, enhanced prediction in our ASO cohort is biologically plausible. AIP, defined as log₁₀(triglycerides/HDL-cholesterol), is a well-validated proxy for atherogenic dyslipidemia and small, dense LDL particles, and has been associated with incident cardiovascular disease and adverse outcomes across multiple populations [24]. Observational work increasingly links higher AIP values to a greater burden of peripheral arterial disease, including lower-extremity PAD, supporting the interpretation that AIP reflects a lipoprotein milieu that both accelerates plaque progression and impairs microvascular perfusion in the distal limb [24]. In our model, AIP remained independently associated with TLRS after adjustment for diabetes, CRP and detailed lesion anatomy, suggesting that it captures residual atherogenic risk not fully reflected by traditional lipid parameters. SII, which combines neutrophil, platelet and lymphocyte counts, integrates inflammation, thrombosis and relative lymphopenia, and has been reported as an independent predictor of restenosis after lower-extremity ASO interventions and of disease severity in lower-extremity arterial disease [17, 18]. This composite nature likely makes SII more closely aligned with the biology of neointimal hyperplasia than single-cell indices such as NLR. By contrast, the TyG index is a robust surrogate of insulin resistance and has been associated with PAD complexity and with PAD risk in meta-analysis [25, 26], but it may represent an upstream metabolic signal that is partly mediated through downstream lipoprotein and inflammatory changes captured by AIP and SII. Once those more proximate pathways and detailed anatomic factors (GLASS stage III, lesion length ≥ 150 mm and residual stenosis > 20%) were accounted for, TyG did not add incremental information in our Extended model, consistent with the observation that long, complex lesions with suboptimal angiographic results and poor runoff are the dominant substrate for restenosis in femoropopliteal and infrapopliteal disease [14, 19]. The exploratory time-window pattern is consistent with the concept that SII may reflect a more immediate thrombo-inflammatory/neointimal response after EVT, whereas AIP may mark a more chronic atherogenic milieu contributing to later recurrent narrowing; this interpretation remains hypothesis-generating because biomarker levels and medications may change over follow-up. Although statin use and intensity were considered, dose selection was non-randomized and may have changed after discharge, so we did not treat statin intensity as a definitive causal modifier of the AIP/SII associations.

Clinically, these findings suggest that a simple lesion-level model combining GLASS stage, lesion morphology and composite systemic indices could support more individualized follow-up and therapy in lower-extremity ASO. Current Global Vascular Guidelines on chronic limb-threatening ischemia and the 2024 ACC/AHA Multisociety guideline for lower-extremity PAD emphasize structured surveillance after revascularization but do not provide a validated tool to stratify restenosis risk at the lesion level [2, 19]. Our Extended model could be used to prioritize high-risk lesions, those with GLASS III anatomy, long length, poor runoff, significant residual stenosis and elevated AIP/SII, for more intensive duplex ultrasound surveillance and closer symptom monitoring, while permitting less intensive imaging in truly low-risk lesions, thereby optimizing resource use. Because AIP and SII are modifiable to some extent through aggressive lipid-lowering, metabolic control and suppression of systemic inflammation, high-risk patients might also be targeted for intensified risk-factor modification (high-intensity statins, tighter glycemic control, structured smoking-cessation program) in line with guideline recommendations for PAD. Finally, our exploratory finding that cilostazol and low-dose rivaroxaban (2.5 mg twice daily) plus aspirin were associated with lower TLRS dovetails with randomized evidence that cilostazol improves femoropopliteal patency and reduces adverse limb events after endovascular therapy [27], and that rivaroxaban plus aspirin reduces major adverse cardiovascular and limb events after lower-extremity revascularization or in stable PAD, with an acceptable bleeding trade-off [28, 29]. Taken together, these data support using the Extended model not only to tailor surveillance but also to identify ASO patients in whom more aggressive pharmacotherapy and, in anatomically complex GLASS III disease, consideration of bypass rather than repeat EVT may offer the best balance of durability and risk. Because drug-coated technologies are intended to reduce neointimal proliferation, their uneven use could influence observed restenosis risk and the apparent performance of inflammatory biomarkers [8, 21]. In our adjusted and interaction analyses, inclusion of drug-coated device use did not materially change the independent associations of AIP and SII, but these findings should be interpreted as hypothesis-generating because device selection was not randomized.

Limitations

This study has several limitations. First, the retrospective single-center design may limit generalizability and permits residual confounding despite multivariable adjustment. Second, treatment strategy was heterogeneous and operator-selected; although drug-coated device use was included as a covariate and tested for interaction with AIP/SII, unmeasured selection factors and device-specific effects may remain. Third, surveillance during the COVID-19 period occasionally required delayed imaging or telephone follow-up, which could have affected event timing. Fourth, formal quality-of-life instruments were not routinely collected, so clinical improvement could only be summarized using Rutherford class. Fifth, patients with multiple treated lesions or bilateral disease were retained in the primary lesion-level analysis with patient-level clustering, and although a single-lesion sensitivity analysis yielded consistent AIP/SII estimates, residual patient-level confounding cannot be excluded. Sixth, discharge medication intensity and chronic anti-inflammatory therapy were abstracted from records but longitudinal dose changes and adherence were not fully captured.

Future work should focus on external validation and refinement of this Extended model in larger, multicenter cohorts that include diverse patterns of lower-extremity peripheral artery disease/arteriosclerosis obliterans, wider ranges of lesion complexity, and different device and pharmacotherapy strategies. Ideally, such validation would prospectively collect uniform duplex and cross-sectional imaging, harmonize follow-up schedules, and evaluate dynamic updating of risk over time as systemic risk factors and treatments change. In parallel, the model could be translated into practical clinical tools that estimates lesion-level 24-month TLRS risk from routinely available variables and suggests risk-tiered surveillance strategies. High-risk lesions could be flagged for intensified duplex ultrasound surveillance and more aggressive secondary prevention or antithrombotic regimens, whereas low-risk lesions might follow a less intensive imaging schedule with symptom-driven review, thereby optimizing resource use and patient burden.

Ultimately, our findings indicate that AIP and SII substantially improve lesion-level risk prediction for TLRS after EVT beyond conventional clinical, anatomic, and procedural factors, and that the Extended model may provide a useful framework for risk-tiered surveillance and more individualized management in patients with LEPAD/ASO, pending rigorous external validation.

Author contributions

The authors confirm contribution to the paper as follows: study conception and design: X. S.; data collection: H. X., X. L., Y. L.; analysis and interpretation of results: H. X., X. L., Y. L.; draft manuscript preparation: H. X., X. L., Y. L., X. S. All authors reviewed the results and approved the final version of the manuscript.

Funding

No funding.

Data availability

Data sets generated during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki. The protocol was reviewed and approved by the Ethics Committee of the Affiliated Hospital of Southwest Medical University, which waived the requirement for informed consent because this was a retrospective analysis of de-identified routinely collected data and posed no more than minimal risk to participants. All data were anonymized prior to analysis.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.