第一部分 腔内治疗主髂动脉闭塞症临床疗效的统计学分析和机器学习预后模型的构建

背景

近年来随着腔内血管外科技术和器械的发展，介入手术成为治疗主髂动脉闭塞症（Aortoiliac occlusive disease，AIOD）的重要方式，治疗理念和方法也有了极大的改变。覆膜支架（Covered stent，CS）与金属裸支架（Bare metal stent，BMS）是最常用的两种腔内手术治疗器械，但目前对其应用的具体场景以及相关的治疗安全性和有效性的对比性研究较少。另外，由于覆膜支架应用越来越多，真正能从CS中获益的患者特征以及中远期疗效仍然存在争议。

目的

回顾性比较单中心使用CS和BMS治疗AIOD的临床结果，采用倾向性评分匹配等统计学方法和机器学习的方法进行分析，评价治疗的安全性和有效性，对比获益的患者特征以及中远期临床疗效。

方法

连续性纳入2016年1月至2022年10月在单中心接受血管腔内治疗的所有AIOD患者，包括根据2007年第二版跨大西洋社会共识（Trans-Atlantic Inter-Society Consensus，TASC-II）制定的所有类型病变患者。收集相关的临床资料和基线数据，分为应用倾向性评分匹配，比较CS和BMS在基线特征、手术因素、30天结局、5年原发性通畅和肢体保留方面的差异。随访结果采用Kaplan-Meier曲线分析比较两组的结局事件发生率。Cox比例风险模型用于确定原发性通畅的独立危险因素预测因素，并对特殊病变进行亚组分析。

将再狭窄患者随机拆分为机器学习预测模型训练集和验证集，采用逻辑回归、随机森林、决策树等常见机器学习算法训练模型，采用ROC曲线预测模型效能。通过Kaplan-Meier生存分析比较再狭窄高危组和低危组患者的一期通畅率。

结果

共209名通过腔内治疗的主髂动脉闭塞症患者纳入研究，CS组135（64.6%）例，BMS组74（35.4%）例， CS组患者的术前ABI更低（0.48±0.26 vs 0.52±0.19；p=0.032），包含更多的复杂主髂动脉病变（TASC D级）患者（47.4% vs 9.5%；p＜0.001），更多的CTO病变（77.0% vs 31.1%；p＜0.001），更多的重度钙化（20.7% vs 14.9%；p＜0.036）。

经过倾向性评分匹配后，共匹配50例患者（25例CS，25例BMS），仅重度钙化和ABI增加在临床数据方面仍具有统计学意义（32.0% vs 8.0%，p=0.034）。手术因素方面，CS组患者更多应用双侧股动脉或联合肱动脉的多部位穿刺入路（60.0% vs 12.0%，p＜0.001），更多的术中支架植入数量（2.3±1.2 vs 1.3±0.7，p＜0.001），更长支架平均长度（9.3±3.3 vs 5.8±2.6，p＜0.001），更多CDT治疗策略（32.0% vs 4.0%，p=0.009）。

两组患者的手术成功率（100% vs 100%，p=1.00），早期（＜30天）死亡率（0% vs 0%，p=1.00）以及5年的初级通畅率（83.4% vs 86.9%，p=0.330）、次级通畅率（96% vs 100%，p=0.570）、肢体保留率（100% vs 100%，p=1.00）均无明显统计学差异。总体手术并发症发生率基本相同（12.0% vs 8.0%，p=0.891），只有ABI增加值（0.45±0.15 vs 0.41±0.22，p=0.038）在两组间有统计学差异。多因素Cox生存分析显示，重度钙化（HR, 1.32; 95% CI, 1.04-1.85; P=0.048）是原发通畅率的唯一独立预测因子。

Adaboost机器学习模型表现最好，训练集及验证集准确率分别为95%、90%，ROC曲线下面积0.96，灵敏度92%，特异度88%。高危组患者术后一期通畅率为51.9%，低危组患者术后一期通畅率为87.0%（p=0.045）。                                                           

结论

在本研究中，BMS和CS对治疗AIOD的有效性相当且均获得了良好的中远期结果，通过倾向性评分匹配，最大程度降低了混杂因素的影响，在未匹配和匹配队列中，CS组和BMS组的一期通畅率总体上相似。CS组患者术后血流动力学改善更加明显，推荐优先用于更复杂的病变，特别是对于重度钙化病变，CS组表现出明显的优势，是影响预后的独立危险因素。同时机器学习预测模型对腔内治疗AIOD术后再狭窄有良好的预测效能。

第二部分 基于信息化平台的社区脑卒中高危人群现况调查及颈动脉狭窄风险因素分析

背景

颈动脉狭窄(Carotid artery stenosis, CAS)是严重危及生命健康的重要疾病，是引起脑卒中的主要原因之一，而脑卒中是我国老年人致死、致残的首位原因。因此早期诊断和治疗CAS对防治脑卒中具有重要的意义。

目的

通过搭建信息化社区居民病例管理平台，了解我国社区居民脑卒中高危人群的流行病学现状，筛选社区颈动脉粥样硬化进展的风险因素，实现对脑卒中的早期干预及全流程信息化管理。

方法

开发构建信息化社区居民健康管理平台，选取2019年1月至2022年12月于北京市西城区西长安街社区卫生服务中心以及北京市第二医院门诊就诊的老年人群，调取基本信息及颈动脉超声检查结果，建立回顾性研究队列并纳入信息化健康管理平台，通过简单随机抽样筛选200例患者进行颈动脉超声复查，对病变进展较快的人群进行单因素及多因素回归分析，筛选出颈动脉斑块进展的高危因素。

结果

成功搭建老年缺血性脑卒中高危人群的识别与全病程健康管理信息化平台，依托平台筛查符合纳入标准的社区人群共3930人，根据中国脑卒中防治指导规范进行脑卒中风险分级，共筛选出脑卒中高危人群1159例（29.5%），中危人群2211例（56.3%），低危人群560例（14.2%）。通过随机抽样抽取样本200例进行超声复查，其中病情进展135人（67.5%），正常65人（32.5%）。单因素分析结果显示，高龄、男性、高血压、糖尿病、外周动脉疾病史以及钙化斑块是病情进展的危险因素，规律的体育锻炼是病情进展的保护因素。多因素回归分析结果显示，年龄（OR＝1.07，95%CI：1.03-1.11，P＜0.01）、男性性别（OR＝3.09，95%CI：1.17-8.12，P＝0.02）、钙化斑块（OR＝4.89，95%CI：1.51-15.83，P＝0.01）是病变进展的独立危险因素（P＜0.05）。

结论

通过搭建信息化病例管理平台，筛查出社区居民脑卒中发病风险较高，而对社区中老年居民进行常规颈动脉超声筛查可以发现钙化斑块等颈动脉狭窄进展的高危因素，通过早期干预可以有效降低脑卒中发生率，具有较高的慢病预防及管理的参考意义。

Part I: Clinical Efficacy of Endovascular Treatment for Aortoiliac Occlusive Disease using Statistical Analysis and Construction of Machine Learning Prognostic Model

Background

Over the past decade, with the development of technology and the improvement of materials, endovascular treatment of aortoiliac disease has become less inferior to traditional open surgery and has provided satisfactory long-term outcomes with low rates of early complications. Covered stents (CS) and bare metal stents (BMS) are the two most commonly used, but few studies have compared the specific scenarios and the associated therapeutic safety and efficacy. In addition, due to the increasing application of CS, the characteristics of patients who can really benefit from CS and the mid-to-long term efficacy are still controversial.

Objective

We retrospectively compared the clinical outcomes of self-expanding covered stents (SECSs) and bare metal stents (BMSs) in the treatment of aortoiliac occlusive disease (AIOD) at a single center between 2016 and 2022.

Methods

All AIOD patients receiving endovascular therapy at a single center from January 2016 to October 2022 were continuously analyzed, including patients with lesions of all classes according to the Trans-Atlantic Inter-Society Consensus II (TASC-II). Relevant clinical and baseline data were collected, and propensity score matching was performed to compare CSs and BMSs in terms of baseline characteristics, surgical factors, 30-day outcomes, 5-year primary patency and limb salvage. The follow-up results were analyzed by Kaplan‒Meier curves. Cox proportional hazard models were used to identify independent risk factor predictors of primary patency and to perform subgroup analyses for specific lesions.

Patients were divided into the training set and validation set of machine learning prediction model. Common machine learning algorithms such as logistic regression, random forest, decision tree and other common machine learning algorithms were used to train the model, and ROC curve were used to predict the model efficiency. Patients were divided into high-risk and low-risk restenosis groups using the best-performing model, and the primary patency rates of the two groups were compared by Kaplan-Meier survival analysis.

Results:

A total of 209 patients with AIOD were enrolled in the study, including 135 (64.6%) patients in the CS group and 74 (35.4%) patients in the BMS group. Surgical success rates (100% vs. 100%, p=1.00), early (< 30-day) mortality rates (0% vs. 0%, p=1.00), cumulative surgical complication rate (12.0% vs. 8.0%, p=0.891), 5-year primary patency rate (83.4% vs. 86.9%, p=0.330), secondary patency rate (96% vs. 100%, p=0.570) and limb salvage rate (100% vs. 100%, p=1.00) did not exhibit significant differences between the two groups. Patients in the CS group had a lower preoperative ankle-brachial index (ABI) (0.48±0.26 vs. 0.52±0.19; p=0.032), more cases of complex AIOD (especially TASC D) (47.4% vs. 9.5%; p < 0.001), more chronic total occlusive (CTO) lesions (77.0% vs. 31.1%; p < 0.001) and more severe calcification (20.7% vs. 14.9%; p < 0.036).

After propensity score matching, 50 patients (25 CSs and 25 BMSs) were selected. The results showed that only severe calcification (32.0% vs. 8.0%, p=0.034) and ABI increase (0.45±0.15 vs. 0.41±0.22, p=0.038) were significantly different between the groups. In terms of surgical factors, patients in the CS group had more use of bilateral femoral or combined brachial artery percutaneous access (60.0% vs. 12.0%, p < 0.001), more number of stents used (2.3±1.2 vs. 1.3±0.7, p < 0.001), longer mean stent length (9.3±3.3 vs. 5.8±2.6, p < 0.001) and more catheter-directed thrombolysis (CDT) treatment (32.0% vs. 4.0%, p=0.009). Multivariate Cox survival analysis showed that severe calcification (HR, 1.32; 95% CI, 1.04-1.85; P=0.048) was the only independent predictor of the primary patency rate.

Adaboost machine learning model has the best performance, the accuracy of training set and verification set are 95% and 90%, respectively. The area under ROC curve is 0.96, the sensitivity is 92% and the specificity is 88%. The primary patency rate of patients classified by machine learning model was 51.9% in high-risk group and 87.0% in low-risk group (p=0.045).

Conclusions

All patients with AIOD who underwent endovascular therapy were included and achieved good  outcomes with both CSs and BMSs. The influence of confounding factors in the two groups was minimized by propensity score matching, and the five-year patency rates were generally similar in the unmatched and matched cohorts. Postoperative hemodynamic improvement was more obvious in patients in the CS group.For more complex lesions, CS is recommended to be preferred.  Especially for severe calcification lesions, which is the only independent predictor of primary patency, CS showed obvious advantages.Further studies with more samples are needed to investigate the role of stent types in AIOD treatment. At the same time, the machine learning prediction model has good predictive efficacy for postoperative restenosis after endovascular treatment of AIOD.

Part II: Investigation of community stroke high-risk population and analysis of risk factors for carotid artery stenosis based on information platform

Background

Carotid artery stenosis(CAS) seriously endangers life and health, and is one of the main causes of stroke which is the first cause of death and disability in the elderly in China. Seriously affects the health of residents and also brings heavy economic burden to the country. Therefore, the early diagnosis and treatment of CAS is of great significance for the prevention and treatment of stroke.

Objective

To understand the epidemiological status of carotid artery disease and stroke risk of Chinese community residents through the information management platform, realize the whole-process information management of stroke.

Method

Select elderly people who visited the Community Health Service Center of West Chang 'an Street in Xicheng District of Beijing from January 2019 to December 2022, collect basic information and ultrasound results, establish a retrospective study cohort and information-based health management platform. Carotid artery ultrasound review was performed on 200 patients randomly selected, univariate and multivariate regression analysis was performed on the population with lesion progression to screen out the high-risk factors.

Result

Successfully built an information platform for the identification of high-risk groups of ischemic stroke in the elderly and the health management of the whole course of the disease. Relying on the platform, a total of 3930 community people meeting the inclusion criteria were screened. 1159 patients (29.5%) were high risk, 2211 patients (56.3%) were middle risk and 560 patients (14.2%) with low risk. 200 samples were selected by random sampling for ultrasound review, among which 135 patients (67.5%) developed disease and 65 patients (32.5%) were normal. Univariate analysis showed that old age, male, hypertension, diabetes, history of peripheral artery disease and calcified plaque were risk factors for disease progression, and regular physical exercise was protective factor. Multivariate regression analysis showed that old age (OR=1.07, 95%CI: 1.03-1.11, P<0.01), male sex (OR=3.09, 95%CI: 1.17-8.12, P=0.02), calcified plaque (OR=4.89, 95%CI: 1.51-15.83, P=0.01) were independent risk factors for lesion progression (P < 0.05).

Conclusion

Community residents have a high risk of carotid atherosclerosis and stroke. Routine carotid ultrasound screening for middle-aged and elderly residents in community can provide early intervention, effectively reduce the incidence of stroke, and has a high reference value for chronic disease prevention and management.