Abstract
目的
腹主动脉瘤是指腹主动脉扩张超过3.0 cm的一种病理状态,手术治疗方式包括开放手术(open surgical repair,OSR)和腔内修复(endovascular aneurysm repair,EVAR)。预测腹主动脉瘤患者OSR后急性肾损伤(acute kidney injury,AKI)的发生有助于术后临床决策。为找到一种更有效的预测方法,本研究对不同机器学习预测模型的效能进行测试。
方法
回顾性收集2009年1月至2021年12月中南大学湘雅医院80例OSR患者的围手术期数据,手术均由血管外科医师实施。选择logistic回归、线性核支持向量机、高斯核支持向量机、随机森林4种常用的机器学习分类模型来实施预测。通过五重交叉验证来分析模型的性能。
结果
33名患者出现AKI,4种分类模型的五重交叉验证结果表明:随机森林是预测AKI最精确的模型,曲线下面积为0.90±0.12。
结论
机器学习模型可以精确预测术后AKI的发生,从而使血管外科医生能更早地处理并发症,并且可能有助于提高腹主动脉瘤OSR的临床疗效。
Keywords: 腹主动脉瘤, 急性肾损伤, 机器学习, 随机森林, 支持向量机
Abstract
Objective
Abdominal aortic aneurysm is a pathological condition in which the abdominal aorta is dilated beyond 3.0 cm. The surgical options include open surgical repair (OSR) and endovascular aneurysm repair (EVAR). Prediction of acute kidney injury (AKI) after OSR is helpful for decision‐making during the postoperative phase. To find a more efficient method for making a prediction, this study aims to perform tests on the efficacy of different machine learning models.
Methods
Perioperative data of 80 OSR patients were retrospectively collected from January 2009 to December 2021 at Xiangya Hospital, Central South University. The vascular surgeon performed the surgical operation. Four commonly used machine learning classification models (logistic regression, linear kernel support vector machine, Gaussian kernel support vector machine, and random forest) were chosen to predict AKI. The efficacy of the models was validated by five‐fold cross‐validation.
Results
AKI was identified in 33 patients. Five‐fold cross‐validation showed that among the 4 classification models, random forest was the most precise model for predicting AKI, with an area under the curve of 0.90±0.12.
Conclusion
Machine learning models can precisely predict AKI during early stages after surgery, which allows vascular surgeons to address complications earlier and may help improve the clinical outcomes of OSR.
Keywords: abdominal aortic aneurysm, acute kidney injury, machine learning, random forest, support vector machine
腹主动脉瘤(abdominal aortic aneurysm,AAA)多发于中老年人,如果不及时治疗,一旦破裂,病死率高,为78%~94%[1]。AAA的治疗方法主要是手术,手术方式有开放手术(open surgical repair,OSR)和腔内修复(endovascular aneurysm repair,EVAR)2种。美国医保患者数据[2]显示OSR在10年内下降了76%,目前,大约80%的AAA患者接受EVAR,并且随着腔内技术的提高和器材的改进,这一数字仍有继续上升的趋势[3]。研究[4-8]发现:与OSR相比,EVAR降低了AAA手术患者30 d内病死率,但是随着回访时间的延长,EVAR的生存率优势逐渐减弱,在术后3至4年总病死率相似。因此,OSR治疗AAA的重要性还是不容忽视[9]。此外,研究[10]显示近肾AAA采用OSR有出色的解剖耐久性。由于移植物和相应辅助装置的前期成本,胸腹主动脉瘤EVAR比OSR的早期成本更高[11],临床实践中受限于早期费用的患者更可能行OSR。
AAA患者OSR围手术期的并发症主要包括心肌梗死、肺炎、肾功能不全、出血、伤口感染等[3]。一项前瞻性多中心研究[12]和一项回顾性队列研究[13]发现:急性肾损伤(acute kidney injury,AKI)是AAA患者OSR后的常见并发症,发生率为37%~39%。术后发生AKI使肾下型AAA修复后的各种不良事件显著增加,如增加住院时间、ICU入住的可能及入住时间,甚至会显著增加死亡风险[14]。构建OSR术后发生AKI的预测模型有助于临床决策,多项研究[15-17]使用一些特殊的生物标志物作为特征,这样构建的模型难以运用于实际。目前,已有研究[18]对胸腹主动脉瘤开放修复术术后AKI的发生建立了机器学习预测模型。但是还需要专门针对AAA接受OSR的患者建立一种有效的方法来预测术后AKI的发生,以便决定手术方式和预防相关并发症。
近年来,基于人工智能的算法成为世界范围内的前沿技术。它们在医疗场景中的应用显示了无限的潜力。相比于需要海量数据的深度学习算法,一些机器学习算法,如支持向量机和随机森林,可能是医学领域中最有效和最稳健的算法[19]。机器学习算法利用特征之间复杂微妙的关系,已用于预测AAA的生长[20-21]、破裂风险[22-23]、手术时机和术后不良事件的发生[24-26]。本研究旨在建立一个精确预测AAA患者OSR术后AKI发生的模型,以便在术后实施预测并指导治疗决策。
1. 资料与方法
1.1. 收集资料
本研究已获得中南大学湘雅医院(以下简称我院)伦理审查委员会批准(审批号:202203079)。连续收集2009年1月至2021年12月于我院行OSR的80例AAA患者,根据既定方案培训一位研究者从我院电子病历系统中按照预先设计的Excel表格收集数据,此研究者预先不知晓研究内容。
基于EIFLE分类标准[27],Twine等[28]提出了关于AAA修复后AKI发生的标准定义,并建议将其用于EVAR和OSR。本研究AKI的定义为动脉瘤肾损伤评分(aneurysm renal injury score,ARISe)≥1,具体如下:1)血肌酐上升,超过基线值26 μmoI/L或术后7 d内尿量<0.5 mL/(kg·h)的时间>6 h;2)血肌酐超过基线值的50%;3)需要间断性或持续性血液透析。
纳入标准:1)主要术式为AAA切除加人工血管置换术的住院患者;2)肾上、肾旁或近肾型AAA患者。排除标准:1)排除破裂及先兆破裂的AAA患者;2)术前存在肾功能不全需要肾替代治疗的患者;3)存在其他严重合并症的患者,如术后急性心力衰竭;4)术后24 h内死亡的患者。
1.2. 开放手术
本中心实施标准的AAA切除加人工血管置换术,手术采用经腹膜或左侧腹膜后入路,术中使用标准剂量的普通肝素实现全身肝素化。理想的主动脉钳夹位置和重建范围是基于术中具体情况及主动脉CT血管造影分析确定的,需要考虑的因素包括:近端动脉瘤;髂动脉闭塞性疾病或动脉瘤性疾病;伴发的肾和肠系膜疾病;静脉解剖异常;钙化灶、血栓或动脉粥样硬化碎片。保留性腺、肾上腺和腰动脉以提供肾的侧支血流,在主动脉颈部存在明显附壁血栓时,需要暂时钳夹肾动脉以减小肾动脉栓塞的风险。手术用到的移植物是法国Intervascular S.A分叉型人工血管,依据术中情况裁剪,远端吻合的位置在主动脉分叉处、髂动脉或者股动脉。手术结束时使用鱼精蛋白对抗普通肝素。
1.3. 统计学处理与机器学习算法
使用计算机编程语言(Python V3.8.8; R V4.1.2)和统计分析软件(SPSS R23.0.0.0)处理和分析数据。为了达到生成最优化算法模型的目的,必须增强对数据的理解,包括对维度、属性、类型、数据描述性统计、数据分布以及数据属性间相关性的了解。如果某个特征的数据范围很大,就很难较好地收敛到一个机器学习算法中,从而降低预测模型的精度。对于这种情况,本研究使用MinMaxScaler函数,转换公式为X_scaled=(X-X.min)/(X.max-X.min),将数据缩放到0~1。
回顾AKI发生有关风险因素的文献[29-31],本研究选择了围手术期的临床数据作为预测模型的特征,包括年龄、性别、体重、BMI、有无肾动脉夹层、有无肾动脉狭窄、术前血肌酐、术中平均动脉压、术中出血量、术中输注的红细胞量、术中输注血浆量、术中输注的血小板量、自体输血量、手术时间,术中尿量、术中是否钳夹肾动脉、肾动脉钳夹时间。其中双肾动脉夹层与狭窄可通过CT血管造影来确定,术前血肌酐为入院时的血肌酐值,术中平均动脉压为手术结束时的平均动脉压,术中尿量定义为每千克体重每小时的尿量。
经过数据缩放和特征选择后,使用4个机器学习模型(logistic回归、线性核支持向量机、高斯核支持向量机和随机森林)来完成预测。Logistic回归使用logistic函数在(0, 1)范围内收敛线性函数。线性支持向量机通过在n维空间中寻找具有最大边界的超平面对数据进行明确的分类。如果特征之间的关系是非线性的,logistic回归和线性核支持向量机模型可能表现不佳。因此,本研究使用一个径向基函数核映射数据到一个更高维度空间,然后测试支持向量机分类器的效能。此外,单个决策树分类器基于一系列分裂规则确定结果,这些规则从树的顶部开始,并持续到一系列分支,对特征空间进行分层以提供预测。单一的决策树模型容易过拟合,缺乏鲁棒性。这可以通过聚合多个决策树的随机森林来解决,随机森林算法关注的是森林中大多数决策树的判断。
使用五重交叉验证来分析模型的性能,其基本思想是将原始数据分为5组,其中4组作为训练数据集,1组作为评估数据集,重复5次,评估数据集就有5种。通过4组训练数据集训练分类器,然后利用评估数据集来评价模型的性能,并绘制每一个模型的受试者操作特征(receiver operator characteristic,ROC)曲线。
符合正态分布的连续性变量用均数±标准差( ±s)描述,离散型变量用例数和频率(%)描述。根据适用标准使用双尾t检验、χ2检验、Fisher精确检验、连续校正χ2检验,单因素分析和多因素分析采用logistic回归。将单因素分析中P<0.05的变量进一步纳入多因素logistic回归。为排除共线性影响,本研究根据已有的临床经验及相关性分析结果,手动筛选多因素logistic回归中的变量。P<0.05为差异有统计学意义。
2. 结 果
2.1. 单因素分析
80例患者中的33例发生AKI,各组人口统计学数据见表1。其中1例患者有肾动脉夹层,发生急性肾功能不全;12例患者有肾动脉狭窄。4例患者在手术过程中钳夹至少一侧肾动脉,其中3例发生AKI,1例未发生AKI,术前肾功能均未发现异常。发生AKI组和未发生AKI组各变量比较,结果显示:年龄(P<0.05)、性别(P<0.05)、术前肾动脉狭窄(P<0.05)、术前血肌酐(P<0.001)、术中出血量(P<0.05)、术中输注红细胞量(P<0.05)、术中输注血浆量(P<0.001)、术中输注血小板量(P<0.05)、自体输血量(P<0.05)、手术时间(P<0.05)差异均具有统计学意义(表1)。
表1.
行开放手术的腹主动脉瘤患者的人口统计学资料
Table 1 Demographics of patients undergoing open surgical repair of abdominal aortic aneurysms
| Groups | n | Age/year | Female/[No. (%)] | Weight/kg | BMI/(kg·m-2) | RAD/[No. (%)] |
|---|---|---|---|---|---|---|
| P | 0.026 | 0.026 | 0.441 | 0.972 | 1.000 | |
| All | 80 | 63.98±9.11 | 20(25.00) | 60.46±9.30 | 21.44±2.85 | 1(1.25) |
| Non-AKI | 47 | 62.09±7.90 | 16(34.04) | 59.79±9.96 | 21.43±2.87 | 1(2.13) |
| AKI | 33 | 66.67±10.12 | 4(12.12) | 61.42±8.32 | 21.45±2.86 | 0(0) |
| Groups | RAS/[No. (%)] | Cr/(μmoI·L-1) | MAP/mmHg | Bleeding/mL | RBCs/Unit | Plasma/mL |
|---|---|---|---|---|---|---|
| P | 0.024 | <0.001 | 0.840 | 0.003 | 0.001 | <0.001 |
| All | 12(15.00) | 99.53±45.85 | 88.11±11.92 | 1 472.50±1 607.60 | 2.21±3.32 | 222.75±320.66 |
| Non-AKI | 3(6.38) | 84.62±26.15 | 88.34±12.65 | 978.72±931.15 | 1.12±2.28 | 97.66±181.41 |
| AKI | 9(27.27) | 120.75±58.44 | 87.79±10.98 | 2 175.76±2 065.34 | 3.77±3.95 | 400.91±388.36 |
| Groups | Platelets/Unit | AV/mL | OT/min | UV/(mL·h-1·kg-1) | SRAC/[No. (%)] | RACT/min |
|---|---|---|---|---|---|---|
| P | 0.012 | 0.018 | 0.044 | 0.169 | 0.376 | 0.310 |
| All | 0.08±0.27 | 802.40±916.17 | 244.59±79.62 | 2.49±0.11 | 4(5.00) | 1.675±7.47 |
| Non-AKI | 0 | 585.11±715.35 | 228.77±67.76 | 2.63±1.00 | 1(2.13) | 0.91±6.27 |
| AKI | 0.18±0.39 | 1 111.88±1 080.94 | 267.12±90.33 | 2.28±1.23 | 3(9.09) | 2.76±8.89 |
AKI: Acute kidney injury; RAD: Renal artery dissection; RAS: Renal artery stenosis; MAP: Mean arterial pressure; AV: Autotransfusion volume; OT: Operation time; UV: Urine volume; SRAC: Supra-renal artery clamping; RACT: Renal artery clamp time.
2.2. 多因素分析
在进行多变量logistic回归之前,使用递归特征消除法得到对AKI影响最大的4个特征:性别、肾动脉狭窄、术中输注血浆量、术前血肌酐。随后进行多变量logistic回归,结果发现男性(OR=8.003,95% CI:1.094~58.541;P<0.05)、肾动脉狭窄(OR=16.099,95% CI:1.994~129.971;P<0.05)、术前血肌酐(OR=1.032,95% CI:1.002~1.062;P<0.05)以及术中输注血浆量(OR=1.005,95% CI:1.001~1.009;P<0.05)是术后AKI发生的独立危险因素(表2、图1)。
表2.
Logistic回归分析术后急性肾损伤的独立危险因素
Table 2 Independent risk factors for postoperative acute kidney injury by logistic regression analysis
| Variables | β | SE | OR(95% CI) | P |
|---|---|---|---|---|
| Intercept | -14.041 | 4.795 | <0.001 | 0.003 |
| Age | 0.090 | 0.049 | 1.094(0.994 to 1.204) | 0.067 |
| Sex | 2.080 | 1.015 | 8.003(1.094 to 58.541) | 0.041 |
| RAS | 2.779 | 1.066 | 16.099(1.994 to 129.971) | 0.009 |
| Cr | 0.031 | 0.015 | 1.032(1.002 to 1.062) | 0.028 |
| Bleeding | <0.001 | <0.001 | 1.000(1.000 to 1.001) | 0.257 |
| RBCs | -0.073 | 0.147 | 0.930(0.697 to 1.240) | 0.620 |
| Plasma | 0.005 | 0.002 | 1.005(1.001 to 1.009) | 0.015 |
| OT | 0.006 | 0.005 | 1.006(0.996 to 1.016) | 0.253 |
RAS: Renal artery stenosis; OT: Operation time.
图1.
多变量逻辑回归的森林图
Figure 1 Forest plot of multivariate logistic regression
RAS: Renal artery stenosis.
2.3. 机器学习算法建立模型
输入各项特征值,利用不同的机器学习算法建立模型,结果显示基于随机森林算法的模型是最有效的(表3、图2),其各项性能评价指标:AUC为0.90±0.12(95% CI:0.75~1.00),对数损失函数值为-0.46±0.12(95% CI:-0.31~-0.63),分类准确度为0.80±0.07(95% CI: 0.71~0.89),精确率为0.89±0.16(95% CI:0.69~1.00),召回率为0.68±0.23(95% CI:0.40~0.96)。进一步分析得到随机森林模型中每个特征的重要性百分比(图3),重要性最高的7个特征总计占68.92%,其中术前血肌酐为21.22%,术中出血量为10.72%,术中尿量为9.56%,手术时间为7.40%,术中自体输血量为7.12%,手术平均动脉压为6.64%,体重为6.26%。
表3.
4种机器学习模型性能( ±s)
Table 3 Performance comparison of 4 machine learning models ( ±s)
| Models | Accuracy | Logloss | Precision | Recall | AUC |
|---|---|---|---|---|---|
| Logistic regression | 0.71±0.10 | -1.04±0.43 | 0.73±0.17 | 0.55±0.20 | 0.76±0.07 |
| Linear SVM | 0.78±0.12 | -0.54±0.08 | 0.86±0.17 | 0.62±0.26 | 0.85±0.09 |
| Gaussian SVM | 0.73±0.12 | -0.58±0.10 | 0.66±0.11 | 0.69±0.28 | 0.77±0.12 |
| Random forest | 0.80±0.07 | -0.46±0.12 | 0.89±0.16 | 0.68±0.23 | 0.90±0.12 |
SVM: Support vector machine; AUC: Area under the curve.
图2.
急性肾损伤预测模型
Figure 2 Acute kidney injury (AKI) prediction model
A: Receiver operator characteristic (ROC) curve of logistic regression to predict AKI; B: ROC curve of linear support vector machine (SVM) to predict AKI; C: ROC curve of Gaussian SVM to predict AKI; D: ROC curve of random forest to predict AKI. The black solid curve is the average curve of 5 curves, and the shade is the average curve ±1 standard deviation.
图3.
随机森林模型的特征贡献
Figure 3 Attribution of the factors to the random forest model
RAD: Renal artery dissection; RAS: Renal artery stenosis; MAP: Mean arterial pressure; AV: Autotransfusion volume; OT: Operation time; UV: Urine volume; RACT: Renal artery clamp time; SRAC: Supra-renal artery clamping.
3. 讨 论
本研究基于机器学习的方法,对数据进行收集与整理,选择模型,优化模型,使用五重交叉验证和ROC曲线等对模型进行评估,结果表明利用简单的临床数据即可以构建较好的模型,其中随机森林是预测AKI的最佳模型。此外,本研究也对AKI发生的风险因素进行了分析。研究[14, 32-33]表明需要肾上钳夹主动脉的患者有增加肾功能不全的风险,但是与肾下钳夹修复AAA的患者相比,30 d病死率相似。尽管如此,还是应努力将近端钳夹位置保持在尽可能低的水平,尤其是术前肾功能不佳的患者。一项单中心观察性队列研究[34]证明肾冷灌注降低选择性开放修复近肾动脉瘤后AKI的发生率,还有观点认为术中使用甘露醇能够减少OSR术后肾功能不全的发生[13]。这些研究给预防术后并发症提供了策略。通过长期实践,本中心总结了一个在手术前评估患者并尽量减少并发症的总体策略:1)改善术前肾功能;2)术中保持器官灌注;3)缩短器官缺血时间;4)建立适当有效的监测系统。
本研究中用到的机器学习是常用的监督学习技术,与传统统计学方法相比,最大的优势在于它可以充分利用特征之间的复杂关系[35]。Manuel等[19]研究表明:在数百种分类器中,随机森林、支持向量机(包括线性核和高斯核)和神经网络分类器的分类效果优于其他分类器。由于医学研究的样本容量大多处于较低水平,常常不能满足神经网络模型的要求。因此,本研究选择了随机森林、支持向量机(线性核和高斯核) 和logistic回归来完成这个二分类任务。样本中AKI发生与否的比例接近1,这对建立模型来说是较理想的。
对特征的选取,不同的研究各不相同。本研究根据以往的临床知识,选择年龄、性别、术前肾动脉狭窄等17个特征来预测AKI。一项关注胸腹主动脉瘤开放修复术的研究[18]表明:相比于单纯的AAA OSR,其更具复杂性,选取的特征也涉及更多方面。但标签的定义没有对AKI的程度进行分类,运用这个模型可能会导致更加积极的临床干预。也有研究[24]结合了人口特征、瘤体几何特征、血流动力学特征来预测AAA的破裂风险。未来可能需要结合不同维度的特征,如基于影像学特征或文本特征等来构建更可靠的模型。
本研究前提假设为样本是独立选取的,并且与静态总体数据分布相同,样本是总体的代表。本研究的大部分样本来自中南地区,为随机抽取的样本。笔者不认为本研究的模型在其他地区也适用,主要因为模型的外效性没有被验证,模型存在的过拟合或欠拟合程度不得而知。要平衡这个问题,需要有更大的样本和多中心的人群。不过,本研究为医学领域提供了一个很好的例子,未来随着多中心大型临床数据中心的建立,其他各种模型可以更好地构建起来,用于指导临床实践或提示疾病风险因素。
综上,机器学习模型可以精确预测术后AKI的发生。这些模型使血管外科医生能够更早地处理并发症,并且可能有助于提高AAA患者OSR的临床疗效。
基金资助
湖南省自然科学基金(2021JJ31102)。
This work was supported by the Natural Science Foundation of Hunan Province, China (2021JJ31102).
利益冲突声明
作者声称无任何利益冲突。
作者贡献
盛昌 论文设计、数据采集和分析,论文撰写与修改;廖明媚 论文指导;周海洋 数据采集;杨璞 论文设计、指导及修改。所有作者阅读并同意最终的文本。
原文网址
http://xbyxb.csu.edu.cn/xbwk/fileup/PDF/202302213.pdf
参考文献
- 1. Brewster DC, Cronenwett JL, Hallett JW, et al. Guidelines for the treatment of abdominal aortic aneurysms. Report of a subcommittee of the Joint Council of the American Association for Vascular Surgery and Society for Vascular Surgery[J]. J Vasc Surg, 2003, 37(5): 1106-1117. 10.1067/mva.2003.363. [DOI] [PubMed] [Google Scholar]
- 2. Suckow BD, Goodney PP, Columbo JA, et al. National trends in open surgical, endovascular, and branched-fenestrated endovascular aortic aneurysm repair in Medicare patients[J]. J Vasc Surg, 2018, 67(6): 1690-1697. 10.1016/j.jvs.2017.09.046. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Chaikof EL, Dalman RL, Eskandari MK, et al. The Society for Vascular Surgery practice guidelines on the care of patients with an abdominal aortic aneurysm[J]. J Vasc Surg, 2018, 67(1): 2-77. 10.1016/j.jvs.2017.10.044. [DOI] [PubMed] [Google Scholar]
- 4. Patel R, Sweeting MJ, Powell JT, et al. Endovascular versus open repair of abdominal aortic aneurysm in 15-years’ follow-up of the UK endovascular aneurysm repair trial 1 (EVAR trial 1): a randomised controlled trial[J]. Lancet, 2016, 388(10058): 2366-2374. 10.1016/S0140-6736(16)31135-7. [DOI] [PubMed] [Google Scholar]
- 5. Lederle FA, Kyriakides TC, Stroupe KT, et al. Open versus endovascular repair of abdominal aortic aneurysm[J]. N Engl J Med, 2019, 380(22): 2126-2135. 10.1056/NEJMoa1715955. [DOI] [PubMed] [Google Scholar]
- 6. De Bruin JL, Baas AF, Buth J, et al. Long-term outcome of open or endovascular repair of abdominal aortic aneurysm[J]. N Engl J Med, 2010, 362(20): 1881-1889. 10.1056/NEJMoa0909499. [DOI] [PubMed] [Google Scholar]
- 7. Becquemin JP, Pillet JC, Lescalie F, et al. A randomized controlled trial of endovascular aneurysm repair versus open surgery for abdominal aortic aneurysms in low- to moderate-risk patients[J]. J Vasc Surg, 2011, 53(5): 1167-1173. 10.1016/j.jvs.2010.10.124. [DOI] [PubMed] [Google Scholar]
- 8. Walsh SR, Tang TY, Sadat U, et al. Endovascular stenting versus open surgery for thoracic aortic disease: systematic review and meta-analysis of perioperative results[J]. J Vasc Surg, 2008, 47(5): 1094-1098. 10.1016/j.jvs.2007.09.062. [DOI] [PubMed] [Google Scholar]
- 9. 陈忠, 王盛. 重视腹主动脉瘤的规范化治疗[J]. 中国普外基础与临床杂志, 2021, 28(11): 1401-1404. [Google Scholar]; CHEN Zhong, WANG Sheng. Attach importance to standardized treatment of abdominal aortic aneurysm[J]. Chinese Journal of Bases and Clinics in General Surgery, 2021, 28(11): 1401-1404. [Google Scholar]
- 10. Tsai S, Conrad MF, Patel VI, et al. Durability of open repair of juxtarenal abdominal aortic aneurysms[J]. J Vasc Surg, 2012, 56(1): 2-7. 10.1016/j.jvs.2011.12.085. [DOI] [PubMed] [Google Scholar]
- 11. Rocha RV, De Mestral C, Tam DY, et al. Health care costs of endovascular compared with open thoracoabdominal aortic aneurysm repair[J]. J Vasc Surg, 2021, 73(6): 1934-1941. 10.1016/j.jvs.2020.09.034. [DOI] [PubMed] [Google Scholar]
- 12. Saratzis A, Joshi S, Benson RA, et al. Editor’s choice-acute kidney injury (AKI) in aortic intervention: findings from the Midlands aortic renal injury (MARI) cohort study[J]. Eur J Vasc Endovasc Surg, 2020, 59(6): 899-909. 10.1016/j.ejvs.2019.09.508. [DOI] [PubMed] [Google Scholar]
- 13. Dubois L, Durant C, Harrington DM, et al. Technical factors are strongest predictors of postoperative renal dysfunction after open transperitoneal juxtarenal abdominal aortic aneurysm repair[J]. J Vasc Surg, 2013, 57(3): 648-654. 10.1016/j.jvs.2012.09.043. [DOI] [PubMed] [Google Scholar]
- 14. West CA, Noel AA, Bower TC, et al. Factors affecting outcomes of open surgical repair of pararenal aortic aneurysms: a 10-year experience[J]. J Vasc Surg, 2006, 43(5): 921-927. 10.1016/j.jvs.2006.01.018. [DOI] [PubMed] [Google Scholar]
- 15. Obata Y, Kamijo-Ikemori A, Inoue S. Clinical utility of urinary biomarkers for prediction of acute kidney injury and chronic renal dysfunction after open abdominal aortic aneurysm repair[J]. Int J Nephrol Renovasc Dis, 2021, 14: 371-384. 10.2147/IJNRD.S322916. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Gombert A, Prior I, Martin L, et al. Urine neutrophil gelatinase-associated lipocalin predicts outcome and renal failure in open and endovascular thoracic abdominal aortic aneurysm surgery[J]. Sci Rep, 2018, 8(1): 12676. 10.1038/s41598-018-31183-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Averdunk L, Rückbeil MV, Zarbock A, et al. SLPI-a biomarker of acute kidney injury after open and endovascular thoracoabdominal aortic aneurysm (TAAA) repair[J]. Sci Rep, 2020, 10(1): 3453. 10.1038/s41598-020-60482-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Zhou CY, Wang R, Jiang WJ, et al. Machine learning for the prediction of acute kidney injury and paraplegia after thoracoabdominal aortic aneurysm repair[J]. J Card Surg, 2020, 35(1): 89-99. 10.1111/jocs.14317. [DOI] [PubMed] [Google Scholar]
- 19. Fernández‐Delgado M CE, Barro S, Amorim D. Do we need hundreds of classifiers to solve real world classification problems?[J]. J Mach Learn Res, 2014, 15(1): 3133-3181. 10.5555/2627435.2697065. [DOI] [Google Scholar]
- 20. Hirata K, Nakaura T, Nakagawa M, et al. Machine learning to predict the rapid growth of small abdominal aortic aneurysm[J]. J Comput Assist Tomogr, 2020, 44(1): 37-42. 10.1097/RCT.0000000000000958. [DOI] [PubMed] [Google Scholar]
- 21. Lee R, Jarchi D, Perera R, et al. Applied machine learning for the prediction of growth of abdominal aortic aneurysm in humans[J]. EJVES Short Rep, 2018, 39: 24-28. 10.1016/j.ejvssr.2018.03.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Turton EP, Scott DJ, Delbridge M, et al. Ruptured abdominal aortic aneurysm: a novel method of outcome prediction using neural network technology[J]. Eur J Vasc Endovasc Surg, 2000, 19(2): 184-189. 10.1053/ejvs.1999.0974. [DOI] [PubMed] [Google Scholar]
- 23. Kleinstreuer C, Li Z. Analysis and computer program for rupture-risk prediction of abdominal aortic aneurysms[J]. Biomed Eng Online, 2006, 5: 19. 10.1186/1475-925X-5-19. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Rengarajan B, Wu W, Wiedner C, et al. A comparative classification analysis of abdominal aortic aneurysms by machine learning algorithms[J]. Ann Biomed Eng, 2020, 48(4): 1419-1429. 10.1007/s10439-020-02461-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Monsalve-Torra A, Ruiz-Fernandez D, Marin-Alonso O, et al. Using machine learning methods for predicting inhospital mortality in patients undergoing open repair of abdominal aortic aneurysm[J]. J Biomed Inform, 2016, 62: 195-201. 10.1016/j.jbi.2016.07.007. [DOI] [PubMed] [Google Scholar]
- 26. Karthikesalingam A, Attallah O, Ma XH, et al. An artificial neural network stratifies the risks of reintervention and mortality after endovascular aneurysm repair; a retrospective observational study[J/OL]. PLoS One, 2015, 10(7): e0129024 [2022-01-06]. 10.1371/journal.pone.0129024. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Bellomo R, Ronco C, Kellum JA, et al. Acute renal failure - definition, outcome measures, animal models, fluid therapy and information technology needs: the Second International Consensus Conference of the Acute Dialysis Quality Initiative (ADQI) Group[J]. Critical Care, 2004, 8(4): R204-R212. 10.1186/cc2872. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Twine CP, Boyle JR. Renal dysfunction after EVAR: time for a standard definition[J]. J Endovascular Ther, 2013, 20(3): 331-333. 10.1583/12-4104C.1. [DOI] [PubMed] [Google Scholar]
- 29. Fernandes M, Majoni M, Garg AX, et al. Systematic review and meta-analysis of preventative strategies for acute kidney injury in patients undergoing abdominal aortic aneurysm repair[J]. Ann Vasc Surg, 2021, 74: 419-430. 10.1016/j.avsg.2020.12.061. [DOI] [PubMed] [Google Scholar]
- 30. Tang Y, Chen J, Huang K, et al. The incidence, risk factors and in-hospital mortality of acute kidney injury in patients after abdominal aortic aneurysm repair surgery[J]. BMC Nephrology, 2017, 18(1): 184. 10.1186/s12882-017-0594-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Dang T, Dakour-Aridi H, Rizwan M, et al. Predictors of acute kidney injury after infrarenal abdominal aortic aneurysm repair in octogenarians[J]. J Vasc Surg, 2019, 69(3): 752-762. 10.1016/j.jvs.2018.05.227. [DOI] [PubMed] [Google Scholar]
- 32. Knott AW, Kalra M, Duncan AA, et al. Open repair of juxtarenal aortic aneurysms (JAA) remains a safe option in the era of fenestrated endografts[J]. J Vasc Surg, 2008, 47(4): 695-701. 10.1016/j.jvs.2007.12.007. [DOI] [PubMed] [Google Scholar]
- 33. Tang Y, Chen JZ, Huang K, et al. The incidence, risk factors and in-hospital mortality of acute kidney injury in patients after abdominal aortic aneurysm repair surgery[J]. BMC Nephrol, 2017, 18(1): 184. 10.1186/s12882-017-0594-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Yeung KK, Jongkind V, Coveliers HM, et al. Routine continuous cold perfusion of the kidneys during elective juxtarenal aortic aneurysm repair[J]. Eur J Vasc Endovasc Surg, 2008, 35(4): 446-451. 10.1016/j.ejvs.2007.11.022. [DOI] [PubMed] [Google Scholar]
- 35. Bzdok D, Krzywinski M, Altman N. Machine learning: supervised methods[J]. Nat Methods, 2018, 15(1): 5-6. 10.1038/nmeth.4551. [DOI] [PMC free article] [PubMed] [Google Scholar]



