A Machine Learning-Based Clinical Decision Support Tool to Predict Abdominal Aortic Aneurysm Prognosis Using Existing Longitudinal Data

总结：基于机器学习的临床决策支持工具，用于预测AAA预后 腹主动脉瘤（AAA）是一种局部扩张的主动脉。如果不治疗，AAA可能 继续破裂，这种情况的死亡率为90%，是第13大主要原因。 在美国，每年报告的死亡人数超过15，000人。在AAA之后， 诊断后，临床医生必须确定其严重性;即，破裂的相对风险与 干预。目前的临床指南，这一决定是基于一个尺寸适合所有 “最大直径标准”，规定当AAA直径达到5.5 cm时， 破裂需要修复动脉瘤。然而，较小尺寸的AAA（< 5.5 cm）已被 断裂率高达23.4%，表明这种基于直径的标准是不合适的 AAA管理。最近完成的一项由NIH资助的临床试验，1U 01-AG 037120：“非侵入性 AAA治疗临床试验”（N-TA 3CT）旨在证明 小AAA的药物治疗。在这次试验中，一个非常独特和有价值的数据集， 每6个月纵向收集一次小AAA患者的数据，持续3年。 该提议旨在检验以下假设：在发现小AAA时， 临床预后-即，根据破裂预测是否以及何时需要临床干预 风险度量-可以使用基于机器学习的算法， 生物力学、形态学和临床数据。为了解决这一假设，我们将追求两个 明确的目标。 目的1是量化N-TA 3CT试验中个体小型AAA的“演变”。的 每个时间点所有患者AAA的生物力学和形态学状态将根据 分别使用有限元分析和形态测量分析在试验期间收集的数据， 这些将与每个时间点每个AAA的临床指数一起沿着。 目标2将是开发和验证机器学习和回归技术，以预测 小AAA的临床预后。来自目标1的数据以及来自N- TA 3CT试验将用于训练机器学习分类模型，以确定是否 可以准确预测动脉瘤预后。将对数据子集进行验证， 评估建议的预测模式的准确性、灵敏度、精确度和特异性。 来自N-TA 3CT试验的独特数据集，结合广泛的经验和方法， 由我们实验室开发的，将使我们能够，第一次，仔细检查和量化自然 小AAA的演变，并随后开发预测模型以改善患者预后。

SUMMARY: A Machine Learning-Based Clinical Decision Support Tool to Predict AAA Prognosis Abdominal aortic aneurysm (AAA) is a localized dilatation of the aorta. If left untreated AAA may go on to rupture, an occurrence which has a 90% mortality rate and is the 13th leading cause of death in the United States, with more than 15,000 annual deaths reported annually. After AAA is diagnosed, a clinician must determine its severity; i.e., the relative risk of rupture compared to the risk of intervention. Current clinical guidelines for this determination is based on the one-size-fits-all “maximum diameter criterion”, which states that when a AAA reaches 5.5 cm in diameter, the risk of rupture necessitates repair of the aneurysm. However, smaller sized AAAs (< 5.5 cm) have been seen to rupture at rates of up to 23.4%, demonstrating that this diameter-based criterion is unsuitable for AAA management. A recently completed NIH-funded clinical trial, 1U01-AG037120: “Non-Invasive Treatment of AAA Clinical Trial” (N-TA3CT) was designed to demonstrate the efficacy of pharmacologic treatment of small AAA. During this trial, a highly unique and valuable dataset was collected longitudinally every 6 months for a 3-year period for patients presenting with small AAA. This proposal is designed to test the hypothesis that, at the time of discovery of small AAA, clinical prognosis – i.e., predicting if and when clinical intervention will be required based on rupture risk metrics – can be facilitated using machine learning-based algorithms using real-time biomechanical, morphological, and clinical data. To address this hypothesis, we will pursue two specific aims. Aim 1 will be to quantify the “evolution” of individual small AAA from the N-TA3CT trial. The biomechanical and morphological status of all patient AAAs at each timepoint will be determined from data collected during the trial using finite element analysis and morphometric analysis, respectively, and these will be tabulated along with clinical indices for each AAA at each timepoint. Aim 2 will be to develop and validate machine learning and regression techniques to forecast the clinical prognosis of small AAA. The data from Aim 1 as well as follow-up reporting data from the N- TA3CT trial will be used to train machine learning classification models to determine whether aneurysm prognosis can be accurately predicted. Validation will be performed on a subset of data to assess the accuracy, sensitivity, precision and specificity of the proposed prediction model. The unique dataset from the N-TA3CT trial, paired with the extensive experience of and methods developed by our lab, will allow us, for the first time, to carefully examine and quantify the natural evolution of small AAA and to subsequently develop a predictive model to improve patient prognosis.