Early Identification of Childhood Cancer Survivors at High Risk for Late Onset Cardiomyopathy: An Artificial Intelligence Approach utilizing Electrocardiography

项目概要/摘要 由于治疗和支持性护理的改善，儿童癌症的五年生存率现在超过85%。 然而，接受蒽环类药物化疗或胸部定向放射治疗的患者存在剂量相关风险 治疗心血管疾病后遗症，包括心肌病、冠状动脉疾病和心脏瓣膜病 疾病，对生活质量和总体生存率产生负面影响。早期认识和干预， 控制儿童癌症幸存者（CCS）的心脏病发病率可以提供改善 剩余生命的质量。为了促进心肌病的早期发现，儿童肿瘤学小组的 指南建议每2至5年用超声心动图（ECHO）进行CCS终身筛查。而 为早期发现心肌功能障碍提供了机会，筛查指南没有确定 收缩功能保留但将来可能发展为心肌病的患者。我们的总体 长期目标是开发一种通用的人工智能（AI）工具，使用ECG描记，可以识别 CCS具有未来心肌病的高风险。我们在圣犹达终身队列（SJLIFE）的一个子集上显示， 研究资料表明，CCS在10年内发生心肌病的高风险可以得到高准确性的预测 (AUC的0.87）通过人工智能（AI）仅使用原始数字心电图（ECG）数据。我们的目标 这个项目的目标是开发一个健壮的（目标1）、可推广的（目标2）和远程应用的（目标3）人工智能工具， 可以从低成本和高度可访问的ECG数据中识别具有心肌病风险的CCS。我们将实现我们的 我们的目标是实现三个具体目标： 目标1.开发一种人工智能工具来预测CCS中未来心肌病的风险：我们将利用来自 3，731名SJLIFE参与者将改进和内部验证一种预测CCS高风险的新型AI工具， 心肌病（定义为射血分数下降< 50% or >10%），在随后的3年，5年和10年。我们 将使用信号处理和深度学习来生成代表ECG的特征，并将这些特征用于 机器学习来预测心肌病 目标二。对Amsterdam LATER队列的一个亚组进行AI工具的外部验证。 我们将在艾玛儿童医院对343名接受儿童癌症治疗的CCS进行外部验证 医院/学术医疗中心在美国。我们将评估人工智能工具性能的一致性 在SJLIFE的LATER队列与hold out测试队列中。 目标3。评估通过智能手表进行远程心肌病预测的可行性。我们将收集 在常规检查期间通过智能手表对SJLIFE参与者子集进行ECG检查，并评估 使用智能手表ECG与临床ECG的AI工具进行风险预测的一致性。 影响：我们的研究结果提供了积极影响CCS健康的潜力，1）确定那些可能受益的人 从更频繁或更先进的心脏成像，和2）指导未来的研究，在远程和真实的时间 晚发性心肌病的预测 0

Project Summary/Abstract Due to improved treatment and supportive care, five-year survival rates for childhood cancer now exceed 85%. However, patients treated with anthracycline chemotherapy or chest-directed radiation have a dose-related risk for adverse cardiovascular sequelae, including cardiomyopathy, coronary artery disease and valvular heart disease, with a negative impact on quality of life and overall survival. Earlier recognition and interventions to manage cardiac morbidity among childhood cancer survivors (CCS) could provide opportunities to improve quality of remaining life. To facilitate early detection of cardiomyopathy, the Children's Oncology Group's guidelines recommend life-long screening of CCS with echocardiography (ECHO) every 2 to 5 years. While offering an opportunity for early detection of myocardial dysfunction, screening guidelines do not identify patients with preserved systolic function who may develop cardiomyopathy in the future. Our overarching long-term goal is to develop a generalizable artificial intelligence (AI)-tool using ECG tracings that can identify CCS at high risk for future cardiomyopathy. We have shown on a subset of St. Jude Lifetime Cohort (SJLIFE) study data that CCS at high risk for cardiomyopathy withing 10 years can be predicted with high accuracy (AUC of 0.87) via artificial intelligence (AI) using raw digital electrocardiography (ECG) data only. Our goal in this project is to develop a robust (Aim 1), generalizable (Aim 2), and remotely applicable (Aim 3) AI-tool that can identify CCS at cardiomyopathy risk from low-cost and highly-accessible ECG data. We will achieve our goal by following three specific aims: Aim 1. Develop an AI tool to predict risk of future cardiomyopathy among CCS: We will utilize data from 3,731 SJLIFE participants to refine and internally validate a novel AI-tool predicting CCS at high risk for cardiomyopathy (defined as ejection fraction < 50% or >10% drop), in the subsequent 3, 5, and 10 years. We will use signal processing and deep learning to generate features representing ECGs and use these features in machine learning to predict cardiomyopathy. Aim 2. Perform an external validation of the AI tool on a subgroup of the Amsterdam LATER Cohort. We will externally validate our AI-tool on 343 CCS treated for childhood cancer at the Emma Children's Hospital/Academic Medical Center in Netherland. We will assess the concordance of the AI-tool performance on the LATER cohort vs hold out test cohort at SJLIFE. Aim 3. Evaluate the feasibility of remote cardiomyopathy prediction via smartwatch. We will collect ECGs on a subset of SJLIFE participants via a smartwatch during their routine exam and assess the. concordance of risk predictions by AI-tool using smartwatch ECG vs clinical ECG. Impact: Our results offer the potential to positively impact CCS health by 1) identifying those who may benefit from more frequent or advanced cardiac imaging, and 2) guiding future studies in remote and real time prediction of late-onset cardiomyopathy. 0