Department of Electronics and Communication Engineering, Visvesvaraya National Institute of Technology, MH, Nagpur, India.
Department of Medicine, Division of Cardiology, Queen's University, Kingston, Ontario, Canada.
J Med Syst. 2020 Nov 11;44(12):208. doi: 10.1007/s10916-020-01675-7.
This study developed an office-based cardiovascular risk calculator using a machine learning (ML) algorithm that utilized a focused carotid ultrasound. The design of this study was divided into three steps. The first step involved collecting 18 office-based biomarkers consisting of six clinical risk factors (age, sex, body mass index, systolic blood pressure, diastolic blood pressure, and smoking) and 12 carotid ultrasound image-based phenotypes. The second step consisted of the design of an ML-based cardiovascular risk calculator-called "AtheroEdge Composite Risk Score 2.0" (AECRS2.0) for risk stratification, considering chronic kidney disease (CKD) as the surrogate endpoint of cardiovascular disease. The last step consisted of comparing AECRS2.0 against the currently utilized office-based CVD calculators, namely the Framingham risk score (FRS) and the World Health Organization (WHO) risk scores. A cohort of 379 Asian-Indian patients with type-2 diabetes mellitus, hypertension, and chronic kidney disease (stage 1 to 5) were recruited for this cross-sectional study. From this retrospective cohort, 758 ultrasound scan images were acquired from the far walls of the left and right common carotid arteries [mean age = 55 ± 10.8 years, 67.28% males, 91.82% diabetic, 86.54% hypertensive, and 83.11% with CKD]. The mean office-based cardiovascular risk estimates using FRS and WHO calculators were 26% and 19%, respectively. AECRS2.0 demonstrated a better risk stratification ability having a higher area-under-the-curve against FRS and WHO by ~30% (0.871 vs. 0.669) and ~ 20% (0.871 vs. 0.727), respectively. The office-based machine-learning cardiovascular risk-stratification tool (AECRS2.0) shows superior performance compared to currently available conventional cardiovascular risk calculators.
这项研究开发了一种基于办公室的心血管风险计算器，使用机器学习 (ML) 算法，该算法利用了集中的颈动脉超声。该研究的设计分为三个步骤。第一步包括收集 18 个基于办公室的生物标志物，包括 6 个临床危险因素（年龄、性别、体重指数、收缩压、舒张压和吸烟）和 12 个颈动脉超声图像表型。第二步包括设计一种基于 ML 的心血管风险计算器，称为“AtheroEdge Composite Risk Score 2.0”（AECRS2.0），用于风险分层，将慢性肾脏病 (CKD) 作为心血管疾病的替代终点。第三步包括将 AECRS2.0 与目前使用的基于办公室的 CVD 计算器（Framingham 风险评分 (FRS) 和世界卫生组织 (WHO) 风险评分）进行比较。这项横断面研究招募了 379 名患有 2 型糖尿病、高血压和慢性肾脏病（1 至 5 期）的亚裔印度患者。从这个回顾性队列中，获取了来自左、右颈总动脉远壁的 758 个超声扫描图像[平均年龄=55±10.8 岁，67.28%男性，91.82%糖尿病，86.54%高血压，83.11%慢性肾脏病]。使用 FRS 和 WHO 计算器的平均基于办公室的心血管风险估计分别为 26%和 19%。AECRS2.0 表现出更好的风险分层能力，其曲线下面积 (AUC) 比 FRS 和 WHO 分别高约 30%（0.871 对 0.669）和 20%（0.871 对 0.727）。基于办公室的机器学习心血管风险分层工具 (AECRS2.0) 与目前可用的传统心血管风险计算器相比具有更好的性能。
