TY - JOUR
T1 - Vascular Age Assessed From an Uncalibrated, Noninvasive Pressure Waveform by Using a Deep Learning Approach
T2 - The AI-VascularAge Model
AU - Mitchell, Gary F.
AU - Rong, Jian
AU - Larson, Martin G.
AU - Korzinski, Timothy J.
AU - Xanthakis, Vanessa
AU - Sigurdsson, Sigurdur
AU - Gudnason, Vilmundur
AU - Launer, Lenore J.
AU - Aspelund, Thor
AU - Hamburg, Naomi M.
AU - Gotal, John D.
AU - Vasan, Ramachandran S.
N1 - Publisher Copyright:
© 2024 Lippincott Williams and Wilkins. All rights reserved.
PY - 2024/1/1
Y1 - 2024/1/1
N2 - BACKGROUND: Aortic stiffness, assessed as carotid-femoral pulse wave velocity, provides a measure of vascular age and risk for adverse cardiovascular disease outcomes, but it is difficult to measure. The shape of arterial pressure waveforms conveys information regarding aortic stiffness; however, the best methods to extract and interpret waveform features remain controversial. METHODS: We trained a convolutional neural network with fixed-scale (time and amplitude) brachial, radial, and carotid tonometry waveforms as input and negative inverse carotid-femoral pulse wave velocity as label. Models were trained with data from 2 community-based Icelandic samples (N=10 452 participants with 31 126 waveforms) and validated in the community-based Framingham Heart Study (N=7208 participants, 21 624 waveforms). Linear regression rescaled predicted negative inverse carotid-femoral pulse wave velocity to equivalent artificial intelligence vascular age (AI-VA). RESULTS: The AI-VascularAge model predicted negative inverse carotid-femoral pulse wave velocity with R2=0.64 in a randomly reserved Icelandic test group (n=5061, 16%) and R2=0.60 in the Framingham Heart Study. In the Framingham Heart Study (up to 18 years of follow-up; 479 cardiovascular disease, 200 coronary heart disease, and 213 heart failure events), brachial AI-VA was associated with incident cardiovascular disease adjusted for age and sex (model 1; hazard ratio, 1.79 [95% CI, 1.50-2.40] per SD; P<0.0001) or adjusted for age, sex, systolic blood pressure, total cholesterol, high-density lipoprotein cholesterol, prevalent diabetes, hypertension treatment, and current smoking (model 2; hazard ratio, 1.50 [95% CI, 1.24-1.82] per SD; P<0.0001). Similar hazard ratios were demonstrated for incident coronary heart disease and heart failure events and for AI-VA values estimated from carotid or radial waveforms. CONCLUSIONS: Our results demonstrate that convolutional neural network-derived AI-VA is a powerful indicator of vascular health and cardiovascular disease risk in a broad community-based sample.
AB - BACKGROUND: Aortic stiffness, assessed as carotid-femoral pulse wave velocity, provides a measure of vascular age and risk for adverse cardiovascular disease outcomes, but it is difficult to measure. The shape of arterial pressure waveforms conveys information regarding aortic stiffness; however, the best methods to extract and interpret waveform features remain controversial. METHODS: We trained a convolutional neural network with fixed-scale (time and amplitude) brachial, radial, and carotid tonometry waveforms as input and negative inverse carotid-femoral pulse wave velocity as label. Models were trained with data from 2 community-based Icelandic samples (N=10 452 participants with 31 126 waveforms) and validated in the community-based Framingham Heart Study (N=7208 participants, 21 624 waveforms). Linear regression rescaled predicted negative inverse carotid-femoral pulse wave velocity to equivalent artificial intelligence vascular age (AI-VA). RESULTS: The AI-VascularAge model predicted negative inverse carotid-femoral pulse wave velocity with R2=0.64 in a randomly reserved Icelandic test group (n=5061, 16%) and R2=0.60 in the Framingham Heart Study. In the Framingham Heart Study (up to 18 years of follow-up; 479 cardiovascular disease, 200 coronary heart disease, and 213 heart failure events), brachial AI-VA was associated with incident cardiovascular disease adjusted for age and sex (model 1; hazard ratio, 1.79 [95% CI, 1.50-2.40] per SD; P<0.0001) or adjusted for age, sex, systolic blood pressure, total cholesterol, high-density lipoprotein cholesterol, prevalent diabetes, hypertension treatment, and current smoking (model 2; hazard ratio, 1.50 [95% CI, 1.24-1.82] per SD; P<0.0001). Similar hazard ratios were demonstrated for incident coronary heart disease and heart failure events and for AI-VA values estimated from carotid or radial waveforms. CONCLUSIONS: Our results demonstrate that convolutional neural network-derived AI-VA is a powerful indicator of vascular health and cardiovascular disease risk in a broad community-based sample.
KW - artificial intelligence
KW - cardiovascular diseases
KW - cohort studies
KW - deep learning
KW - vascular stiffness
UR - http://www.scopus.com/inward/record.url?scp=85180535450&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85180535450&partnerID=8YFLogxK
U2 - 10.1161/HYPERTENSIONAHA.123.21638
DO - 10.1161/HYPERTENSIONAHA.123.21638
M3 - Article
C2 - 37901957
AN - SCOPUS:85180535450
SN - 0194-911X
VL - 81
SP - 193
EP - 201
JO - Hypertension
JF - Hypertension
IS - 1
ER -