Assessment of arterial stiffness in populations at increased cardiovascular risk

<p dir="ltr">Introduction</p><p dir="ltr">Arterial stiffness is an established marker of cardiovascular risk. However, its full potential for clinical applications remains unconfirmed due to the lack of accessible assessment methods. Photoplethysmography (PPG), which measures instantaneous blood volume changes in the skin, may be used for evaluation of arterial stiffness. PPG is common in medical pulse-oximeters to measure oxygen saturation levels and pulse rate, and has also recently been incorporated into consumer wearable devices such as smart watches and smart rings. Furthermore, machine learning may be applied on PPG signals and clinical data, to improve assessment of arterial stiffness and to facilitate identification of individuals having an increased cardiovascular risk, diagnose or condition.</p><p dir="ltr">This thesis aimed to investigate assessment of arterial stiffness in populations at increased cardiovascular risk, using novel and simplified methods to enhance cardiovascular risk stratification. Specifically, we aimed at (1) evaluating an improved overnight PPG-based measure for assessing cardiovascular risk, (2) improving simple PPG- and ECG-based methods for assessing arterial stiffness, (3) evaluating whether arterial stiffness or machine learning-based PPG analysis could improve risk stratification in patients with suspected obstructive coronary artery disease (CAD), (4) predicting the clinically challenging blood pressure phenotype masked uncontrolled hypertension (MUCH) post-myocardial infarction (MI) by applying machine learning methods.</p><p dir="ltr">Methods and results</p><p dir="ltr">In Study I, associations between the novel overnight finger PPG-based stiffness index (OSI) and markers of cardiovascular risk and ambulatory blood pressure (ABP) were investigated in a population with confirmed or suspected hypertension (n=79). OSI was positively correlated with cardiovascular risk scores (SCORE2/SCORE2-OP: ρ = 0.40, <i>P</i> = 0.002 and Framingham: ρ = 0.41, <i>P</i> < 0.001), and with office (r = 0.34, <i>P</i> = 0.002), awake (r = 0.40, <i>P</i> < 0.001), and asleep pulse pressure (r = 0.47, <i>P</i> < 0.001), and ambulatory arterial stiffness index (r = 0.37, <i>P</i> < 0.001). OSI correlated with systolic ABP (asleep r = 0.55, awake r = 0.42; both <i>P</i> < 0.001) and diastolic ABP (asleep r = 0.36, <i>P</i> = 0.001). Generally, OSI showed stronger correlations compared to a previously studied overnight PPG- based marker of arterial stiffness.</p><p dir="ltr">In Study II, finger PPG and finger blood pressure were collected in generally healthy participants (n=33). Carotid-femoral pulse wave velocity (cfPWV; SphygmoCor) and brachial single cuff-based aortic pulse wave velocity (aoPWV; Arteriograph) were reference methods. PPG waveform features were extracted and engineered, and machine learning was applied for prediction model development. PPG-based models predicted cfPWV (root mean square error [RMSE] 0.70, <i>R</i>² 0.74) and aoPWV (RMSE 0.52, <i>R</i>² 0.92) well, which was comparable to repeatability and agreement of the reference methods. The novel PPG amplitude ratio, "Am b/Am pl", emerged as a key feature in modelling, showing strong correlations with cfPWV and aoPWV (r = - 0.81 and -0.75, respectively; both <i>P</i> < 0.001).</p><p dir="ltr">In Study III, patients investigated with coronary computed tomography angiography (CCTA) for suspected symptomatic new-onset chronic coronary syndrome (CCS) were assessed with aoPWV by Arteriograph and index finger PPG (n=141). Arterial stiffness measures were compared with clinical risk models in their discriminatory ability for obstructive CAD (CAD-RADS [CAD-reporting and data system] >3, indicating at least one moderate stenosis). aoPWV and PPG-derived cfPWV were not predictive of CAD-RADS ≥3. Machine learning identified three PPG features (waveform area "Ar OS", waveform area ratio "IPA" and time span "Tm N") that provided discriminatory ability for CAD-RADS ≥3 comparable to the risk factor-weighted clinical likelihood model (receiver operating characteristic area under the curve [AUC] 0.73 [95% confidence interval 0.61-0.85] vs 0.72 [0.62-0.82]), when implemented in a random forest model.</p><p dir="ltr">In Study IV, patients underwent ambulatory blood pressure monitoring (ABPM) following a recent hospitalisation for a MI (n=99). The blood pressure phenotype MUCH (office blood pressure <140/90 mm Hg at ABPM start but mean 24-h blood pressure ≥130/80 mm Hg, and on antihypertensive medication) was evaluated in machine learning applied to 62 clinical registry variables. Seventeen patients (18%) were found to have MUCH post-MI. The discharge diagnoses diabetes and hypertension, and kidney dysfunction were identified as key predictors of MUCH. The best machine learning model achieved a mean cross- validation AUC of 0.82 for predicting MUCH.</p><p dir="ltr">Conclusions</p><p dir="ltr">We developed and evaluated improved arterial stiffness assessment using the easy-to-use finger PPG method. Machine learning was successfully applied to PPG signals and clinical variables to enhance arterial stiffness estimation, and to identify obstructive CAD in patients with suspected new-onset CCS, and MUCH after MI. These findings confirm the potential value of using the widely accessible PPG method and machine learning for improved cardiovascular risk stratification.</p><h3>List of scientific papers</h3><p dir="ltr">This thesis is based on the following papers, which will be referenced by their corresponding Roman numerals:</p><p dir="ltr">I. <b>Hellqvist H,</b> Rietz H, Grote L, Hedner J, Sommermeyer D, Kahan T, Spaak J. Overnight stiffness index from finger photoplethysmography in relation to markers of cardiovascular risk and vascular ageing. Heart Vessels. 2025;40(10):895-904. <a href="https://doi.org/10.1007/s00380-025-02537-3" rel="noreferrer" target="_blank">https://doi.org/10.1007/s00380-025-02537-3</a></p><p dir="ltr">II. <b>Hellqvist H,</b> Karlsson M, Hoffman J, Kahan T, Spaak J. Estimation of aortic stiffness by finger photoplethysmography using enhanced pulse wave analysis and machine learning. Front Cardiovasc Med. 2024;11:1350726. <a href="https://doi.org/10.3389/fcvm.2024.1350726" rel="noreferrer" target="_blank">https://doi.org/10.3389/fcvm.2024.1350726</a></p><p dir="ltr">III. <b>Hellqvist H,</b> Karlsson M, Löfmark H, Kahan T, Spaak J. Assessment of arterial stiffness and machine learning analysis of finger photoplethysmography for prediction of obstructive coronary artery disease in patients with suspected chronic coronary syndrome. [Manuscript]</p><p dir="ltr">IV. <b>Hellqvist H,</b> Erlinge D, Lindahl B, Jernberg T, Oldgren J, James S, Al- Khalili F, Kahan T, Spaak J. Prevalence and prediction of masked uncontrolled hypertension in patients recently hospitalized for myocardial infarction. Eur Heart J Open. 2025;5(6). <a href="https://doi.org/10.1093/ehjopen/oeaf138" rel="noreferrer" target="_blank">https://doi.org/10.1093/ehjopen/oeaf138</a></p>