Physiological Waveform Features Based Aging Clocks Associate with Cardiovascular Disease

Pal, Barney and Cannesson (UCLA) build PPG- and arterial-pressure-waveform 'aging clocks' from more than 840 per-beat time, frequency and statistical features on 17,327 surgical patients. LightGBM predicted chronological age with r = 0.92 (R² 0.81 for PPGAge, 0.82 for ABPAge). Patients whose predicted age ran ahead of their real age by at least the model MAE had higher odds of cardiovascular disease (PPGAge OR 1.87, 95% CI 1.59–2.21; ABPAge OR 2.28), and 'delayed agers' had lower odds (PPGAge OR 0.68).

• Form factor: perioperative monitor (not stated in abstract)

Aging affects the morphology of photoplethysmography (PPG) and arterial blood pressure (ABP) waveforms. Here, we demonstrate that machine learning (ML) models trained on features derived from PPG or ABP waveforms can predict an individual's age. We further investigated whether the discrepancy between predicted and chronological age (Age-Gap) could serve as a measure of cardiovascular health by testing its association with cardiovascular disease (CVD). From a large perioperative dataset of 17,327 surgical patients, high-fidelity PPG and ABP waveforms were analyzed, and over 840 time-, frequency-, and statistical-domain features per cardiac cycle were extracted using a signal-processing-based tool. These features were then used to develop ML-based aging clocks, PPGAge and ABPAge, trained on PPG and ABP features, respectively. For each aging clock, we evaluated five ML models: Light Gradient Boosting Machine (LightGBM), eXtreme Gradient Boosting (XGBoost), Random Forest, Lasso, and Elastic Net. Individuals were classified as having accelerated, normal, or delayed aging based on their Age-Gap relative to the model-specific mean absolute error (MAE, rounded up to the nearest whole year): accelerated aging was defined as Age-Gap [≥] +MAE, normal aging as -MAE [≤] Age-Gap < +MAE, and delayed aging as Age-Gap < -MAE. LightGBM provided the best age predictions for both aging clocks, with Pearson correlation coefficients (r) of 0.92 and 0.92 and R-squared values of 0.81 and 0.82 for PPGAge and ABPAge, respectively. Moreover, we found that accelerated aging was significantly associated with higher odds of CVD for both aging clocks, PPGAge (odds ratio, 1.87 [95% CI, 1.59-2.21], p<0.001) and ABPAge (odds ratio, 2.28 [95% CI, 1.51-3.43], p<0.001). Additionally, delayed aging was significantly associated with lower odds of CVD for PPGAge (odds ratio, 0.68 [95% CI, 0.58-0.80], p<0.001). Our findings suggest that aging clocks based on high-fidelity PPG or ABP waveform features may serve as novel biomarkers for identifying individuals with higher odds of CVD.

Verbatim from Semantic Scholar (opens in a new tab) · Abstract via Semantic Scholar; © the authors/publisher

No results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.