PPG Markers for Heart Failure Prediction: Waveform Features, Risk Stratification & Prognosis
**Heart failure affects over 64 million people worldwide and remains one of the leading causes of hospitalization, yet early detection and continuous ...
- PPG & Signal Processing
Heart failure affects over 64 million people worldwide and remains one of the leading causes of hospitalization, yet early detection and continuous monitoring remain significant clinical challenges. Photoplethysmography (PPG) has emerged as a promising non-invasive technology for heart failure risk stratification and prognosis because the PPG waveform encodes information about cardiac output, vascular compliance, autonomic function, and hemodynamic status, all of which are disrupted in heart failure pathophysiology.
This article examines the scientific evidence for PPG-derived biomarkers in heart failure prediction, from pulse wave morphology features to machine learning classification models. We cover the physiological basis of each marker, the clinical validation evidence, and the current limitations that must be addressed before PPG-based heart failure screening reaches clinical practice. For foundational PPG concepts, see our introduction to PPG technology. For details on the signal processing methods underlying these analyses, visit our PPG algorithms overview.
Physiological Basis: Why PPG Reflects Heart Failure
The PPG signal measures pulsatile blood volume changes in the microvascular bed, typically at the fingertip, earlobe, or wrist. This signal is shaped by the interaction between cardiac pump function, arterial wave propagation, vascular impedance, and autonomic regulation. Heart failure disrupts each of these mechanisms, producing measurable changes in the PPG waveform.
In left ventricular systolic dysfunction, reduced ejection fraction directly decreases the pulse pressure driving the PPG systolic peak. The slower rate of ventricular ejection alters the systolic upstroke time and slope. Elevated left atrial and pulmonary venous pressures change the diastolic component through altered venous return dynamics. Compensatory neurohormonal activation, including sympathetic nervous system upregulation and renin-angiotensin-aldosterone system activation, modifies peripheral vascular tone and thus the reflection wave characteristics visible in the PPG diastolic phase.
Arterial stiffness, which increases in heart failure due to chronic pressure and volume overload, alters pulse wave velocity and changes the timing and amplitude of reflected waves visible in the PPG contour. This is the fundamental mechanism underlying the augmentation index (AIx) and reflection index (RI) features extracted from PPG waveforms, both of which respond to vasoactive drugs, ageing, and large artery stiffness (Millasseau et al., 2006; doi:10.1097/01.hjh.0000239277.05068.87).
Autonomic dysfunction is a hallmark of heart failure, with reduced parasympathetic tone and elevated sympathetic activity. This manifests in the PPG signal as reduced heart rate variability (HRV), altered pulse amplitude variability (PAV), and changes in the low-frequency to high-frequency (LF/HF) spectral ratio of pulse rate variability. For an overview of the HRV metrics involved, see Shaffer & Ginsberg (2017, doi:10.3389/fpubh.2017.00258).
PPG Waveform Morphology Features
Pulse Contour Analysis
The shape of the PPG pulse wave contains information about cardiac function and vascular properties that is directly relevant to heart failure assessment. Key morphological features include:
Systolic peak amplitude and upstroke time. The amplitude of the systolic peak reflects pulse pressure, which decreases in heart failure with reduced ejection fraction (HFrEF). The systolic upstroke time (time from pulse onset to systolic peak) increases when left ventricular contractility is impaired. Elgendi (2012, doi:10.2174/157340312801215782) reviewed the characteristic features of the fingertip PPG waveform, the artifacts that corrupt it, and the indexes derived from it.
Dicrotic notch position and depth. The dicrotic notch, caused by aortic valve closure and the reflected pressure wave, shifts in position and depth with changes in aortic compliance and peripheral resistance. In heart failure, the dicrotic notch may become attenuated or absent due to reduced aortic compliance and altered wave reflection patterns.
Augmentation index (AIx). The AIx quantifies the contribution of the reflected wave to the systolic pressure peak. In PPG, this is computed as the ratio of the augmentation pressure (difference between the reflected wave peak and the systolic inflection point) to the total pulse amplitude. Elevated AIx indicates increased arterial stiffness and wave reflection magnitude, both associated with heart failure risk.
Reflection index (RI). The RI is calculated as the ratio of the diastolic peak amplitude to the systolic peak amplitude. This feature reflects arterial tone and wave reflection.
Second Derivative PPG (SDPPG) Analysis
The second derivative of the PPG waveform, also known as the acceleration plethysmogram (APG), decomposes the pulse wave into five characteristic waves (a, b, c, d, e) that correspond to different phases of the cardiac cycle and vascular response. Several SDPPG ratios have proven informative for cardiovascular risk assessment:
The b/a ratio reflects the ability of the arteries to accommodate the ejected blood volume and is used as an index of vascular aging.
The d/a ratio reflects late systolic wave reflection and is sensitive to vascular tone. Takazawa et al. (1998) showed that d/a fell after angiotensin (from -0.40 to -0.62) and rose after nitroglycerin (to -0.25) in 39 patients with simultaneous ascending aortic pressure recordings (doi:10.1161/01.HYP.32.2.365).
The aging index (AGI), defined as (b - c - d - e) / a, integrates information from all five SDPPG waves. In the same study it increased with age (r = 0.80) and was higher in subjects with a history of diabetes, hypertension, hypercholesterolemia, or ischemic heart disease. In 524 hypertensive patients, Bortolotto et al. (2000) found the aging index was higher in those with atherosclerotic alterations, although carotid-femoral pulse wave velocity reflected arterial changes better (doi:10.1016/S0895-7061(99)00192-2).
For more detail on PPG waveform feature extraction methods, see our PPG signal processing algorithms documentation.
Heart Rate Variability and Autonomic Markers
Heart rate variability derived from PPG (often termed pulse rate variability, PRV) provides a window into autonomic nervous system regulation, which is profoundly disrupted in heart failure. While PRV from PPG is not identical to HRV from ECG, a systematic review found sufficient accuracy at rest, although physical activity and some mental stressors impair agreement (Schäfer & Vagedes, 2013; doi:10.1016/j.ijcard.2012.03.119).
Time-Domain HRV Features
SDNN (standard deviation of NN intervals) is consistently reduced in heart failure patients. In the UK-HEART study of 433 outpatients with chronic heart failure, Nolan et al. (1998) found SDNN to be an independent predictor of all-cause mortality: the risk ratio for a 41.2-ms decrease in SDNN was 1.62 (95% CI 1.16-2.44), and annual mortality was 5.5% for SDNN > 100 ms versus 12.7% for 50-100 ms (doi:10.1161/01.CIR.98.15.1510). In 17 young subjects undergoing tilt testing, Gil et al. (2010) found no significant differences between PRV and HRV time-invariant indices, with correlations above 0.97 (doi:10.1088/0967-3334/31/9/015).
RMSSD (root mean square of successive differences) primarily reflects parasympathetic activity and decreases with heart failure severity. Reduced RMSSD is associated with increased risk of ventricular arrhythmias in heart failure populations. Using classification trees on ECG-derived HRV features from public databases, Pecchia et al. (2011) reported 96.39% accuracy for detecting heart failure and 79.31% for distinguishing severe from mild heart failure (doi:10.1109/TBME.2010.2092776).
pNN50 (percentage of successive RR intervals differing by more than 50 ms) is typically reduced in heart failure, reflecting diminished vagal modulation.
Frequency-Domain HRV Features
The spectral decomposition of pulse rate variability reveals characteristic shifts in heart failure. The LF/HF ratio (low-frequency 0.04-0.15 Hz to high-frequency 0.15-0.40 Hz power ratio) increases in heart failure due to sympathetic predominance and vagal withdrawal. Very low frequency (VLF) power, reflecting thermoregulatory, renin-angiotensin, and endothelial influences, also decreases and has prognostic value.
In 330 patients with chronic heart failure followed for 3 years, Guzzetti et al. (2005) found that depressed night-time VLF power (<= 509 ms^2) was independently related to death from progressive pump failure, while reduced night-time LF power (<= 20 ms^2) was linked to sudden death (doi:10.1093/eurheartj/ehi067). These frequency-domain features are readily extractable from PPG signals using standard spectral estimation techniques.
Nonlinear HRV Metrics
Nonlinear dynamics analysis of pulse rate variability has shown particular promise for heart failure characterization. Sample entropy (SampEn) quantifies signal complexity and regularity, with reduced values in heart failure indicating more regular, less complex cardiac dynamics. Sample entropy was introduced by Richman and Moorman (2000) as a less biased alternative to approximate entropy for physiological time series (doi:10.1152/ajpheart.2000.278.6.H2039).
Detrended fluctuation analysis (DFA) separates short-term (alpha1) and long-term (alpha2) fractal scaling properties. The short-term scaling exponent alpha1 has emerged as one of the strongest HRV-based predictors of adverse outcomes in heart failure. In 499 patients with heart failure and LVEF <= 35%, Mäkikallio et al. (2001) found that a short-term fractal scaling exponent below 0.90 carried a risk ratio of 1.9 for mortality, and the reduced exponent remained an independent predictor after multivariate adjustment (RR 1.4) (doi:10.1016/S0002-9149(00)01312-6).
For a comprehensive guide on HRV analysis and age-adjusted reference ranges, see our HRV chart by age.
Pulse Transit Time and Pulse Arrival Time
Pulse transit time (PTT), the time for the arterial pulse wave to travel between two anatomical sites, and pulse arrival time (PAT), measured from the ECG R-wave to the PPG pulse arrival, are both influenced by cardiac output and arterial compliance. In heart failure, reduced cardiac output and increased arterial stiffness alter these timing parameters.
Poon and Zhang (2005) estimated blood pressure from pulse transit time in 85 subjects over an average of 6.4 weeks, with errors of 0.6 +/- 9.8 mmHg systolic and 0.9 +/- 5.6 mmHg diastolic against reference readings (doi:10.1109/IEMBS.2005.1615827). Continuous PAT monitoring has been proposed as a method for detecting heart failure decompensation, as rising filling pressures and declining cardiac output produce systematic PTT/PAT changes days before symptomatic exacerbation.
For more on cuffless blood pressure estimation, which shares the same PPG timing features, see our cuffless blood pressure technology overview.
Machine Learning Models for Heart Failure Classification
Feature-Based Classification
Traditional machine learning approaches use hand-crafted PPG features (morphological, HRV, timing) as inputs to classifiers such as random forests and gradient boosting. Commonly used inputs include SDNN, RMSSD, systolic upstroke time, SDPPG ratios, and the reflection index. Published PPG-only heart failure classifiers are mostly small, single-center studies, and their reported accuracy has not yet been confirmed in independent prospective cohorts.
Deep Learning Approaches
Deep learning models that operate directly on raw or minimally processed PPG waveforms, such as 1D convolutional and recurrent networks, may capture features not explicitly defined by domain knowledge. Saliency methods such as Grad-CAM can show which parts of the waveform a model relies on, which helps with physiological interpretation.
Multi-Modal and Ensemble Approaches
Multisensor wearables show what continuous noninvasive monitoring can achieve. In the LINK-HF study of 100 patients monitored for up to 3 months after a heart failure admission, a disposable multisensor chest patch combined with a personalized machine-learning baseline detected precursors of heart failure rehospitalization with 76% to 88% sensitivity and 85% specificity, with a median of 6.5 days between the first alert and readmission (Stehlik et al., 2020; doi:10.1161/CIRCHEARTFAILURE.119.006513). That patch was chest-worn rather than PPG-based, but it shows the value of trend-based alerting that PPG wearables could adopt.
Combining PPG features with basic clinical variables (age, sex, BMI, medication status) is a common strategy for improving on PPG-only models.
Prognostic Stratification and Remote Monitoring
Beyond initial detection, continuous PPG monitoring offers the possibility of heart failure prognostic stratification and early decompensation detection. Several key findings support this application:
Trend analysis over days to weeks captures the gradual hemodynamic deterioration that precedes acute exacerbations. Physiological trends such as resting heart rate and HRV can shift in the days before clinical decompensation. In the CHAMPION trial, Abraham et al. (2011) showed that management guided by an implanted wireless pulmonary artery pressure sensor reduced heart-failure hospitalizations in NYHA class III patients (HR 0.70 at 6 months; 39% reduction over a mean 15 months of follow-up) (doi:10.1016/S0140-6736(11)60101-3). PPG-based approaches aim to achieve similar benefits non-invasively.
NYHA class discrimination from PPG features has been demonstrated by multiple groups. Pecchia et al. (2011), cited above, distinguished severe from mild heart failure using ECG-derived HRV; equivalent PPG-based severity classification still awaits validation in larger cohorts.
Ejection fraction estimation from PPG is an active area of research. Clinical accuracy remains insufficient for standalone EF estimation. The combination of PPG features with cardiac bioimpedance may improve these correlations.
Current Limitations and Research Gaps
Despite promising results, several limitations must be addressed before PPG-based heart failure prediction reaches clinical deployment:
Sample size and generalizability. Most published studies use cohorts of 50-300 subjects from single centers. Multi-center, prospective validation studies with diverse populations (varied ages, ethnicities, comorbidities) are essential. Heart failure with preserved ejection fraction (HFpEF) is particularly understudied in the PPG literature, despite representing approximately half of all heart failure cases.
Signal quality in the target population. Heart failure patients frequently have comorbidities (atrial fibrillation, peripheral vascular disease, diabetes) that degrade PPG signal quality. Peripheral edema, common in decompensated heart failure, can significantly alter PPG morphology independent of cardiac function changes. Robust motion artifact removal and signal quality assessment algorithms are prerequisites for reliable feature extraction.
Medication confounders. Beta-blockers, ACE inhibitors, diuretics, and other heart failure medications directly affect heart rate, vascular tone, and pulse morphology. Models must either account for medication effects or demonstrate robustness across different pharmacological regimens. See our article on PPG drug response monitoring for more on this topic.
Regulatory pathway. PPG-based heart failure screening or monitoring would require regulatory clearance as a medical device (FDA Class II or higher). The evidence requirements for such clearance, including prospective trials demonstrating clinical benefit, significantly exceed what has been published to date.
Conclusion
PPG-derived biomarkers for heart failure prediction represent a convergence of signal processing, cardiovascular physiology, and machine learning that holds genuine clinical promise. The physiological basis is sound: heart failure produces measurable changes in pulse morphology, autonomic regulation, and hemodynamic timing that are all encoded in the PPG waveform. Machine learning models combining these features have shown promise in small research studies, and deep learning may discover features beyond traditional domain knowledge.
The path to clinical implementation requires larger multi-center validation studies, robust handling of the comorbidities and medications common in the heart failure population, and regulatory-grade evidence of clinical benefit. For researchers and engineers working in this space, the combination of continuous PPG monitoring with trend analysis and early warning algorithms represents the most promising near-term application, leveraging the unique advantage of PPG as a non-invasive, continuous, wearable-compatible measurement modality.
For further reading on PPG signal analysis methods relevant to heart failure research, explore our algorithms documentation and our guide to PPG conditions monitoring.
References
- Millasseau, S. C., Ritter, J. M., Takazawa, K., & Chowienczyk, P. J. (2006). Contour analysis of the photoplethysmographic pulse measured at the finger. Journal of Hypertension, 24(8), 1449-1456. https://doi.org/10.1097/01.hjh.0000239277.05068.87
- Takazawa, K., et al. (1998). Assessment of vasoactive agents and vascular aging by the second derivative of photoplethysmogram waveform. Hypertension, 32(2), 365-370. https://doi.org/10.1161/01.HYP.32.2.365
- Bortolotto, L. A., Blacher, J., Kondo, T., Takazawa, K., & Safar, M. E. (2000). Assessment of vascular aging and atherosclerosis in hypertensive subjects: second derivative of photoplethysmogram versus pulse wave velocity. American Journal of Hypertension, 13(2), 165-171. https://doi.org/10.1016/S0895-7061(99)00192-2
- Schäfer, A., & Vagedes, J. (2013). How accurate is pulse rate variability as an estimate of heart rate variability? International Journal of Cardiology, 166(1), 15-29. https://doi.org/10.1016/j.ijcard.2012.03.119
- Nolan, J., et al. (1998). Prospective study of heart rate variability and mortality in chronic heart failure (UK-HEART). Circulation, 98(15), 1510-1516. https://doi.org/10.1161/01.CIR.98.15.1510
- Gil, E., et al. (2010). Photoplethysmography pulse rate variability as a surrogate measurement of heart rate variability during non-stationary conditions. Physiological Measurement, 31(9), 1271-1290. https://doi.org/10.1088/0967-3334/31/9/015
- Pecchia, L., Melillo, P., & Bracale, M. (2011). Remote health monitoring of heart failure with data mining via CART method on HRV features. IEEE Transactions on Biomedical Engineering, 58(3), 800-804. https://doi.org/10.1109/TBME.2010.2092776
- Guzzetti, S., et al. (2005). Different spectral components of 24 h heart rate variability are related to different modes of death in chronic heart failure. European Heart Journal, 26(4), 357-362. https://doi.org/10.1093/eurheartj/ehi067
- Richman, J. S., & Moorman, J. R. (2000). Physiological time-series analysis using approximate entropy and sample entropy. American Journal of Physiology - Heart and Circulatory Physiology, 278(6), H2039-H2049. https://doi.org/10.1152/ajpheart.2000.278.6.H2039
- Mäkikallio, T. H., et al. (2001). Fractal analysis and time- and frequency-domain measures of heart rate variability as predictors of mortality in patients with heart failure. American Journal of Cardiology, 87(2), 178-182. https://doi.org/10.1016/S0002-9149(00)01312-6
- Poon, C. C. Y., & Zhang, Y. T. (2005). Cuff-less and noninvasive measurements of arterial blood pressure by pulse transit time. IEEE EMBC 2005, 5877-5880. https://doi.org/10.1109/IEMBS.2005.1615827
- Stehlik, J., et al. (2020). Continuous wearable monitoring analytics predict heart failure hospitalization: the LINK-HF multicenter study. Circulation: Heart Failure, 13(3), e006513. https://doi.org/10.1161/CIRCHEARTFAILURE.119.006513
- Abraham, W. T., et al. (2011). Wireless pulmonary artery haemodynamic monitoring in chronic heart failure: a randomised controlled trial. The Lancet, 377(9766), 658-666. https://doi.org/10.1016/S0140-6736(11)60101-3
Related articles
Pulse Wave Analysis Using PPG: Waveform Features, Vascular Health, and Clinical Applications
Complete guide to pulse wave analysis from PPG signals. Covers augmentation index, stiffness index, derivatives, fiducial points, and vascular assessment.
PPG Waveform: Anatomy, Features, and Clinical Meaning
The PPG waveform encodes far more than heart rate. Learn the anatomy of each pulse wave, what dicrotic notch position means, and how waveform shape reveals cardiovascular health.
PPG for Cardiac Monitoring: From Heart Rate to Heart Failure Detection
PPG is becoming a first-line cardiac monitoring tool. This guide covers validated applications: heart rate, HRV, atrial fibrillation, heart failure, and emerging ECG-free arrhythmia detection.
Median Filter for PPG: When It Helps and When It Hurts Waveform Features
Practical guide to median filter use on PPG: remove impulse spikes, protect waveform morphology, choose kernel sizes, and combine with accelerometer gates.
PPG Diabetes Cardiovascular Risk
Type 2 diabetes mellitus carries a two-to-four times higher cardiovascular risk than the general population, driven largely by accelerated vascular ag...