Modeling the post-extrasystolic potentiation to determine heart failure

Premature ventricular contractions (PVC) occur naturally in the human heart and lead to a well-studied increase in cardiac contractility, referred to as post-extrasystolic potentiation (PESP), which behaves independently of many otherwise confounding factors. It is well-known that the degree of potentiation differs between healthy and failing hearts. In this paper, we present a method for non-invasive and non-obtrusive assessment of the PESP, leveraging photoplethysmography (PPG) measurements, revealing information on the degree of potentiation and thereby cardiac functioning. We modeled the PESP as a second-order system and we compared the parameters of this system between patients with differing ejection fractions. The method was developed on an internal dataset acquired during percutaneous, electrophysiological procedures. The parameters of the proposed second-order system are determined for each of these patients and the results show a correlation between cardiac ejection fraction and the damping factor, one of the parameters of the system. This model-based approach was demonstrated on the earlier mentioned non-obtrusive PPG measurements, as well as invasive blood pressure measurements. Modeling the PESP as a second-order model using information obtained from the non-invasive PPG signal reveals information showing a promising step towards enabling continuous monitoring of cardiac contractility using non-invasive means. The presented method could be applied in non-invasive, non-obtrusive PPG recording devices such as a smartwatch. This provides a continuously monitored estimate of cardiac contractility as an additional clinical parameter, as well as decreasing the time between reduction of contractility and potential hospital admission.

Photoplethysmography (PPG) is an optical measurement of changes in blood volume in the tissue underneath the location of the sensor and reflects the pulsatile activity of the heart (Allen, 2007; Alian and Shelly, 2014). PPG is widely available due to the prevalence of smartwatches and other similar wearable devices, which commonly include PPG measurements to provide wearers with information about their cardiac activity. The acquisition of PPG is low-cost, requiring a sensor that only consists of a light-emitting diode (LED) and a photo diode for detecting incoming light, non-invasive as the LED and photo diode can be placed against the skin, and non-obtrusive as the technology can be placed in devices such as a smartwatch, a ring, or a finger cap. Several vital signals can be determined from PPG signals, such as heart rate and heart rate variability (Galli et al., 2022; Biswas et al., 2019). Moreover, it has been shown that measuring PPG can be performed remotely, based on video recordings (Tasli et al., 2014).

However, PPG as a measurement modality suffers from several limitations (Loh et al., 2022; Park et al., 2022). Since the most common sensor location is the wrist, the recordings are often contaminated by motion artifacts unintentionally caused by the wearer. In addition, PPG signals lack an interpretable physical unit as it measures the relative changes in blood volume rather than an absolute quantity. The amount of light detected by the sensor depends on several factors, such as skin pigmentation, level of sensor-skin contact, sensor angle, posture, as well as physiological changes in the cardiovascular system, such as peripheral vascular resistance, blood pressure, and cardiac output. As a result, the absolute amount of light received by the photo diode is often considered not meaningful. Consequently, most PPG-based monitoring applications focus on relative measures, such as ratios in signal amplitude, rather than the absolute amplitude itself.

Post-extrasystolic potentiation (PESP) refers to an increase of the contractility of the heart that occurs in the beat following the compensatory pause after an extrasystolic beat (Cooper, 1993; Sinnecker et al., 2015). An example of such an extrasystolic beat is a premature ventricular contraction (PVC), which typically has only a short filling time and expels little blood out of the heart. This is because the cardiac muscle cells have not fully recovered from the previous beat and from the limited venous return during the shortened diastole. The subsequent compensatory pause following the PVC leads to an augmented cardiac output in the post-extrasystolic beat in comparison to regular heartbeats. This augmentation could in part be explained by the Bowditch effect, where cardiac inotropy is dependent on the heart rate (Balcazar et al., 2018) and the Frank-Starling principle, where the stroke volume is dependent on the left ventricle end-diastolic pressure (Delicce and Makaryus, 2024). However, the PESP is a case on its own, acting separately from the previously mentioned effects (Cooper, 1993; Sprenkeler and Vos, 2016).

PVCs are fairly common with a high prevalence in the general population, increasing with age, presence of an underlying heart disease or comorbidities such as hypertension or coronary artery disease (Klewer et al., 2022). The actual number of occurrences might be higher than reported, as PVCs are commonly only detected during routine cardiac evaluations and remain asymptomatic for most people. In subjects without an underlying cardiopulmonary disease, PVCs are often benign. However, a higher occurrence of PVCs and an enhanced PESP are typical findings in the case of heart failure (HF) (Steger et al., 2016; Phillips et al., 1990).

After the potentiation of the post-extrasystolic beat, the effect of potentiation of the heartbeats reduces in the next few heartbeats. After a few heartbeats, cardiac contractility and output have returned to what they were just before the PVC. Hardman (1994) measured that the PESP in human cardiac tissue degrades following an inverse exponential function with a time constant estimated to be around 800 milliseconds. Sprenkeler and Vos (2016) also showed that in the case of HF, the mechanical restitution is slower, corresponding to a larger time constant. Furthermore, Sinnecker et al. (2014) showed that oscillations can occur in the maximum arterial blood pressure amplitude due to the PESP in the case of severe HF. This oscillating behavior is attributed to a disruption in the calcium cycling (Siri-Angkul et al., 2021), more specifically to the sarcoplastic reticulum containing less calcium due to leaky ryanodine receptors and a reduced calcium uptake, which is partly compensated by the upregulation of the plasma membrane sodium-calcium exchanger (Hasenfuss and Pieske, 2002). This oscillating behavior was also demonstrated in animal studies (Wier and Yue, 1986; Kowlgi et al., 2020).

Although the oscillating maximum arterial blood pressure might have strong potential for diagnosing the severity of HF, the findings by Sinnecker et al. (2014) still rely on invasive measurement of blood pressure, limiting its possibilities for continuous monitoring. As mentioned before, the amplitude of PPG signals is affected by various factors, including cardiac contractility and cardiac output, as well as blood pressure. We therefore hypothesize that the oscillating pressure behavior, that follows a PVC in patients with severe HF, should also be reflected in oscillations of the amplitude in PPG signals.

HF is a complex clinical syndrome that occurs when the heart is unable to pump or receive blood effectively, leading to inadequate perfusion of organs and tissues (Golla et al., 2025). Heart failure with a reduced ejection fraction (HFrEF) is a subset of possible HF types, in which the heart has a reduced ejection fraction (EF) and is thus unable to expel most of the blood in the heart. HFrEF, also known as systolic heart failure, is diagnosed in an estimated 11.5 million people worldwide, with its occurrence expected to rise over the coming years (Murphy et al., 2020), and is determined as having an EF lower than 40 percent. As a result of HFrEF, compensatory myocardial remodeling occurs in an attempt to restore blood flow, but does result in reduced cardiac contractility, which leads to an even further reduced blood flow and increased regurgitation, lowering the EF even more (Trinh et al., 2025).

Detecting the change in time constant as well as a possible occurrence of oscillations in the decay of the potentiation using PPG could allow non-invasive tracking of heart activity and its degradation in the case of HF. Allowing continuous tracking and providing additional information in possible medication efficacy during rehabilitation or as an early warning of degrading cardiac function could strongly reduce the amount of hospitalizations due to acute heart failure. Research has also shown acute heart failure treatment profits from early diagnosis (Searle et al., 2016).

In this paper, we propose a method to describe the oscillations in blood pressure and PPG amplitude with a second-order system. Second-order systems commonly describe a mass-spring-damper system or an RLC circuit (Plonus, 2020) and these systems can describe both the increase in time for mechanical restitution as described by Sprenkeler and Vos (2016), as well as the oscillatory behavior as a response to an extrasystolic beat seen in the case of severe HF, as described by Sinnecker et al. (2014). We furthermore propose to reduce the number of model parameters in this second-order system to a single parameter that represents the damping of the system and we will investigate how this parameter relates to the ejection fraction (EF) in a group of patients who were diagnosed with HF with reduced ejection fraction.

In Section 2, the methodology for data acquisition and processing, as well as the proposed model as a second-order system are discussed. In Section 3, the resulting parameters of the models are compared for the acquired data. The results are then discussed in Section 4, as well as highlighting implications and the validity of made assumptions. Finally, in Section 5, the paper is concluded and summarized.

Data acquisition

The method was developed on an internal dataset acquired during percutaneous, electrophysiological ablation procedures. During these procedures, the heart is paced at a fixed interbeat interval. The fixed interbeat interval ensures a constant cardiac output, which reduces the influences of the mentioned Bowditch effect and Frank-Starling principle. This regular pace is then purposefully disrupted by introducing a PVC, after which the pacing is temporarily stopped to allow the heart to ‘respond’ with a compensatory pause, which is part of the post-extrasystolic response of the heart to a PVC. This response of the heart to the PVC leads to the potentiation of the pulses following the PVC. The degree of potentiation depends on both the length of the pause and the time interval between the PVC and the pulse that occurred before it. An example of the PESP can be seen in Figure 1.

ECG and arterial blood pressure (aBP) are recorded using the GE Mac-Lab Hemodynamic Recording System, at a sampling frequency of 1000 Hz. The PPG is recorded with the Philips Elan device, which uses two green LEDs and a photodiode to record the heart activity at a sampling frequency of 32 Hz. The device was positioned on the wrist, similar to a wristwatch. The ejection fractions were obtained using echocardiography maximally 6 months before the procedure, as is clinical routine, and is used as a substitute of cardiac contractility. We anticipate that no significant changes in EF have occurred between the echocardiography and the procedure as the patients had to be in a clinically stable situation for inclusion.

The dataset consists of seven patients, for a total of 394 PVCs that were followed by a PESP that was undisturbed by large artifacts. An overview of the included patients, including their measured ejection fraction (EF) and the amount of labeled PVCs taken into consideration for the development and evaluation of the algorithm can be found in Table 1.

Both aBP and PPG signals reflect variations in the contractility of the heart. We therefore hypothesize that a similar return to baseline occurs in the amplitude of these signals after a PVC and subsequent potentiation. In the methodology below, we will therefore generalize our description as much as possible, making it applicable to both aBP and PPG signals. Where our methodology differs between these two signals, we will mention this explicitly.

Pre-processing

For the PPG signal, the raw PPG signal is filtered using a second-order bandpass filter to suppress interferences outside the frequency band of interest (i.e., between 0.07 and 12 Hz), as described by Peláez-Coca et al. (2022). These interferences include baseline wander, as well as high-frequency noise, irrelevant to the current application. Similarly, the aBP signal was filtered using a second-order low-pass filter to suppress frequencies above 12 Hz that hold no relevant information for our application.

Next, pulses are detected using a peak detector, as described by Papini et al. (2017). Subsequently, the start of each individual pulse is defined as the start of a systolic upslope. The individual pulses are defined as the signal between two consecutive starts.

Pulses were then annotated by an expert based on the ECG, indicating each pulse as either a PVC pulse, a potentiated pulse (the first pulse following the PVC), or a normal pulse. We chose this procedure because a PVC is typically easily recognizable in the ECG because of the large electrical activation. In the PPG and aBP signals, a PVC is often more difficult to distinguish because of the small amount of blood that is expelled, leading to little variation in the aBP and PPG signals. Figure 1a shows an example of such an event with its labeling.

Our proposed method aims to model the recovery of the heart towards normal cardiac functioning after a PVC. As a result, any PVC that is not followed by a sequence of normal heartbeats, for instance because of the occurrence of another PVC before the heart has returned to normal functioning, cannot be used to model this recovery. For that reason, we chose to only include PVCs if and only if the pulse preceding the PVC was considered normal by the expert annotator and none of the five pulses that followed the PVC was labeled as a PVC.

PESP modeling

In this paper we propose to model the degradation of the potentiation over time caused by the extrasystole by using a second-order system. This system can be described in the general form where y (t) is the model output, x (t) is the model input, with the parameters ζ the damping factor, ω n the natural frequency, and K the gain of the system. For each included PVC, we model the input signal x t as a Dirac delta pulse with an offset/bias b in its amplitude: The output y (t) of the model describes the decay of the PESP, following the amplitudes of the PVC and the following five pulses. Here, we defined the amplitude as the absolute difference between the systolic peak value and the start diastolic value of the pulse, as indicated in Figure 1. The output y (t) is therefore a curve that shows the progression of the potentiation in the amplitude of either the PPG or aBP signal, depending on which signal was used to compute the pulse amplitudes.

The model parameters ζ, ω n, K, and b can subsequently be found by minimizing the mean squared error (MSE) between the recorded pulse amplitudes y i, t r u e and its modeled counterpart y i, m o d e l, where y i, m o d e l is the value of y (t) at the time t that corresponds to the position of the peak of the i -th pulse:

The minimization of the MSE is done using the limited memory BFGS algorithm implemented in the SciPy package in Python, using a maximum of 50 iterations. As a constraint in the minimization of the MSE, we assume that ζ is larger than zero, because a negative damping–while mathematically allowed–is physically impossible. In the optimization problem, the parameters are first determined for each potentiation as an initialization step for the final optimization problem, where all PESPs of the patient are considered together for the minimization of the error as defined in Equation 3. This sequential optimization strategy is proposed to deal with strong variations in the amplitude of the pulses, which, especially for the PPG, can be significant. Such variations can be caused by, for example, posture changes or the tightness with which the recording device was strapped to the wrist. As a result of these amplitude variations, the gain K and bias b parameters are expected to vary between different instances of the potentiation. We therefore expect that any information carried by the gain K and the bias b of the second-order model will be affected by strong confounders. We chose to ignore these parameters from further analysis and consider their variation in the minimization of the MSE (Equation 3) as a way of dealing with inter-patient differences in signal amplitude.

Time normalization

Following the suggestion by Sinnecker et al. (2014), we furthermore propose to replace the time index t in Equation 1 by a normalized time index τ. This means that the i -th beat after a PVC will occur at normalized time index τ = i and consequently the PVC occurs at τ = 0. It is important to note that when we now compare responses, the assumption is made that the heart rate is relatively constant between responses. As mentioned before the responses are themselves at least partially dependent on the amount of blood filling the left ventricle (Delicce and Makaryus, 2024), which also assumes a constant cardiac output. However, in the case of heart failure, this dependency is lessened, making the assumption less strict.

As shown by Sinnecker et al. (2014), when oscillations in the pulse amplitudes are present, they occur between consecutive pulses, meaning that under the condition that the beats occur at their normalized time index, the natural frequency ω n must be equal to π. This reduces Equation 1 to:

Undamped, the response of the system as defined in Equation 4 to an input x (τ) as defined in Equation 2 is comparable to a sine wave, which attains its extreme values at τ = 1 2, 1 1 2, 2 1 2, …. Therefore, to have the extreme values instead happen at the values of τ = 1,2,3, … and thereby coincide with the pulse peaks, the input is shifted by a half of the time index. As such, the input x (τ) is newly defined as

The response of the updated system as defined in Equation 4, using the input x(τ) as defined in Equation 5. This can be seen in Figure 2, showing two examples of a model response in case of overdamping and in case of underdamping.

The proposed simplifications yield an optimization problem the parameters ζ, K, and b, of which only one (i.e., ζ) describes the behavior of the PESP decay between patients. When the system is critically damped (ζ = 1) the system’s response reaches a value determined by the gain of the system K and returns back to the original level that is determined by the bias b. In the case of overdamping (ζ > 1), as shown in Figure 2a, the system’s output rises quickly in response to the input, but, because of the damping, does not reach the value determined by the gain K. After the initial rise, it decays back to the original level following an exponential curve. Increasing the value of ζ causes this response to rise more slowly to a smaller peak amplitude, and delays the decay to the original level b. In the case of underdamping (0 < ζ < 1), as shown in Figure 2b, the system rises quickly in response to the input pulse, but exceeds the value as determined by the systems gain K. Because of the underdamping, it will also quickly return back to its baseline value, but then ‘overshoot’ its target and oscillate around the original level, with again a decaying amplitude. Further reducing the damping factor towards zero means the overshoot becomes stronger and that it will take longer for oscillations to reduce.

Bootstrapping

To reduce the effect of artifacts and outliers and enable the evaluation of the robustness of the proposed method against outliers, the method was executed a total of 256 times per patient on a randomly sampled subset of the data, containing only 70 percent of the PESP responses, rounded to the nearest integer. This method can be considered a form of bootstrapping to check the stability of the result and produces a distribution for ζ per patient. The distribution and the final determined value of ζ can then be compared, and depending on the value of ζ, a corresponding estimate of the ejection fraction can be made.

The described method was applied to both the aBP signals and the PPG signals. In Figures 3, 4, the result of fitting the second-order system to several of the measured peak amplitudes in the PPG signal can be seen for two different patients.

The resulting distributions of the damping factor ζ for all patients are shown in Figure 5. It can be observed that for both the aBP signal and the PPG signal the damping factor ζ increases with decreasing ejection fraction. However, at an ejection fraction of about 35 %, from hereon referred to as the collapsing point, this relation changes and the damping factor ζ reduces to values below 1, indicating the presence of oscillations in the pulse amplitudes. After this point, there is no obvious increase or decrease of the damping factor with a further decreasing ejection fraction. The value of ζ however stays below 1, indicating that the oscillating behavior persists within patients with lower ejection fractions.

From the comparison between panels 5a and 5b, it can be seen that from the wide interquartile ranges that the variance of the damping factors is larger in the case of PPG pulses than for aBP pulses. This is in line with our expectations because PPG signals typically suffer from lower signal to noise ratios and the amplitudes of the PPG pulse are affected more by aforementioned confouders. However, the same trend of increasing ζ with decreasing ejection fraction and a collapsing point around an ejection fraction of 35 % is still visible.

In this paper, we propose a method to describe the PESP that occurs after an extra-systolic beat. Following a PVC, the amplitudes of subsequent pulses are taken as input to fit a second-order model that, in response to a Dirac delta pulse disturbance, models the decay of pulse amplitudes and thereby, the decay of the potentiation of the cardiac muscle. To simplify the model (i.e., reduce the number of model parameters) and reduce the chance of overfitting the model to artifacts or outliers, we proposed to normalize the time axis by using beat numbers instead of interbeat intervals and thereby assume the natural frequency of the model to be equal to π. We obtained multiple parameters describing this second-order model, namely, the gain, bias, and damping factor. This is done for a small dataset that was obtained during percutaneous electrophysiological ablation procedures, in which the heart is paced and PVCs were triggered manually. We showed that the damping parameter of the second-order model, fitted to the pulse amplitudes after the PVC, shows a clear correlation with the measured ejection fraction of the patient. This suggests that the damping parameter can non-invasively and non-obtrusively indicate a measure of a reduced ejection fraction in a patient, possibly providing an early warning about degradation of cardiac function.

The simplicity of the model is also fit for implementation on wearable devices. While optimization problems can become very resource intensive, the current implementation fits the data on one parameter taken from small amount of pulses after a PVC. This ensures that the amount of data that has to be fit by the model does not explode in size. There is also no need for running in real-time, as the process only has to update on the occurrence of a PVC, which are uncommon in the general population.

As mentioned before, the PPG measurement is inherently more noisy than other measurements such as aBP. However, it does have the upside of being non-invasive and non-obtrusive, as well as becoming more and more commonplace due to the rise of devices such as smartwatches. The fact that the method shows similar values in the obtained damping factor for both the invasive blood pressure measurements and the non-invasive non-obtrusive PPG measurements shows great promise in its applicability. The influence of noise on PPG signals can be seen in Figures 3, 4; the determined amplitudes of consecutive pulses seem to be influenced by the noise, leading to variation around the fitted curves. Because of these varying pulse amplitudes, the parameters of the fitted curves on an individual sequence of the PESP can vary. In an extreme case, like demonstrated in Figure 3 (second row, center panel), even though the patient has a high ejection fraction and the overall damping is higher than 1, an individual noisy PESP event can still be fitted with a curve that suggests underdamping (ζ < 1). This demonstrates the need for multiple potentiations to be considered for a patient to reduce the influence of noise on the results. While this is also true for the pulse amplitudes and second-order model fits obtained from the invasive aBP signal, this is mostly important for the noisier non-invasive PPG signal.

When we consider the relation between the ejection fraction and the damping factor, as shown in Figure 5, we see that for lower ejection fractions the damping factor increases with decreasing ejection fraction. However, when the functionality of the heart has degraded to a certain level, this increase in damping ‘collapses’, and instead we observe a damping factor between 0 and 1. This means that oscillations occur, which can also be seen in the data (Figures 2b, 4) and has been reported before by Sinnecker et al. (2014). In their research, Sinnecker et al. split their cohort in two groups, survivors and non-survivors, and found that only the group of non-survivors showed oscillations in the systolic blood pressure after a PVC. This suggests that these oscillations in the potentiated response to a PVC can be indicative of severe heart failure.

The collapsing point where the system changes to an underdamped system seems to occur around an EF of 35 %. In the dataset, two patients have this reported ejection fraction, and one patient does not show these oscillations, while the other one does. In part this can be ascribed to low size of our dataset, but it also indicates that the EF does not tell the whole story in the state of heart failure. That is to say, EF does not fully describe the state of the cardiac tissue in HFrEF, just as the NYHA-score only describes the perceived impact on the quality of life of the patient.

When comparing Figures 5a and 5b, it can be seen that the values of the damping parameter obtained from the aBP signal are very similar to the ones obtained from the PPG signal. This is to be expected, as they both describe the same phenomenon and the degree of damping is considered to be independent of the signal amplitude. In the optimization of the model parameters, the variations in pulse amplitudes are accounted for by the scaling factor K, ensuring that the damping ζ only describes the relative amplitude changes. This makes our proposed model tolerant to variations in pulse amplitude that can be caused by either confounders such as posture changes or tighter wristbands in PPG signals or by monitoring with different modalities (e.g., PPG versus aBP).

As a limitation to our study, it should be noted that the amount of patients included in our dataset is low and we cannot draw strong conclusions. However, our results are in line with what we would expect from the pathophysiology of HFrEF, and similar behavior was also found by Sinnecker et al. (2014). This suggests that our findings are not coincidental and modeling the pulse amplitudes of aBP or PPG signals might indeed provide insight in deterioration of heart failure conditions. It is important to take the exploratory nature of this study into account, and findings will need to be validated in a larger cohort. To that end, the data is recorded during an electrophysiological procedure in which access to the arteries is also required, which is not an uncommon procedure. Because of this, the cohort can be extended within reason, taking into account the manual labeling that is part of the study.

Even though the dataset consists of 394 separate PVCs and their responses, it cannot be guaranteed these events are fully independent. Though PVCs are understood to occur independent of each other, due to the small amount of different sources of PVCs, other, less understood mechanisms could be in play. Therefore, the results described in this paper have to be further validated in larger cohort studies before being applied to clinical and everyday care.

Another limitation of our proposed method is that it is currently still based on the PESP, an extreme and enhanced disruption of the normal heart rhythm and behavior. While showing promising results, it means the current method remains local to PVC-induced disturbances, as well as the assumption of an otherwise normal heart rhythm. However, a similar process, yet on a smaller scale, occurs whenever the heart rate changes. Methods could therefore look at extracting similar cardiac information based on these more subtle changes in the cardiac potentiation after heart rate variations, or investigate whether similar behavior can be found in abnormal heart rhythms like atrial fibrillation.

Furthermore, the current implementation has been tested on data recorded in a controlled environment, reducing the amount and the severity of artifacts that are present in the PPG signal. More studies are needed to determine whether the phenomenon can still be seen when using data that is recorded during every day activity, in which the increased level of noise and artifacts might hinder correct retrieval of pulse parameters. Such studies might also determine whether certain assumptions made during the development of this algorithm are still valid in less controlled measurement conditions, like the assumption of constant cardiac output before the occurrence of a PVC. To that end, methods to reduce motion artifacts in PPG signals like the one described by Basso et al. (2026) could assist in this endeavor.

The last limitation that we would like to mention here is that the detection of PVCs in the PPG signal is non-trivial. While the large change in electrical behavior of the heart is easily detectable in an ECG signal, due to the reduced amount of blood that is expelled during a PVC, it can often appear in the PPG signal as an increased diastole of the previous pulse, making it difficult to detect. Especially in the current dataset, the low sampling frequency can further obfuscate the difference and exacerbate the difficulties of developing a detection algorithm. Presumably, an algorithm could be developed focusing on the sudden change in interbeat interval, but such an implementation might have problems with over-detecting the amount of PVCs, especially because they occur infrequently in the general population. However, approaches such as the one described by Regis et al. (2019) show promise in the automatic classification of premature beats, such as PVCs, in the PPG signal. Such an algorithm would also help increase the dataset size, as large publicly available datasets, such as the MIMIC-III dataset (Johnson et al., 2016), could be included for additional validation of the proposed method without requiring manual labeling of beats.

In this paper, we propose a method to characterize the potentiation that occurs in the myocardial tissue after a premature ventricular contraction. The post-extrasystolic potentiation is modeled using a second-order system that describes the decay of pulse amplitudes after the initial potentiation. We showed that our method can be applied to cardiac pulses that are obtained from arterial blood pressure measurements, but also to pulses that are recorded non-invasively and non-obtrusively using PPG. The damping in the second-order system, described by the damping parameter ζ, suggests that a distinction can be made between patients with different degrees of heart failure, where the damping changes to an underdamped system for patients with a highly reduced ejection fraction. While these results are promising in showing that the cardiac state can be characterized using non-invasive and non-intrusive continuous measurements, results have been obtained in a small cohort of patients and was collected in a highly controlled environment. This pilot study shows a strong promise of characterizing myocardial potentiation using PPG measurements, but a larger-scale study need to be conducted to further validate the findings before applying it in a clinical setting. Additionally, further research can be done on the applicability of our method in real-life situations, unlocking earlier detection of a degradation in the potentiation, but in which the data can be obfuscated and corrupted by various noise sources.