This study involves a secondary analysis of an existing dataset [31]. The primary analysis focused on evaluating the accuracy of PPG-derived HR variability (HRV) [31], while in this analysis, the baroreceptor sensitivity is quantified by measuring HR changes following alterations in arterial line pressure or less invasive alternatives.

The interventional dataset was used to analyze the differences and similarities between BRS derived from invasive arterial line BP, VCP, and PPG. More details can be found in [31]. In summary, 28 male healthy volunteers (aged 52, SD 7 y; BMI 27, SD 4 kg/m², SBP 130, SD 12 mm Hg). Participants were equipped with four sensing modalities (see Figure 1): (1) arterial line inserted into radial artery on the nondominant arm, (2) infrared PPG at the index finger of the same arm (Biopac PPG100C, 240 Hz), (3) Finapress Nova at middle finger of the same arm, and (4) ECG in lead II configuration (Biopac ECG100C, 240 Hz). Participants performed a protocol of 3 repetitive sessions in supine position with the arm resting alongside the body. Each session included several interventions, namely two times paced breathing at 7 breaths for 3 minutes. A handgrip intervention during which the participant was asked to squeeze in a handgrip as much as possible for one and a half minutes using their dominant hand. In addition, a dedicated rest period where the participants were asked to close their eyes for 2 minutes. Extra (unlabeled) transition time was allocated in between activities.

In addition, a single 24-hour recording on a healthy volunteer was used to demonstrate BRS trends under free living conditions. In a normotensive male (33 y), a wearable prototype developed by imec was placed that recorded ECG and PPG for 24 hours. The ECG was placed in lead II configuration, and the transmissive PPG sensor Nonin 8000J was placed on the left index finger. In addition, an ambulatory BP measurement device (Suntech Medical Oscar2) recorded cuff-based oscillometric BP from the left upper arm in intervals of 15 and 30 minutes during the day and night, respectively. This was a regular office day, including 8 hours of sleep, 2 walks, and office work behind a desk.

The interventional study dataset was collected at Ziekenhuis Oost-Limburg in Genk, Belgium, and has been approved by the institutional review board (ethical committee approval 16/039U). All enrolled participants were compensated with a US $135 voucher. The ambulatory 24-hour feasibility data have been approved by the institutional review board of imec The Netherlands in Eindhoven, the Netherlands.

Both studies were conducted under the principles of the Declaration of Helsinki. All eligible participants were given the right to refuse participation and the right to withdraw from the study at any time. Written informed consent was collected from all participants before participation. To protect the participants’ privacy, all data collected from this study were kept confidential and anonymized and were only accessible to the members of the research team.

The dataset consisted of 28 male volunteers. Per participant, 3 sessions were recorded of several interventions, including the interventions analyzed here: rest, paced breathing, and handgrip. A total of 420 segments were analyzed to extract BRS values of different (surrogate) features. By averaging over the repeated measures, we obtained 140 data points for each participant and each intervention.

For all participants and interventions, the mean of the BRS_SBP,ABP and BRS_SBP,VCP were 8.1 (SD 3.0) milliseconds/mm Hg and 6.6 (SD 3.0) milliseconds/mm Hg, respectively, with intraparticipant precision of 15% for BRS_SBP,ABP and 21% for BRS_SBP,VCP, respectively. The PPG-based features show mean values of 3.2 (SD 1.9) au for BRS_PeakAmp,PPG, 93 (SD 60) au for BRS_Upstroke,PPG, 1.8 (SD 1.7) milliseconds for BRS_RDRatio,PPG, and 7.7 (SD 4.1) milliseconds/milliseconds for BRS_PAT,PPG.

The intraparticipant precision for the PPG-based features is on average higher: 54% for BRS_PeakAmp,PPG, 56% for BRS_Upstroke,PPG, and 44% for BRS_RDRatio,PPG. The intraparticipant precision of the BRS_PAT,PPG approaches that of the traditional BRS index (23% for BRS_PAT,PPG, compared to 20% and 26% for BRS_SBP,ABP and BRS_SBP,VCP, respectively). In all cases, the interparticipant variation (37% for BRS_SBP,ABP, 46% for BRS_SBP,VCP, 59% for BRS_PeakAmp,PPG, 64% for BRS_Upstroke,PPG, 91% for BRS_RDRatio,PPG, and 53% for BRS_PAT,PPG) exceeded the intraparticipant precision.

As described in the methods section, analysis was done in a step-by-step approach.

Figure 5 shows the exploratory 24-hour BRS recording from a healthy participant in free-living conditions. Except for a 1-hour walk around 5 PM and a short walk before 12 AM, the participant spent most of the day doing sedentary work. The participant was in bed between 12 and 8 AM. The lowest BRS was observed in the afternoon around 5 PM when the participant went for a walk, as also indicated by relatively high HR and low HRV. The activities were self-reported.

The BRS indices show a clear pattern over the 24 hours, with both BRS_PAT,PPG and BRS_RDRatio,PPG being at a lower level during the day and increasing at night, effectively during sleep. Given the inverse relation between PAT and SBP, the mirrored pattern of increasing SBP and decreasing PAT (and vice versa) is clearly evident, also during walking activities with contributing HR. The correlation coefficients between PAT and RDRatio with SBP were high at −0.90 and −0.63 (both P≤.05), respectively throughout the 24-hour trajectory. This confirms the essential validity of the observed trends based on the processed and qualified data. Interestingly, the correlation between SBP with the derived BRS indices (BRS_PAT,PPG and BRS_RDRatio,PPG) was equally high and significant (r=−0.74 vs r=−0.75; both P≤.05, respectively), while their mutual correlation was better (r=0.84; P≤.05). At a close look, the rhythmic oscillations during the night cannot only be seen in BRS_PAT,PPG and BRS_RDRatio,PPG but also in HRV_LF. Unlike HRV_LF, HRV_HF does not show any significant oscillations during the night, but instead displays a clear difference in absolute BRS level between day and night, which could not be observed in HRV_LF. Coherence of BRS_PAT,PPG also shows a slight increase during the night. Except for a few datapoints during the walking activities, where the measurements were affected by motion artifacts, the percentage of BRS_PAT,PPG above the coherence quality threshold of 0.14 never dropped below 80%, while the qualified coherence % of BRS_RDRatio,PPG was significantly lower, as directly reflected in the coherence profile of BRS_RDRatio,PPG. At all times, even during the walking events where motion artifacts are to be expected, sufficient data is above coherence threshold ensuring valid median values throughout the day.

This study illustrates that both the BRS based on PAT and the RDRatio derived from PPG serve as appropriate substitutes for the gold-standard BRS obtained from arterial lines. This is substantiated by the highest correlation observed during rest, a comparable coefficient of variation within participants, as well as anticipated alterations during activities. Furthermore, the feasibility of these measures was successfully tested over a 24-hour period under free-living conditions. This underscores their potential applicability in real-world scenarios.

Baroreflex sensitivity based on arterial line SBP is best correlated with BRS determined from pulse arrival time derived from PPG (PAT,P) signal: 66% of the variation in BRS_SBP,ABP is explained by BRS_PAT,PPG. The correlation of other surrogates to BRS_SBP,ABP is slightly lower. In almost all cases, the interparticipant variation is higher than the intraparticipant precision, which suggests that these BRS features can be used to identify differences between individuals.

The baroreflex plays a crucial role in maintaining BP, acting through various pathways to modulate HR, vascular resistance, and cardiac contractility. The challenge in determining the BRS noninvasively is measuring (systolic) arterial pressure as the gold-standard method, arterial line, is invasive or obtrusive. The volume clamp method is an often-used noninvasive substitute to determine arterial pressure. It is known to overestimate the central ABP [36] and shown to overestimate BRS in the low frequencies [37]. Nevertheless, it has been used as a reference to validate other BRS methods, like those derived from PPG [10,24-26]. Comparing BRS using arterial pressure derived from the invasive arterial line with volume clamp method and PPG reveals the disputes and benefits of the methods. The BRS based on SBP derived from the volume clamp method shows the best correlation of 0.78 with BRS based on SBP from the invasive arterial line and a correlation of 0.59 with the BRS based on the systolic peak in the PPG signal. The latter is lower compared to results from others, reporting a correlation of 0.77 [10]. Nevertheless, our study population is considerably older (compared to 28.5 y) and has relatively high BP, which would lower the BRS and, in turn, increase the influence of noise, thereby reducing the correlation. An underestimation of the BRS based on SBP from the volume clamp method compared to the arterial line–based SBP of 24% was reported [13], similar to the 19% reported here.

The 3 modalities, arterial line, volume clamp, and PPG, have differences and similarities important to consider. The volume clamp method uses a PPG signal as well. Although the PPG signal is not used to measure the arterial pressure directly, it is used to maintain a constant volume by adjusting the cuff pressure, such that it equals the finger arterial pressure. Therefore, in contrast to the arterial line method, both PPG and volume clamp methods are influenced by peripheral perfusion and, hence, temperature, motion, and contact pressure. This could potentially cause errors when relating peripheral to systemic hemodynamic changes [15]. These errors would be visible between arterial line and volume clamp or PPG-derived features but be similarly present between PPG and volume clamp–derived features.

The range of BRS features derived from the PPG used here is also reported previously, like pulse upstroke, peak amplitude, pulse arrival time, and rise time [10,24]. In addition, models estimating SBP using PPG use PAT as the most important feature [14]. Pulse arrival time and rise-decay ratio from PPG correlate equally well with BRS_SBP,ABP and BRS_SBP,VCP. Primary peak and pulse upstroke from PPG correlate well with BRS_SBP,VCP, but notably less with BRS_SBP,ABP, suggesting volume clamp underestimates the performance of BRS_PAT,P and BRS_RdRatio,P and overestimates the performance of others like BRS_Upstroke,P. These results suggest that BRS_SBP,V can be used as a method to measure BRS in a noninvasive way; however, care should be taken to check the performance of other (PPG-derived) BRS indices using this method.

Coherence was used as a measure of BRS reliability. It assesses the similarity of two signals in the frequency domain; if the HR and arterial pressure have similar frequency components as a result of the baroreflex, the BRS becomes more reliable.

BRS changes from rest to controlled activities show how well a feature tracks the baroreflex effects of the interventions. The correlation between SBP and HR increases for paced breathing intervention; therefore, an increase in BRS during paced breathing is expected. Especially at 6 breaths per minute, at which the HR and BP oscillation amplitude are increased [38]. Expectedly, during paced breathing, BRS increases for BRS_SBP,ABP and all BRS surrogates compared to that at rest. However, the correlation between BRS_SBP,ABP and BRS_PAT,PPG decreases with paced breathing. It has been shown that during slow breathing of around 6 breaths/minute the BRS is overestimated, since, in this case, other mechanisms in which respiration influences HR also concentrate in the LF frequency band [39]. During the handgrip intervention, the coherence is increased, likely because of a thoracic pressure increase damping the oscillatory pressure effect of respiration (data not shown). An overestimation of the BRS changes based on volume clamp and PPG compared to BRS_SBP,ABP is found, which is more prominent during paced breathing compared to handgrip or handgrip recovery.

The adequate robustness found for the BRS surrogates during controlled activities suggests a wider applicability for BRS monitoring, which was further assessed by means of a 24-hour recording under free-living conditions. Based on the structured analysis, the BRS based on PAT and RDRatio was considered the most promising to test for the 24-hour ambulatory; it had the lowest intraparticipant variation and highest correlation with BRS_SBP,ABP during rest.

A wearable prototype for continuous ECG and PPG signal acquisition allowed for computation of beat-to-beat PAT, RDRatio, and RR intervals, and thereby enabled the observation of characteristic patterns in BRS_PAT,PPG, BRS_RDRatio,PPG, and HRV.

The BRS indices were found to be higher during the night as compared to daytime, which is in line with previous experiments where BRS was obtained from an arterial line [28,40,41]. This expected behavior of increasing nighttime BRS (and a coherence up to 0.6) could be explained by the absence of other inhibitory influences on the baroreflex like emotional behavior and somatic afferent influences stimulated by muscle contraction, as proposed by [40].

In preceding 24-hour BRS studies, oscillatory patterns of BRS during nighttime were less prominent [28,30,40], likely because these studies either average over participants or longer time windows and typically report one datapoint per hour (unlike the 15 min interval in this study). Meanwhile, studies focusing on sleep stages do show an increase in baseline BRS during the night and oscillations in BRS and HRV_LF between rapid eye movement (REM) and non-REM sleep stages [42]. Supported by the findings from the interventional study and the coherence with HRV_LF, the observed patterns in the proposed BRS indices BRS_PAT,PPG and BRS_RDRatio,PPG undoubtedly reflect nocturnal BRS oscillations. Furthermore, the frequency of oscillations also appears to be in line with the typical duration and cycle times of adult sleep stages [43]. However, this remains to be further investigated with proper reference technology and in a larger population.

Furthermore, the proposed BRS indices may add value beyond existing HRV metrics. That is, both BRS indices show the baseline increase and nocturnal oscillations, whereas HRV_LF does not show a clear baseline increase and HRV_HF does not show oscillations, and its baseline increase may also be driven by different nocturnal respiration levels. Meanwhile, oscillations of BRS_PAT,PPG tend to compare higher with HRV_LF; yet, BRS_RDRatio,PPG seems to be more indicative of higher frequency contributions.

Regarding the reliability of the proposed indices, BRS_PAT,PPG shows constantly high coherence throughout the 24 hours. Even during activity, a sufficient percentage of qualified samples is present, which may be further enhanced with basic signal and feature qualification strategies. Although the coherence of BRS_RDRatio,PPG is substantially lower, it also remains above the threshold constantly with sufficient qualified samples. In terms of usability and technology integration, this may become a relevant compromise given that BRS_RDRatio,PPG holds the theoretical advantage to be computed using a (peripheral) PPG, hence without an ECG.

The nature of the PPG sensor also allows for free-living recordings, enabling the monitoring of the BRS of the patients on day-to-day activities. The obtained results from the 24-hour study are encouraging future research, considering the wide range of clinical applications for longitudinal BRS monitoring: as a prognostic tool for heart attacks and arrhythmias not only as single point measurement but also during sleep [11], and for cardiac mortality in patients with renal failure [44], or as a predictor of outcome after surgery [45]. Furthermore, present knowledge may be enhanced in day-to-day assessment of spontaneous BRS, which previously relied on 8-minute recordings on two consecutive days [46]. Variations in the 24-hour recordings between young and elderly people have also been reported [28]. Establishing a 24-hour recording could therefore show not only the BRS in short BP changes but also on circadian BP patterns. Ultimately, longitudinal BRS monitoring bears large potential for hypertension diagnostics, in particular for primary hypertension whose origin is widely unknown.

The study on arterial lines does present certain limitations, primarily due to the relatively limited sample size and the inclusion of only male participants. Further research is required to examine the impact of factors such as age and arterial stiffness on the BRS indices, as well as to explore their interrelationships. Such comprehensive analysis necessitates the involvement of larger and more diverse cohorts. Overall, these preliminary patterns of the BRS_PAT,PPG and BRS_RDRatio,PPG over a 24-hour period under free-living conditions support the findings from the controlled interventional study, demonstrating that it is possible to use PAT and RDRatio derived from the PPG signal to estimate the BRS. However, the key limitation of the 24-hour study is the confinement to a single participant, but the findings give rise to further expand this study to investigate circadian and nocturnal BRS patterns in both healthy participants, and given the clinical relevance, also patient cohorts.

BRS determined from pulse arrival time or RDRatio in a PPG signal is best correlated with BRS based on arterial line SBP. The BRS_PAT,PPG and BRS_RDRatio,PPG also follow the BRS_SBP,ABP during different physical activities. Furthermore, it allows for 24-hour BRS recordings, in which the expected circadian rhythm patterns are present.