### Protocols and participants

Twelve healthy male volunteers who had no history of cardiac disease were non-smokers, and had BMIs between 20 and 25 kg m^{−2}, were recruited. Participants had a mean age of 22.1 years (± 3.1), and a mean BMI of 23.2 kg m^{−2} (± 4.2). None of the participants took drugs or any medication. The study protocol complied with the Declaration of Helsinki, was approved by the local Ethics Committee (Hôpital Erasme—CCB: B406201630013), and was registered on ClinicalTrials.org (April 11, 2017), identifier *NCT03107351*. The prototype of the KCG device used in this clinical trial was authorised by the Belgian Federal Agency for Medicine and Health Products (FAMHP). Written informed consent was obtained from each participant prior to the experimental testing procedure.

Sensors were placed on participants to connect to the ECG, BCG, and SCG recording systems. They had to lie in a supine position on an inversion table for 10 min for stabilisation. The recording was then started, and continuous data were acquired for the next 7 min. Participants were then put in HDT of 6 degrees for 10 min of stabilisation followed by 7 min of recording. In the final position in HUT of 80 degrees, the participants underwent another 10 min of stabilisation, followed by 7 min of recording.

Participants were asked to remain as still as possible, and to not talk or fall asleep for the duration of the recording.

The influence of breathing on the BCG signal is well known [3, 24, 36]. Hence, we accounted for the influence of respiration using an Imposed and Controlled Breathing (ICB) protocol. The ICB consisted of breathing for a fixed duration during the inspiration and expiration phases matching the normal inspiratory over total breath ratio (35%). An audio system with inspiratory and expiratory sounds was used to guide the participants. Participants were briefed with breathing instructions prior to performing 10 repetitions of 4-, 6-, 8-, and 10-s breathing cycles in succession. Post-hoc analysis showed that they maintained these cycles accurately.

### Data acquisition and devices

Two devices (Dev1 and Dev2) were used to simultaneously record the ECG, SCG, and BCG signals of the participants. Dev1 is the Cardiovector (Medical Computer Systems Ltd., Zelenograd, Russian Federation), a system used to monitor autonomic function on the International Space Station [37]. It allows the synchronous performance of: (a) electrocardiography (ECG), (b) impedance cardiography (ICG) in tetrapolar configuration, (c) plethysmography (PTG) using a nasal thermistor to evaluate breathing, (d) 6-DOF SCG (3-axis linear accelerations and 3-axis angular velocities, LIS344ALH and LPY403AL, STMicroelectronics), and (e) 6-DOF BCG (3-axis linear accelerations and 3-axis angular velocities, LIS344ALH and LPY403AL, STMicroelectronics). The acceleration was acquired at 0.5 mg and an RMS noise of 500 μg·Hz^{−1/2}, while the angular rate was acquired at a 100 mdps resolution and 10 mdps·Hz^{−1/2} RMS noise. An 80 Hz first-order analogue low-pass filter was applied to both linear accelerations and angular rates. All signals were sampled at 1 kHz and stored on a computer. The Dev1 had separate small plastic covers containing the accelerometer and gyroscope sensors that could be individually attached to the body. More details on the system can be found in [37].

Dev2 was the kinocardiograph (HeartKinetics SRL, Gosselies, Belgium), a portable device with two modules. Each module contained a 3-axis accelerometer and 3-axis gyroscope sensor (LSM6DSL, STMicroelectronics), and was attached to the body with standard sticky gel ECG electrodes (3 M). The acceleration and angular rate sensitivity range of the sensor were set to ± 2 g and ± 250 dps, respectively, with a resolution of 0.061 mg/LSB and 4.375 mdps/LSB and an RMS noise of 80 μg·Hz^{−1/2} and 4 \(\mathrm{mdps}\cdot{Hz}^{-1/2}\) with an output bandwidth of 416 Hz. The Dev2 was controlled with a smartphone or a tablet connected via Bluetooth. It collects a two-lead ECG at 200 Hz (ADS1292R, AD Instruments) together with 3-DOF linear accelerations and 3-DOF rotational velocities from each detector. A micro-controller (STM32F411, STMicroelectronics) was used to acquire synchronously data from each chip and operate the Bluetooth low energy (SPTLE-RF, STMicroelectronics). A total of 12-DOF linear acceleration and angular velocity signals were recorded at 50 Hz, and a 25 Hz first-order analogue low-pass filter was applied.

Dev2 consists of two parts which corresponds to the SCG and BCG sensors. For both BCG and SCG, the Dev1 sensors were adhesively attached on top of the Dev2 sensors, which were then positioned on the body with their ECG electrodes. The BCG sensors were placed in the lumbar lordosis curvature, between the second and the third lumbar vertebrae and close to the participant’s centre of mass. The SCG sensors were placed on the manubrium of the sternum, below the clavicle, and in the superior mediastinum region where the great vessels emerge from the heart (Fig. 5).

Data acquisition from both the Dev1 and Dev2 devices was initiated at the same time, and continuous data were acquired for the duration of the recording.

### Data analyses

All data analyses and statistics were performed on a custom-built KCG analysis toolbox written in Matlab 2018b (Mathworks, NL).

Signals acquired using Dev1 and Dev 2 were synchronised based on the ECG signal. The delay between the recordings of the two devices was determined by cross-correlating the ECG signals from both devices to find the point of correspondence. Signal synchronisation was validated visually for all records. The EA method, which is described in the next section, is based on the R waves detected on the ECG. However, there may have been slight differences in the ECG readings of the two devices and in the synchronisation of the ECG of both Dev1 and Dev2. Therefore, the ECG of Dev1 was used for the rest of the analyses for both devices.

The SCG and BCG signals of Dev2 were up-sampled to 1 kHz. Detection of the P, Q, R, S, and T waves was conducted using a modified Pan and Tompkins algorithm [38], and a pattern matching algorithm on the ECG channel. The detections were validated visually. The heart rate (HR) was calculated based on the RR intervals. Afterwards, the initial record was divided into 4 sub-records, one for each of the 4 phases of the ICB protocol (4-, 6-, 8-, and 10-s breathing cycles), and each sub-record was processed independently. In this study, only data from the 10-s breathing cycle protocol are presented. This was done to obtain the longest record possible within a constant breathing cycle.

#### Ensemble average method

The time reference for the EA is a fixed interval that took into account the cardiac activity preceding atrial depolarisation occurring before the R peak. Indeed, the beginning of the *n*th interval started before the P wave, at time \({k}_{n,\mathrm{start}}= {R}_{n}- \Delta\), where \({R}_{n}\) is the time of the *n*-th R peak and ∆ is 200 ms. The end of the cardiac cycle of interest was assumed to be \({k}_{n,\mathrm{end}}= {R}_{n}+ \mathrm{max}[{RR}_{i}]\) where \({RR}_{i}\) represent all the RR intervals of the current record. The mean was taken on all the beats to obtain an average ECG signal. Based on this, an EA was calculated for each channel of the BCG and SCG recordings.

#### Kinocardiography data analysis

Participant height and weight were used to assess their inertial parameters [39]. Linear accelerations (\(\vec{a}\)) first undergo single-time integration to provide velocity (\(\vec{v}\)). From this, the linear and rotational kinetic energy transmitted by cardiac contraction to the body was computed using the following equations on the ensemble-averaged signals:

$$K_{{{\text{Lin}}}} = \frac{1}{2}m(v_{x}^{2} + v_{y}^{2} + v_{z}^{2} )$$

(1)

$$K_{{{\text{Rot}}}} = \frac{1}{2}\left( {I_{xx} \omega_{x}^{2} + I_{yy} \omega_{y}^{2} + I_{zz} \omega_{z}^{2} } \right),$$

(2)

where \(m\) is the body mass of the subject, \(K_{{{\text{Lin}}}}\) is the linear kinetic energy, \(v_{x}\), \(v_{y}\), \(v_{z}\) are components of the velocity vector \(\vec{v}\), \(K_{{{\text{Rot}}}}\) is the rotational kinetic energy, \(I_{xx}\), \(I_{yy}\),\({\text{ and }}I_{zz}\) are the orthogonal components of the moment of inertia, and \(\omega_{x}\), \(\omega_{y} ,\) and \(\omega_{z}\) are components of the measured angular velocity \(\vec{\omega }\) coming from the gyroscope.

These computations gave the following ensemble-averaged signals: BCG \(K_{{{\text{Lin}}}}\), BCG \(K_{{{\text{Rot}}}}\), SCG \(K_{{{\text{Lin}}}}\), and SCG \(K_{{{\text{Rot}}}}\). The time integral (*iK*) of *K* over the ensemble-averaged cardiac cycle (CC) was computed as follows:

$${\text{SCG}} \;{{\text{iK}}_{{{\text{Lin}}}}} = \mathop \int \limits_{{{\text{CC}}}}^{{}} {\text{SCG }}K_{{{\text{Lin}}}} (t) {\text{d}}t$$

(3)

$${\text{SCG iK}}_{{{\text{Rot}}}} = \mathop \int \limits_{{{\text{CC}}}}^{{}} {\text{SCG }}K_{{{\text{Rot}}}} (t){\text{ d}}t$$

(4)

$${\text{BCG iK}}_{{{\text{Lin}}}} = \mathop \int \limits_{{{\text{CC}}}}^{{}} {\text{BCG }}K_{{{\text{Lin}}}} (t){\text{d}}t$$

(5)

$${\text{BCG iK}}_{{{\text{Rot}}}} = \mathop \int \limits_{{{\text{CC}}}}^{{}} {\text{BCG K}}_{{{\text{Rot}}}} (t){\text{d}}t.$$

(6)

Where CC, which denotes the entire cardiac cycle, defined as starting with the P wave of the cycle i and ending with the P wave of the cycle *i* + 1, was delimited based on the ensemble-averaged ECG. The integral was indeed performed on this specific CC interval and not the whole ensemble-averaged signal. This leads to the scalar metrics BCG \({{{iK}}_{{{\text{Lin}}}}} ,\) BCG \({{{iK}}_{{{\text{Rot}}}}}\), SCG \({{{iK}}_{{{\text{Lin}}}}}\), and SCG \({{{iK}}_{{{\text{Rot}}}}}\).

Detailed information on the signal processing of multidimensional BCG and SCG records can be found in our previous study [6].

### Statistical analyses

#### Influence of sample rate and sensor comparison

Based on the 1 kHz acquisitions of Dev1, the sample rates BCG and SCG data were decreased by a factor of 20 by keeping the first sample and then every 20th sample after the first. Signals at 50 Hz were therefore obtained. Bland-Altmann plots [40] were generated for each KCG metric calculated on the initial 1 kHz acquisition versus the down-sampled version at 50 Hz acquisition. To compare KCG metrics computed on basis of SCG and BCG signals acquired by different sensors, Bland–Altmann plots were also generated for each metric to compare Dev1 and Dev2 acquisitions. The measures were considered similar when Bland–Altmann plots displayed (1) an absence of positive or negative trend and (2) the presence of more than 90% of the points between ± 1.96 standard deviation (STD).

#### Ensemble average window length effect

Based on the 100 s signal from the 10 repetitions of 10 s breathing cycle, each record was categorised into 8 sections which corresponded to the first 20 s, 30 s, 40 s, 50 s, 60 s, 70 s, 80 s, and 90 s of the signal, respectively, to account for the influence of duration on the EA metrics. Data were grouped according to the different positions (supine, HDT, and HUT). To quantify the difference in the EA metrics caused by the window lengths, a plot of the differences expressed as a percentage of the value [(Method A − Method B)/mean %)] with 95% Confidence Interval (CI), as described by Giavarina [40], was generated.

#### Influence of position

The KCG parameters were entered into a linear mixed-effects model [41] together with position and time as fixed effects to assess their interactions. The random effects were the intercepts for the subjects. Indeed, position and time are considered as fixed effects,, because their impact on haemodynamics is expected to be seen on the variables of interest by a change of the intercept. This fixed effect on the intercept is subject to some randomness caused by inter-individual differences, hence the choice to consider subjects as a random effect. Visual inspection of the residual plots was performed to detect any deviation from homoscedasticity or normality. *p* values were obtained using likelihood ratio tests of the full model with the effect in question, against the model without this effect. Statistical significance was set at α = 0.05 (two-tailed hypothesis tests). KCG parameters were compared between positions by a paired *t* test for data with a normal distribution, or a Wilcoxon signed-rank test for skewed data. A Lilliefors test was used to test whether the difference between the compared sample populations was normally distributed. A Bonferroni correction was applied to account for multiple comparisons. Therefore, as 4 metrics were computed, a *p* value less than 0.0125 was considered to compute 95% confidence intervals.