Are cuffless devices challenged enough? Design of a validation protocol for ambulatory blood pressure monitors at the wrist: the case of the Aktiia Bracelet

Abstract

The US and European guidelines for the diagnosis and management of hypertension recommend the introduction of systematic home and night Blood Pressure (BP) monitoring. Fully-automated wearable devices can address the needs of patients and clinicians by improving comfort while achieving measurement accuracy. Often located at the wrist and based on indirect BP measurements, these devices must address the challenges of ambulatory scenarios. New validation strategies are needed, but little guidance has been published so far.
In this work, we propose an experimental protocol for the validation of cuffless wrist BP monitors that addresses ambulatory environment challenges in a controlled experimental setting. The protocol assesses the robustness of the measurement for different body postures, the ability of the device to track BP changes, and its ability to deal with hydrostatic pressure changes induced by different arm heights. Performance testing using Aktiia Bracelet is provided as an illustration. The results of this pilot study indicate that the Aktiia Bracelet can generate accurate BP estimates for sitting and lying positions and is not affected by hydrostatic pressure perturbations.
Clinical Relevance - Automated cuffless BP monitoring is opening a new chapter in the way patients are being diagnosed and managed. This paper provides a guidance on how to assess the clinical utility of such devices when used in different body positions.

Full text

Abstract The US and European guidelines for the diagnosis
and management of hypertension recommend the introduction
of systematic home and night Blood Pressure (BP) monitoring.
Fully-automated wearable devices can address the needs of
patients and clinicians by improving comfort while achieving
measurement accuracy. Often located at the wrist and based on
indirect BP measurements, these devices must address the
challenges of ambulatory scenarios. New validation strategies
are needed, but little guidance has been published so far.
In this work, we propose an experimental protocol for the
validation of cuffless wrist BP monitors that addresses
ambulatory environment challenges in a controlled
experimental setting. The protocol assesses the robustness of the
measurement for different body postures, the ability of the
device to track BP changes, and its ability to deal with
hydrostatic pressure changes induced by different arm heights.
Performance testing using Aktiia Bracelet is provided as an
illustration. The results of this pilot study indicate that the Aktiia
Bracelet can generate accurate BP estimates for sitting and lying
positions and is not affected by hydrostatic pressure
perturbations.
Clinical Relevance Automated cuffless BP monitoring is
opening a new chapter in the way patients are being diagnosed
and managed. This paper provides a guidance on how to assess
the clinical utility of such devices when used in different body
positions.
I. INTRODUCTION
A. Hypertension and prediction of cardiovascular risk
Hypertension is a significant risk factor for cardiovascular
morbidity and mortality. However, the definition of
hypertension varies across countries and its interpretation
depends on the BP measurement technique that is used [1,2].
Three methods co-exist today when it comes to estimating BP
[3], i.e. Office Blood Pressure Monitoring, Home Blood
Pressure Monitoring (HBPM) and Ambulatory Blood Pressure
Monitoring (ABPM).
Professional (office) BP Monitoring relies on a single
measurement that can be either performed by a health
practitioner at the upper arm using auscultation or calculated
as a mean of multiple measurements performed by an
automated oscillometric device [4]. Even though the
measurement conditions are controlled by the clinician, they
can be affected by a general stress related to the medical visit,
a white-coat effect [5].
Contrary to the office monitoring, self-measured HBPM is
performed in an out-of-clinic scenario and usually spans over
J. Solà, A. Vybornova, S. Fallet, E. Olivero, B. De Marco, O.
Grossenbacher, N. Ignjatovic, B. Ignjatovic, M. Favre-Bulle, N. Levinson, N.
Siutryk, V. Chapuis, M. Bertschi are with Aktiia SA, Neuchâtel, Switzerland
a period of several weeks, providing a better clinical solution.
The measurement is generally taken twice per day with an
oscillometric cuff at the upper arm or at the wrist. The patient
is instructed to initiate the measurement when seated and
relaxed, with the device at the heart level.
Compared to the HBPM, the ABPM increases the resolution
of the BP patterns during a typical day. The patient is
monitored out-of-clinic over 24 hours using an oscillometric
device at the upper-arm. The device automatically triggers the
measurements during the day (every 20 minutes) and during
the night (every 1h). ABPM provides clinically relevant data,
such as BP variability and nocturnal hypertension.
With more than two decades of available longitudinal data
compiled on ambulatory patients, there is enough evidence to
state that the best predictor of end-organ damage and
cardiovascular events are the BP measurements at home, and
in particular during sleep [6]. Based on this evidence, the EU
and the US guidelines for the management of hypertension
integrate now the systematic use of ABPM [1,2].
B. Fully Automated Cuffless BP Monitoring
Although ABPM has multiple advantages over other clinically
used BP monitoring techniques, the measurement relies on the
inflation of the cuff around the arm that causes patients
discomfort and pain. This limitation leads to a reduced
compliance and a lower number of measurements available
per patient, in particular during the night [5,7-9]. There is thus
a clear need for technologies to satisfy both patients and
clinicians, capable of performing BP measurements accurately
and seamlessly int
A new category of fully automated wrist-located cuffless
devices is currently under development [3]. These devices give
hope for widespread longitudinal day and night BP
monitoring. Either based on semi-occlusive or non-occlusive
sensing technologies, these devices will be the first of a kind,
providing both day and night BP profiles without any lifestyle
interference. However, when validating these devices, one
should consider that the measurement may be taken at
different body positions that will generate important
hydrostatic pressure artifacts.
C. Posture-related BP changes and hydrostatic BP bias
The measurement of BP at body positions other than
impact on the measured
values:
- Posture-related BP changes: BP is intrinsically different at
different body positions, e.g. it is normally higher while
(+41 32 552 20 52, e-mail: josep@aktiia.com, www.aktiia.com).
B. S. Alpert, MD is with University of Tennessee, Memphis, Tennessee, USA
(retired).
J. Solà, A. Vybornova, S. Fallet, E. Olivero, B. De Marco, O. Grossenbacher, N. Ignjatovic,
B. Ignjatovic, M. Favre-Bulle, N. Levinson, N. Siutryk, V. Chapuis, M. Bertschi and B. Alpert

standing than when lying supine. While HBPM is
supposed to eliminate this effect by normalizing the
measurement conditions, ABPM reports the BP values in
different positions, without necessarily providing position
data for each measurement.
- Hydrostatic BP bias: at a given moment BP is also different
at each segment of the arterial tree. In particular, in sitting
or standing position, BP is lower at the level of the head
and higher in the feet, due to the hydrostatic pressure
effect. ABPM and HBPM at the upper arm are minimally
affected by this phenomenon (the measurement location is
always close to the heart level), HBPM at the wrist deals
with this effect by requesting the patient to place the wrist
the Note that a difference of 10 cm in
height below the heart level corresponds to a BP increase
of 7.5 mmHg.
Hydrostatic BP bias is thus an effect that will play a major
role in the usability of fully automated cuffless BP monitors.
In particular, for BP monitors that are located at the wrist,
different arm positions will generate different BP readings
even in situations where the underlying BP at the heart level
was unchanged. While these readings might be locally
accurate, they do not depict actionable clinical information as
the BP values required to diagnose or treat a hypertensive
patient. Figure 1 illustrates some typical measurement
conditions of daily life that can lead to hydrostatic BP bias
larger than 20 mmHg.
In order to be successfully integrated into current clinical
practices, fully automated cuffless BP monitors must find a
way to deal with the hydrostatic BP bias and accurately track
the BP changes in different body postures. Posture-related BP
changes are especially critical in the context of indirect BP
measurement, such as pulse wave analysis (PWA) or pulse
transit time (PTT). Indeed, the posture change affects the state
of a vascular bed and there is a risk that, depending on the
algorithm being used, the device is unable to track these
changes accurately.
II. DESIGN OF AN EXPERIMENTAL PROTOCOL TO TEST
AMBULATORY BP MONITORS AT THE WRIST
When it comes to validate the accuracy of BP monitors
there are currently several standards and guidelines: the
ISO81060-2 international standard [10], the IEEE1708
standard [11], and the ISO81060-3 draft international standard
[12]. However, there are few, if any, details on how to
ambulatory scenario, i.e. being able to accurately follow the
BP changes due to posture change and being unaffected by
hydrostatic bias. We suggest here a guidance on how to
address the robustness to hydrostatic BP changes during a
clinical validation.
A. Experimental protocol
During the clinical validation two devices should be used:
a reference device (REF) that provides BP measurements at
the heart level, and the device under test (DUT). It is important
to note that the reference device should itself satisfy the
requirements of being able to accurately track posture-related
BP changes and be unaffected by hydrostatic bias. Then,
during the data recording session(s), at least the following
steps should be performed in addition to the ones described by
the applicable standard:
0. Initialization: applicable only if the DUT is a calibrated
device [3].
1. Baseline assessment: REF and DUT devices record BP
in standard sitting and relax conditions [10].
2. BP change intervention: REF and DUT record BP
while a physical or drug intervention is implemented in
order to modify BP [12].
3. Body position intervention: REF and DUT record BP
while subjects change their body position, including
realistic scenarios foreseen by the intended use of the
device, e.g.
4. Hydrostatic bias intervention: while keeping the arm
where REF is located in a standard measurement
position, the arm where DUT is raised and/or lowered
to different levels.
B. Data analysis
In addition to the data analysis procedures described in the
applicable standard, the retrospective analysis of the collected
data should at least include the following steps:
1. Analysis of the performance of the DUT to track the
BP changes induced by using drug or physical
interventions. It is important to report the amount of
REF BP change that was actually induced during the
intervention.
2. Analysis of the performance of the DUT to track the
induced BP changes from changing body position.
3. Analysis of the ability of the DUT readings to remain
stable (similar to the REF) during hydrostatic bias
intervention.
4. Analysis of the acceptance rate of the DUT across the
different experimental conditions. For most of the
devices performing indirect BP estimations (i.e. based
on PWA or PTT) some measurements are
automatically discarded by the algorithms because the
data are too noisy or corrupted by motion artifacts.
Thus, the acceptance rate describes the percentage of
actual measurements that the DUT successfully
performs (independently of their accuracy).
Figure 1: Three examples of the daily-life situations with the wrist
below the heart level. The difference of height of 1 cm corresponds to
a BP difference of 0.75mmHg, therefore the BP overestimation can be
as high as 22 mmHg in the examples shown.

III. EXAMPLE OF AUTOMATICITY TESTING USING THE AKTIIA
AUTOMATED AND CUFFLESS BRACELET
A. Context
Aktiia SA is developing a fully automated wrist cuffless BP
monitor. The Aktiia Bracelet relies on off-the-shelf optical
sensors that perform green reflective photoplethysmography
(PPG) measurements on the skin vasculature of the wrist.
Figure 2 illustrates the size and shape of the device. From the
recorded optical signals, a library of Optical Blood Pressure
Monitoring (OBPM) algorithms infers systolic BP (SBP) and
diastolic BP (DBP). The technical characteristics of the device
have already been described [13], and several pilot
investigations and clinical trials have been published [14-16].
Because the device is fully automated (no button needs to be
pressed to perform a measurement) and generates no
discomfort to the patient (optical readings), then Aktiia
Bracelet is particularly well-suited for long-term day and night
monitoring of patients in daily-life conditions.
B. Materials and methods
In order to test the full automaticity performance of Aktiia
Bracelet a first pilot study was performed following the
guidance described in this paper.
Ten healthy volunteers (5 males and 5 females) participated
in a pilot study. The anthropological data describing the study
population are reported in Table 1. After giving informed
consent, the reflective PPG signals were recorded at the
the Aktiia Bracelet and the
reference signals were recorded with a volume-clamp device
finger cuff positioned on the central phalange of a middle
finger on the contralateral hand (Nexfin, BMEYE BV, The
Netherlands). The choice of volume-clamp for this pilot study
was made because of its capability to track beat-to-beat BP
changes, and for its previous published results of BP variation
tracking during body position changes [17].
According to the procedures described in Section II, the
following measurement conditions were studied:
- sitting with both arms positioned on the table, at the heart
level
- standing up with both arms positioned on a stable
support, at the heart level
- lying supine
- sitting with both arms positioned on the table, at the heart
level, while performing an isometric leg extension
- sitting with the right arm positioned on the lap, ~20 cm
below the heart level
The PPG data were retrospectively analyzed using the Aktiia
OBPM Library of Algorithms. The assessment of
automaticity performance was implemented by evaluating
whether the Aktiia Algorithm was accurate across different
body positions and during BP changes induced by the
procedures. Accordingly, the acceptance rates from different
scenarios were calculated, and the means and standard
deviations of the error between the Aktiia Algorithm BP
estimates and the reference at each scenario were computed.
C. Results
The acceptance rates of Aktiia Algorithm across the different
body position and maneuvers are depicted in Figure 3. During
the supine position the maximum percentage of
measurements was achieved (for 67% of the available epochs
Aktiia Algorithm could generate BP estimates), and during
sitting-related positions an acceptance rate of ~50% was
achieved. However, during the standing position, the
minimum percentage of measurements was achieved: for only
27% of the available measurements the Aktiia Algorithm
generated BP estimates. The known amount of motion
artifacts during the standing position forces the Aktiia
Algorithm to reject the PPG signals because of low reliability.
The means and standard deviations of the error for DBP and
SBP across different body positions are also provided in
Figure 3.
It is important to note that:
- Concerning the standard deviation of the error for both
DBP and SBP, all body positions and maneuvers except
standing achieved similar performances. In the standing
position the standard deviation of the error for SBP
exceeded 10 mmHg.
- Concerning the mean of the error for both SBP and DBP,
all body positions and maneuvers achieved similar
performances. Important to note here is the non-biasing
results achieved in the arm on lap position. Because in
this position the arm was placed ~30 cm below the heart
level one would expect a systematic and significant bias
of ~20 mmHg
Figure 2.: Illustration of the Aktiia Bracelet: an Automated Cuffless BP
monitor at the wrist produced by Aktiia SA.
Table 1: Statistics on study population and achieved BP changes
during the different interventions.
Age
(years)
BMI
(kg/m2) (mmHg) (mmHg)
Mean 31,18 22,62 29,15 19,13
SD 7,40 2,47 11,38 6,73
Range 25 7,93 38,30 19,52

D. Discussion
The results of this pilot study indicate that the Aktiia
Algorithm is capable of generating accurate BP estimates, in
particular during supine posture (67% of the time) and sitting-
related positions (~50% of the time). As expected, the Aktiia
Algorithm is only capable of generating BP estimates in the
standing position a reduced percentage of time (~25%). The
results also indicate that the Aktiia Algorithm is not affected
by hydrostatic pressure perturbations induced by changes in
arm position with respect to the heart level, with no significant
bias observed when the measuring arm is lowered by ~20 cm
below the heart level.
IV. CONCLUSION
In this work, we propose an experimental protocol for cuffless
BP monitors at the wrist that addresses the challenges of the
ambulatory environment in a controlled experimental setting.
The protocol includes a BP change intervention, a body
posture change intervention, and a hydrostatic bias
intervention. In addition to reporting the accuracy metrics, it
is important to report the acceptance rate of the device for
each measuring condition. An example of the protocol
implementation is provided by testing the performance of the
Aktiia Bracelet device.
Cuffless wrist-based devices have a great potential to make
ambulatory monitoring more appealing to the patients and
provide the clinicians with a better BP profile on daily,
weekly, and monthly scales. However, as a scientific
community, we must be vigilant to validate these devices
objectively. This protocol provides realistic testing of the
performance of a cuffless wrist-based device and paves the
way to improving the validation of similar devices in the
future.
V. REFERENCES
[1] ESC, ESC/ESH Guidelines for the management of arterial
hypertension. European Heart Journal (2018)
[2] AHA, Guideline for the Prevention, Detection, Evaluation, and
Management of High Blood Pressure in Adults. Hypertension 2017
[3] J. Solà and R. Delgado-Gonzalo, Eds., The Handbook of Cuffless
Blood Pressure Monitoring: A Practical Guide for Clinicians,
Researchers, and Engineers. Springer International Publishing, 2019
[4] M. G. Myers, M. Godwin, M. Dawes, A. Kiss, S. W. Tobe, and J.
Hypertens.
Dallas Tex 1979, vol. 55, no. 2, pp. 195 200, Feb. 2010, doi:
10.1161/HYPERTENSIONAHA.109.141879.
[5] et al.
J. Hypertens., vol. 31, no.
9, pp. 1731 1768, Sep. 2013, doi: 10.1097/HJH.0b013e328363e964.
[6] Y. Yano and K. Ka
Hypertens. Res. Off. J. Jpn.
Soc. Hypertens., vol. 35, no. 7, pp. 695 701, Jul. 2012, doi:
10.1038/hr.2012.26.
[7] -four-hour ambulatory blood pressure
measurement in clinical practice and research: a critical review of a
J. Intern. Med., vol. 269, no. 5,
pp. 478 495, May 2011, doi: 10.1111/j.1365-2796.2011.02356.x.
[8] et al. pertension
recommendations for conventional, ambulatory and home blood
J. Hypertens., vol. 21, no. 5, pp. 821 848,
May 2003, doi: 10.1097/00004872-200305000-00001.
[9] -
press N. Engl. J. Med., vol. 354, no. 22, pp. 2368
2374, Jun. 2006, doi: 10.1056/NEJMra060433.
[10] ISO 81060-2:2018: Non-invasive sphygmomanometers Part 2:
Clinical investigation of intermittent automated measurement type,
ISO 2018
[11] IEEE 1708-2014 - IEEE Standard for Wearable Cuffless Blood
Pressure Measuring Devices, IEEE 2014
[12] ISO/DIS 81060-3: Non-invasive sphygmomanometers Part 3:
Clinical investigation of continuous automated measurement type,
ISO 2019
[13] J. Solà, M. Bertschi, and J. Krauss, Measuring Pressure: introducing
OBPM, the optical revolution for blood presure monitoring. IEEE
Pulse, June 14 2018
[14] J. Solà, A. Vybornova, and S. Fallet et al, Aktiia Bracelet: Monitoring
of Blood Pressure Using Off-The-Shelf Optical Sensors . Proc. IEEE
EMBC2019
[15] C. Pellaton, A.Vybornova,, and S. Fallet et al, Accuracy testing of a
new optical device for noninvasive estimation of systolic and diastolic
blood pressure compared to intra-arterial measurements. Blood
Pressure Monitoring, October 31 2019, doi:
10.1097/MBP.0000000000000421
[16] NIH, Aktiia OPBM Single-center Prospective Observational Study
Against Double Auscultation. ClinicalTrials.gov, NCT04027777
[17] LM. Sipkens, K. Tresked, et al., Application of Nexfin noninvasive
beat-to-beat arterial blood pressure monitoring in autonomic function
testing, Blood Pressure Monitoring, 2011. doi:
10.1097/MBP.0b013e32834b4431
Figure 3: Upper panel, error distribution and acceptance rate (Acc) for
different scenario for DBP. Lower panel, error distribution and
acceptance rate on different scenario for SBP. The shaded red areas
illustrate the hydrostatic bias that one would expect during the
robust to hydrostatic pressure changes (~ +20 mmHg).
Mean of the error for DBP
Standard deviation of the error for DBP
Mean of the error for SBP
Standard deviation of the e rror for SBP