Content uploaded by Josep Solà
Author content
All content in this area was uploaded by Josep Solà on Jun 12, 2017
Content may be subject to copyright.
Solà et al, 2017 - Performance of Systolic Blood Pressure estimation from radial Pulse Arrival Time (PAT) in anesthetized patients 1
Presented at EMBEC2017 - European Medical and Biological Engineering Conference 2017, Tampere (FI), 11-15 June 2017
Performance of Systolic Blood Pressure estimation
from radial Pulse Arrival Time (PAT) in anesthetized patients
J. Solà1, A. Vybornova1, F. Braun1, M. Proença1, R. Delgado-Gonzalo1, D. Ferrario1, C. Verjus1,
M. Bertschi1, N. Pierrel2 and P. Schoettker2
1 Systems Division, CSEM, Neuchâtel, Switzerland
2 Anesthesia Department, CHUV, Lausanne, Switzerland
Abstract— The performance of estimating Systolic Blood
Pressure (SBP) in anesthetized patients via Pulse Arrival Time
(PAT) techniques was studied with respect to the minimum re-
quired time in between two recalibration procedures.
Materials: a clinical trial [NCT02651558] involving 14 pa-
tients was conducted measuring PAT from an ECG and an ar-
terial line inserted into the radial artery.
Methods: comparison of BP estimates from PAT measure-
ments against invasive BP values was performed in terms of
mean error and standard deviation of the error
(AAMI/ANSI/ISO 81060-2), cumulative percentage of readings
falling within 5, 10 and 15 mmHg (BHS criteria), and MAD -
Mean Absolute Difference (IEEE Std 1708). Two calibration
strategies were explored, involving time between recalibration
periods ranging from 10 seconds to 8 minutes.
Results: assuming an affine calibration function between
PAT and SBP, different slope (Mean Slope: -1.45, CI: -1.64 to
-1.27 mmHg/ms) and offset values (Mean Offset: 575, CI: 517 to
633 mmHg) were found in between patients. In addition, given
a patient, affine calibration functions at different anesthesia
phases also showed to be variable. When assessing agreement in
terms of existing international standards it was found that PAT-
based SBP estimates complied with requirements when time be-
tween two calibrations was smaller than 60 seconds.
Conclusions: the use of anesthetic agents compromises the
implementation of PAT-based techniques to estimate SBP.
Keywords— Blood Pressure (BP), Pulse Arrival Time (PAT),
calibration function, anesthetized patients.
I. INTRODUCTION
There is large demand for new techniques to monitor Sys-
tolic Blood Pressure (SBP) in anesthetized patients in a con-
tinuous and non-invasive way. Current anesthesia protocols
require either the use of invasive arterial line monitoring
(providing continuous SBP measurements), or the inflation
of oscillometric cuffs (providing intermittent SBP measure-
ments every 60 seconds). Although controversial, Pulse Ar-
rival Time (PAT) has been suggested in the past to monitor
SBP [1]. The goal of this study is to provide figures of merit
of PAT-based techniques under the effects of anesthesia.
II. MATERIALS
Preliminary data from an on-going clinical trial [Clinical-
Trials.gov NCT02651558] were analysed. Patients were
placed on the operating table and monitoring was carried out
via the Philips monitor IntelliVue MX800. Monitoring con-
sisted of a 3-lead electrocardiogram, pulse oximetry at the
finger of the right hand, and an invasive recording of pressure
signals via a dedicated catheter inserted into the right radial
artery (BD Arterial Cannula 20G/1.1mmx45mm, Becton
Dickinson Infusion Therapy Syst. Inc., Utah USA).
Recording of raw data was performed using iXTrend Ex-
press (Ixellence GmbH, Wildaum, Germany) during 15
minutes of general anesthesia induction phase with target-
control infusion of Propofol as induction agent, and Remifen-
tanyl as opioid. Rocuronium was administered as a paralys-
ing agent before tracheal intubation. Decreases in BP were
managed with boluses of Ephedrin or Neosynephrin to coun-
teract the hemodynamic effect of anesthetic agents, and
adapted to patients physio-pathology and specifics. Data
from first 14 enrolled patients were used for the analysis.
Fig. 1 Correlation plots between PAT and Systolic BP for the 14 re-
cruited subjects: colored dots depict measurement points, and dashed lines
depict their associated affine calibration functions.
2 Solà et al, 2017 - Performance of Systolic Blood Pressure estimation from radial Pulse Arrival Time (PAT) in anesthetized patients
Presented at EMBEC2017 - European Medical and Biological Engineering Conference 2017, Tampere (FI), 11-15 June 2017
III. METHODS
The recorded ECG and arterial line signals were processed
every 10 seconds to calculate two time series. First, a time
series of systolic BP values: average systolic BP value within
each 10 seconds window. Second, a time series of PAT val-
ues: time delay between the ECG R-wave and the onset of
the pressure pulse waveform, calculated via Chiu’s method
[2] on the ensemble averaged arterial pulse (as for [3]).
For each patient, the calculated SBP and PAT time series
were post-processed in order to estimate an affine calibration
function relating SBP to PAT during time windows of differ-
ent lengths (ranging from 20 seconds to the entire recording).
Affine calibration functions of all patients were then com-
bined into a general-population calibration function averag-
ing the coefficients of each patient-dependent calibration.
Finally, for each patient, the PAT time series were trans-
formed into predicted SBP time series (pSBP) by applying
two calibration strategies. A first calibration strategy as-
sumed pSBP to be a constant value during a “time in-between
recalibration” window. The constant value was a cuff meas-
ured SBP value at the beginning of the window. This strategy
mimics a constant calibration strategy, as for a standard cuff-
based monitoring of a patient. A second calibration strategy
calculated pSBP within a window by applying an affine cal-
ibration function to the PAT time series. The calibration
function was assumed to be the general-population calibra-
tion function, with a corrected offset in order to match the
measured SBP value at the beginning of each window. This
strategy mimics a PAT-based estimation of SBP with recali-
brated offset values at every new performed SBP measure-
ment via an oscillometric cuff.
IV. RESULTS
Figure 1 provides a visual example of patient-dependent
calibration functions estimate over entire recordings.
A general-population calibration function was then calcu-
lated, relating PAT and SBP time series in the form of SBP =
Slope ⋅PAT +Offset, with average coefficients:
• Slope: -1.45, 95% CI: -1.64 to -1.28 mmHg/ms
• Offset: 575, 95% CI: 517 to 633 mmHg
Figure 2 provides examples of the temporal evolution of
slope estimates calculated from the PAT and the SBP series
within the analysis windows of 120 s. Dashed lines depict the
patient-dependent overall calibration function estimated over
the entire recording, bold black lines depict a reliable calibra-
tion function estimated at a given analysis window, and red
lines depict an unreliable local calibration function (i.e. asso-
ciated to non-significant correlation between SBP and PAT
time series). Temporal evolution of the slope for these illus-
trative patients demonstrates the large variability of local af-
fine calibration functions during anesthesia, associated to re-
current use of anesthetic agents acting on both arterial
muscular tonus and cardiac contractibility.
Figure 3 provides a performance assessment of the two
calibration strategies according to the figures of merit sug-
gested by the following existing international protocols:
• IEEE Std 1708 [4]: the Mean Absolute Difference
score (MAD).
• AAMI/ANSI/ISO 81060-2 [5]: the mean error and
the standard deviation of the error.
• BHS Criteria [6]: the cumulative percentage of read-
ings falling within 5, 10 and 15 mmHg.
Fig. 2 Temporal evolution of local affine calibration functions between PAT and Systolic BP, with their associated slope coefficients. For a particular
patient, calibration functions were fitted for the entire recording (dashed lines), and within analysis windows of 120 seconds (solid lines).
Solà et al, 2017 - Performance of Systolic Blood Pressure estimation from radial Pulse Arrival Time (PAT) in anesthetized patients 3
Presented at EMBEC2017 - European Medical and Biological Engineering Conference 2017, Tampere (FI), 11-15 June 2017
.
Fig. 3 Performance assessment according to IEEE Std 1708, ISO 81060-2:2013 and BHS Criteria at predicting SBP following two strategies.
Upper panel depicts a standard cuff-based monitoring of a patient. Lower panel depicts PAT-based estimation of SBP with recalibrated offset val-
ues at every performed oscillometric cuff measurement. Mean and 95% CI values are calculated at different time in-between calibration windows.
4 Solà et al, 2017 - Performance of Systolic Blood Pressure estimation from radial Pulse Arrival Time (PAT) in anesthetized patients
Presented at EMBEC2017 - European Medical and Biological Engineering Conference 2017, Tampere (FI), 11-15 June 2017
For the standard cuff-based BP monitoring strategy of a
patient, it is found that requirements of ISO, BHS, and IEEE
are only met if the time between consecutive measurements
is lowered to 60 seconds. After this one-minute threshold, the
performance continuously decreases as time in between
measurements is increased.
For the PAT-based estimation of SBP with recalibrated
offset values, it is found that the same one-minute threshold
applies in order to meet the requirements of ISO, BHS and
IEEE. However, after this one-minute threshold, the perfor-
mance degrades until reaching a plateau around the two-mi-
nute time in between calibrations. This plateau is reached at
approximately:
• IEEE Std 1708, MAD of 11 mmHg.
• ISO 81060-2, mean error of -2.5 mmHg, and stand-
ard deviation of the error of 11 mmHg.
• BHS Criteria, cumulative percentage of readings
falling within 5, 10 and 15 mmHg of respectively of
40%, 60% and 70%.
Figure 4 provides particular examples of SBP prediction
for the two studied strategies, when applying a two-minute
“time in-between recalibration” window. The illustrated pa-
tients correspond to those depicted in Figure 2.
V. DISCUSSION AND CONCLUSIONS
The presented results point at the fact that the use of anes-
thetic agents compromises the implementation of PAT-based
techniques to estimate SBP during anesthesia.
At least during anesthesia induction, the standard use of
anesthetic agents on a patient creates large variability of local
calibration functions. In order to fulfill the requirement of in-
ternational standard for BP measurements, such variability
would force recalibration procedures to be performed at an
interval of 60 seconds. Furthermore, within this 60 s win-
dows, performances of a PAT-based SBP monitoring did not
outperform the current SBP monitoring implemented anes-
thesia induction protocols, which may require, depending on
patients' physiologic status, an oscillometric cuff measure-
ment to be performed every 60 seconds.
The analyzed data provides valuable statistical analysis of
the coefficients of an affine calibration function relating SBP
and PAT time series in anesthetized patients. These results
might be used in the future to support the design of new clin-
ical trials in the field of cuffless blood pressure monitoring.
REFERENCES
1. Mukkamala et al, Toward Ubiquitous Blood Pressure Monitoring
via Pulse Transit Time: Theory and Practice. IEEE Trans Biomed
Eng. 2015 Aug;62(8):1879-901
2. Chiu et al, Determination of pulse wave velocities with comput-
erized algorithms. Am Heart J. 1991 May;121(5):1460-70
3. Sola, Continuous non-invasive blood pressure estimation. ETHZ
Thesis No 20093, 2011.
4. IEEE Std. 1708-2014 - IEEE Standard for Wearable Cuffless
Blood Pressure Measuring Devices.
5. ISO, ANSI/AAMI/ISO 81060-2:2013 Non-invasive sphygmoma-
nometers — Part 2: Clinical investigation of automated measure-
ment type
6. ESH, Working Group on Blood Pressure Monitoring of the Euro-
pean Society of Hypertension International Protocol for valida-
tion of blood pressure measuring devices in adults, Blood Press
Monit. 2002 Feb;7(1):3-17
Author: Josep Solà
Institute: CSEM
Street: Jacquet-Droz, 1
City: CH-2002 Neuchâtel
Country: Switzerland
Email: Josep.Sola@csem.ch
Fig. 4 Three examples of SBP prediction during anesthesia when
applying strategy 1, i.e. standard cuff-based monitoring strategy, and
strategy 2, i.e. PAT-based estimation of SBP with recalibrated offset
values. Time in-between recalibration was set to two minutes.
