Download Hemodynamic sensing using subcutaneous photoplethysmography

Am J Physiol Heart Circ Physiol 295: H2560 –H2572, 2008.
First published October 10, 2008; doi:10.1152/ajpheart.00574.2008.
Innovative Methodology
Hemodynamic sensing using subcutaneous photoplethysmography
Robert G. Turcott and Todd J. Pavek
St. Jude Medical, Sylmar, California
Submitted 30 May 2008; accepted in final form 29 September 2008
THE USE OF CHRONICALLY IMPLANTED, microprocessor-based cardiovascular devices has accelerated in recent years due to
expanding indications of conventional therapies, such as those
provided by automatic defibrillators (18), as well as the development of new therapeutic and diagnostic technologies that
have broadened the range of clinical application. Examples of
the latter include biventricular pacing (8, 20, 36), hemodynamic monitoring (2, 38), and electrocardiographic monitoring
using implantable loop recorders (7). The increasing role of
implantable devices and a desire for improved device performance are driving a need for increasingly sophisticated physiological sensing. The ideal sensor would enable a number of
distinct applications including disease monitoring and optimization of device function and delivered therapy. It would have
minimal impact on the longevity and volume of the device and
would be integrated into the device itself, rather than requiring
intravascular placement, which increases the cost and risk of
complications.
Photoplethysmography (PPG) uses light to noninvasively
detect changes in microvascular blood volume (12). It is an
attractive sensing technology for use in implanted devices
because it functions from a location outside the bloodstream,
can be implemented in embodiments that consume little power
and volume, requires no moving parts, and has the potential to
support a variety of diagnostic and therapeutic applications.
PPG has been applied to a number of clinical problems,
including noninvasive monitoring of arterial oxygenation as
the enabling technology of the pulse oximeter (30, 37, 40),
beat-to-beat blood pressure measurement (5, 28, 41, 42), atrioventricular (AV) pacing delay optimization (9, 14, 15, 27), and
blood flow monitoring after free tissue transfer (26). The
mechanism of PPG and its clinical applications have recently
been extensively reviewed (3, 22, 39). Despite its long history,
PPG has not previously been used in a subcutaneous location.
Many hemodynamic monitoring goals could be realized with
the knowledge of acute changes in arterial blood pressure.
While it is not presently feasible for chronically implanted
devices to acquire this information, the ability of PPG to detect
acute changes in blood volume raises the possibility that it
serves as a surrogate of pressure in hemodynamic sensing
applications. In general, the relationship between vascular
pressure and volume is nonlinear (19) and is complicated by
time-dependent changes in vasomotor tone (32). As described
in the APPENDIX, however, there is a mathematical basis for the
hypothesis that for small changes in pressure, changes in the
PPG waveform are directly proportional to changes in mean
arterial pressure (MAP) over short timescales.
Two particular applications of hemodynamic sensing in
pacemakers and implantable defibrillators are pacing interval
optimization, in which AV and interventricular pacing intervals are tailored to optimize cardiac function, and targeted
antiarrhythmia therapy, in which the aggressiveness of the
therapy is determined by the hemodynamic consequences of
the arrhythmia. A subcutaneous PPG sensor was developed to
test these brady- and tachycardia hemodynamic sensing applications as well as the linearity of induced changes in MAP and
peripheral vascular volume. Aortic pressure (AoP) and subcutaneous PPG signals were simultaneously recorded during brief
episodes of rapid pacing in healthy dogs to simulate tachyarrhythmias and during AV delay (AVD) changes in dogs with
induced heart block. Optimum AVDs were estimated from the
changes in pressure and PPG waveforms and tested for concordance.
In summary, the hypotheses of the study are as follows: 1) a
subcutaneous PPG sensor operating in a backscatter configuration can detect changes in vascular volume; 2) rapid pacing
and changes in AVD induce transient changes in both arterial
pressure and PPG waveforms; 3) transient changes in MAP and
PPG are directly proportional; and 4) optimum AVDs estimated from transient changes in MAP and PPG are concordant.
Address for reprint requests and other correspondence: R. Turcott, Stanford
Univ. Medical Center, Division of Cardiovascular Medicine, Falk CVRC; MC
5406, Stanford, CA 94305-5406 (e-mail: [email protected]).
The costs of publication of this article were defrayed in part by the payment
of page charges. The article must therefore be hereby marked “advertisement”
in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
defibrillator; pacemaker; sensor; atrioventricular delay; arrhythmia
H2560
0363-6135/08 $8.00 Copyright © 2008 the American Physiological Society
http://www.ajpheart.org
Downloaded from http://ajpheart.physiology.org/ by 10.220.33.3 on June 15, 2017
Turcott RG, Pavek TJ. Hemodynamic sensing using subcutaneous photoplethysmography. Am J Physiol Heart Circ Physiol
295: H2560 –H2572, 2008. First published October 10, 2008;
doi:10.1152/ajpheart.00574.2008.—Pacemakers and implantable
defibrillators presently operate without access to hemodynamic information. If available, such data would allow tailoring of delivered
therapy according to perfusion status, optimization of device function,
and enhancement of disease monitoring and management. A candidate
method for hemodynamic sensing in these devices is photoplethysmography (PPG), which uses light to noninvasively detect changes in
blood volume. The present study tested the hypotheses that PPG can
function in a subcutaneous location, that the acute changes in blood
volume it detects are directly proportional to changes in arterial
pressure, and that optimum pacing intervals identified by it are
concordant with those determined by arterial pressure. Aortic pressure
and PPG were simultaneously recorded in 10 dogs under general
anesthesia during changes in atrioventricular (AV) delay and bursts of
rapid pacing to simulate tachyarrhythmias. Direct proportionality
between transient changes in pressure and PPG waveforms was tested
using regression analysis. Scatter plots had a linear appearance, with
correlation coefficients of 0.95 (SD 0.03) and 0.72 (SD 0.24) for
rapid-pacing and AV delay protocols, respectively. The data were
well described by a directly proportional relationship. Optimum AV
delays estimated from the induced changes in aortic pressure and PPG
waveforms were concordant. This preliminary canine study demonstrates that PPG can function subcutaneously and that it may serve as
a surrogate for acute changes in arterial pressure.
Innovative Methodology
HEMODYNAMIC SENSING USING PHOTOPLETHYSMOGRAPHY
MATERIALS AND METHODS
Fig. 1. Subcutaneous photoplethysmography (PPG) sensor. A: photograph
showing arrangement of light-emitting diodes (LEDs) and photodiode.
B: functional block diagram. LED control allows continuous or pulsatile
illumination of one or both LEDs. The transimpedance amplifier has 4
selectable gain settings. See text for details. IR, infrared.
AJP-Heart Circ Physiol • VOL
ablation, continuous atrial-triggered right ventricular (RV) pacing was
provided.
Instrumentation and data acquisition. Arterial pressure was recorded using a high-fidelity micromanometer-tipped pressure catheter
(4F, SPC-340; Millar Instruments, TX) placed in the ascending aorta
via a femoral arteriotomy. Active fixation pacemaker leads (model
1388, 1488; St. Jude Medical) were placed in the right atrial (RA)
appendage and the RV apex via jugular venotomy. The PPG sensor
was placed in a subcutaneous pocket in a relatively flat portion of the
neck. This anatomic location, selected for mechanical stability in the
acute preparation, is in contrast to the infraclavicular fossa typically
used in patients receiving pacemakers or implantable defibrillators.
Optical components were oriented in the direction (superficial or
deep) that gave the strongest signal. When no direction yielded a
clearly superior signal, the optics were oriented in the deep direction.
Surface ECG, AoP, RA intracardiac electrogram, and PPG signals
were continuously logged using a 12- or 16-bit analog-to-digital data
acquisition card and custom software written in a C programming
environment (DAQCard-1200 or DAQCard-6036E, and LabWindows; National Instruments, Austin, TX). The sampling rate was 100
Hz (1 animal) or 500 Hz (9 animals).
Rapid pacing protocol. In five dogs with intact conduction systems,
tachyarrhythmias were simulated using rapid pacing delivered for
10 –15 s to the RA or RV (3 dogs), or the RV alone (2 dogs),
alternating with 20-s recovery periods of intrinsic rhythm in which
pacing was withheld. Pacing was provided by a custom system
operating under computer control via a data acquisition/counter-timer
card and real-time software written in a C programming environment
(DAQCard-1200, LabWindows). Each recording comprised ⬃860 s
of data, formed by six repetitions of four rapid pacing cycle lengths
(250, 293, 353, and 444 ms) delivered in random order.
Postacquisition data processing and analysis were performed using
software developed in Matlab (MathWorks, Natick, MA). The average values of the AoP and PPG waveforms were calculated over one
respiratory period immediately preceding the onset of rapid pacing
and over one respiratory period beginning 5 s after the onset of rapid
pacing. The changes in AoP (⌬P) and PPG (⌬v) were determined from
these values. As a control, identical calculations were performed using
data obtained during baseline periods of intrinsic rhythm. Data were
excluded if the PPG offset was unstable, e.g., from hemorrhage into
the pocket or motion artifact.
AVD pacing protocol. RV pacing was triggered by RA-sensed
events using the pacing circuit described in Rapid pacing protocol.
During data collection, baseline pacing periods with an AVD of 90 or
180 ms were delivered for 20 s, alternating with test AVD periods of
5 s, allowing measurement of the acute change (4, 33). A single
recording, representing ⬃400 s of continuously logged data, consisted
of three repetitions of five unique test AVDs (30, 60, 90, 120, and 150
ms) delivered in a random order. Pacing at baseline AVD was
delivered continuously between data acquisitions.
AoP and PPG signals were digitally band-pass filtered using a 30-s
finite impulse-response filter with cutoff frequencies of 0.1 and 1.0
Hz, which allows changes in MAP that evolve over seconds to pass
but blocks lower and higher frequencies, including cardiac pulsations.
After we corrected for the time-lag imposed by digital filtering,
changes in AoP (⌬P) and PPG (⌬v) were calculated as the differences
in the signal averages obtained over one respiratory cycle immediately
preceding and immediately after the AVD change. As a control, ⌬P
and ⌬v were also calculated using data obtained during periods of
baseline pacing in which the AVD was held constant. Data were
excluded from analysis if PVCs or PACs occurred within 10 s before
the AVD change, within one respiratory cycle after the change, or if
the PPG offset was unstable.
The transient changes induced in AoP (⌬P) and PPG (⌬v) by
changes in AVD were plotted against the corresponding AVD for
each data set. Third-degree polynomials were fit to the data, with the
295 • DECEMBER 2008 •
www.ajpheart.org
Downloaded from http://ajpheart.physiology.org/ by 10.220.33.3 on June 15, 2017
PPG sensor. A custom PPG sensor was developed for subcutaneous placement. The sensor, shown in Fig. 1, contains red and infrared
(IR) light emitting diodes (LEDs; measured wavelengths 653 and 941
nm, respectively), a photodiode for detection of backscattered light,
and associated circuitry for LED current control, signal conditioning,
and gain control. The electrical components were encapsulated in
epoxy, and the junction between the silicone-jacketed cable and epoxy
was sealed with medical adhesive. The LEDs were surrounded by
titanium cylinders that extend from circuit board to the surface of the
epoxy to prevent direct transmission of light between LED and
photodiode and to maximize the amount of optical power delivered to
the overlying tissue. The photodiode was placed adjacent to the LEDs
and oriented to receive the backscattered light, similar to the arrangement used for fetal and esophageal pulse oximetry (13, 23). The
photodiode was configured in photovoltaic mode and feeds a transimpedance amplifier that provides conversion of photodiode current to
voltage with a low-pass cutoff frequency of 60 kHz. The data
presented here were acquired with the IR LED driven continuously at
20 mA and a gain of 0.25 V/␮A.
Subjects and surgical preparation. Data were acquired from 10
mongrel canines in accordance with the Animal Welfare Act, with
protocols approved by the Institutional Animal Care and Use Committee of St. Jude Medical. Dogs of either sex (20 –30 kg) were fasted
overnight. They were sedated with acepromazine (15 mg sc) and
induced with thiopental sodium (15–22 mg/kg iv). Isoflurane (0.8 –
2.5%) was used for maintenance with positive pressure ventilation
(tidal volume 10 –15 ml/kg, 100% oxygen). AV node/His ablation was
performed using a radio frequency ablation catheter (7F VascoFlector,
VascoLator; VascoMed, Germany) in five animals to achieve complete heart block and allow pacemaker control of the AVD. After
H2561
Innovative Methodology
H2562
HEMODYNAMIC SENSING USING PHOTOPLETHYSMOGRAPHY
a direct proportionality, then while ␴ˆ e2 will by definition be greater than ␴ˆ e1,
it should be similar in magnitude. We specifically require ␴ˆ e2/␴ˆ e1 ⱕ 2 to
conclude that the data are compatible with a directly proportional relationship.
To summarize, to conclude that a data set is consistent with a
directly proportional relationship between ⌬v and ⌬P, we require the
following: 1) the qualitative appearance of scatter and residuals plots
is consistent with a linear relationship; 2) the lower bound of the 95%
CI of r is ⱖ0.7; and 3) residuals of the two predictive models are
similar in magnitude, i.e., ␴ˆ e2/␴ˆ e1 ⱕ 2.
Signal amplitudes. The pacing-induced changes in the AoP and
PPG waveforms were compared with the amplitudes of other signal
components. The mean pressure, PPG offset, and peak-to-peak amplitudes of respiration and pulse were measured directly from the
unprocessed recordings using all data sets, and the amplitudes of the
changes induced by rapid pacing and AVD change were characterized
by the estimated SDs ␴ˆ ⌬P and ␴ˆ ⌬v of ⌬P and ⌬v associated with the
pacing maneuver over individual data sets. The averages and SDs of
these measures were then calculated across all data sets.
RESULTS
Rapid pacing protocol. The rapid pacing protocol was conducted in five animals. An example of signals recorded during
an episode of rapid RV pacing using cycle length 250 ms is
presented in Fig. 2. As seen in the middle trace, a rapid drop in
arterial average and pulse pressures occurs with the onset of
pacing. A corresponding reduction in average and pulse peripheral blood volume is seen in the PPG waveform presented
Fig. 2. Effects of rapid right ventricular (RV) pacing (cycle length 250 ms) on aortic pressure (AoP) and PPG waveforms. The onset of rapid pacing induces
a sudden drop in both average and pulse pressures (middle trace), which are reflected in the PPG signal (bottom trace). Horizontal lines in the AoP and PPG
waveforms indicate the regions over which the average values were calculated, and their vertical offsets correspond to the calculated average. The resulting
change was ⌬P ⫽ ⫺10.8 mmHg for pressure and ⌬v ⫽ ⫺4.7 mV for PPG.
AJP-Heart Circ Physiol • VOL
295 • DECEMBER 2008 •
www.ajpheart.org
Downloaded from http://ajpheart.physiology.org/ by 10.220.33.3 on June 15, 2017
locations of the maxima of the polynomials taken as the estimated
optimum AVD. Optima derived from PPG and AoP were compared.
Statistical analysis. Our hypothesis is that transient changes in AoP
and PPG waveforms are not just linearly related but in fact are directly
proportional. This was tested using regression analysis, with ⌬P
treated as the response variable and ⌬v as the predictor variable. The best
fitting (least-squares sense) parameters of two linear models were
obtained for each data set. In one model, two unconstrained parameters were used, yielding the line of best fit f1(⌬v) ⫽ m̂1⌬v ⫹ b̂; in the
other model the intercept b̂ was fixed at zero, and the best-fitting direct
proportionality f2(⌬v) ⬅ m̂2⌬v was obtained. These linear relationships were included on scatter plots of ⌬P vs. ⌬v, which were
qualitatively inspected along with residual plot e1 ⫽ ⌬P ⫺ f1(⌬v).
Each data set was dichotomized according to the appearance of these
plots. A set was deemed to be qualitatively linear if the scatter plot
unambiguously suggested a linear relationship and, allowing for
statistical fluctuation, no significant trend in data points away from the
line of best fit or in residuals away from e1 ⫽ 0 was apparent. If the
plots failed to show an unambiguous dependence of ⌬P on ⌬v, or if
the relationship was clearly nonlinear, the data set was deemed to be
qualitatively not linear.
The estimates of the marginal SDs ␴ˆ ⌬P and ␴ˆ ⌬v were used to characterize
the variability of the pacing-induced changes in the signals. Linearity was
quantitatively evaluated using the Pearson correlation coefficient r. To conclude that a direct proportionality describes the relationship, we require as a
necessary condition that r is large, with a lower confidence bound of its 95%
confidence interval (CI) ⱖ0.7. Finally, residual SEs ␴ˆ e1 and ␴ˆ e2 were
obtained. These are mathematically equivalent to estimates of the SDs of
residuals using f1(⌬v) ⫽ m̂1⌬v ⫹ b̂ and f2(⌬v) ⫽ m̂2⌬v, respectively, as
predictive models. If the underlying relationship between the variables is truly
Innovative Methodology
HEMODYNAMIC SENSING USING PHOTOPLETHYSMOGRAPHY
AJP-Heart Circ Physiol • VOL
Fig. 3. Scatter plot showing the relationship between the changes in average
AoP ⌬P and PPG ⌬v induced by rapid pacing. A: collection of ⌬P obtained
from all pacing episodes of a single recording is plotted against the corresponding values of ⌬v (F), along with the line of best fit (solid line) and
best-fitting line of direct proportionality (dotted line). Qualitatively, the data
exhibit a strong linear relationship, an observation supported by the large
correlation coefficient of r ⫽ 0.97 (95% CI: 0.95– 0.98). No significant trends
in the data points away from the line of best fit are apparent. The line of best
fit is offset slightly relative to the best-fitting proportionality, but the data are
well described by both models. B: plot of residuals. Points are distributed about
e1 ⫽ 0 mmHg with no significant trends.
the SD of its residuals, was essentially equal to that of the
best-fitting line, yielding a ratio of ␴ˆ e2 to ␴ˆ e1 that was ⱕ1.4 in
12 of the 13 data sets and ⬍2.0 in all.
All but one recording in the rapid pacing protocol met the
three criteria we used to conclude the data are consistent with
a directly proportional relationship between ⌬P and ⌬v: scatter
and residual plots that are qualitatively linear, lower bound of
the 95% CI of r that is ⱖ0.7, and a ratio of residual SEs that
is ⱕ2.0. If analysis is limited to 兩⌬P兩 ⱕ 10 mmHg, then all data
sets would satisfy the requirements for direct proportionality.
295 • DECEMBER 2008 •
www.ajpheart.org
Downloaded from http://ajpheart.physiology.org/ by 10.220.33.3 on June 15, 2017
in the bottom trace. The vertical offsets of the horizontal lines
in the middle and bottom traces indicate the averages of the
AoP and PPG waveforms calculated before and after the onset
of rapid pacing, and their horizontal placements indicate the
time periods over which the averages were obtained. In this
example, the change in AoP was ⌬P ⫽ ⫺10.8 mmHg and the
change in PPG was ⌬v ⫽ ⫺4.7 mV.
Figure 3 presents an example of the scatter and residual plots
used to qualitatively examine the relationship between the
average changes in AoP and PPG induced by rapid pacing. In
Fig. 3A, the collection of ⌬P obtained from all rapid pacing
episodes of a single recording (dog 3, set 2) is plotted against
the corresponding values of ⌬v. The line of best fit f1(⌬v) ⫽
m̂1⌬v ⫹ b̂ is included in Fig. 3 (solid line), along with the
best-fitting direct proportionality f2(⌬v) ⫽ m̂2⌬v (dotted line).
The data exhibit a strong linear relationship, with the points
tightly clustered about f1 and no significant trend away from it,
an observation supported by the appearance of Fig. 3B, in
which the residuals are seen to fall in a grossly uniform way
about e1 ⫽ 0. The data are described by a large correlation
coefficient, r ⫽ 0.97 (95% CI: 0.95– 0.98), and while the
best-fitting direct proportionality f2 is clearly displaced relative
to the line of best fit f1, the additional residual SE associated
with it is on the order of that associated with f1, giving ␴ˆ e2/␴ˆ e1 ⫽ 1.4.
Thus this data set satisfies the three criteria described in
MATERIALS AND METHODS that we used to test for a directly
proportional relationship between ⌬v and ⌬P. Of note, the two
shortest RV cycle lengths resulted in the two tightly clustered
groups of three points each at ⌬v ⫽ ⫺15 and ⫺5 mV. The
displacement of these clusters away from the other 42 data
points resulted in long moment arms and the potential to
overestimate r. However, that these points are tightly clustered
and fall in line with the axis of the remaining 42 points
suggests that they are not outliers artificially raising the value
of r. Indeed, when excluded from analysis the correlation
remains strong, with r ⫽ 0.91 (95% CI: 0.83– 0.94). Nevertheless, a better estimation of the true value of r would require
more complete sampling across the range of ⌬v.
The results for the ensemble of data recorded during the rapid
pacing protocol are presented in Table 1. The linearity of the scatter
and residual plots was qualitatively evaluated according to the criteria
described in MATERIALS AND METHODS, with the results shown under
the heading “Linear?” in Table 1. All but one of the 13 data sets met
our definition of qualitative linearity; the exception exhibited trends in
the data points away from the line of best fit at large values of ⌬P, i.e.,
⌬P ⬎ 10 mmHg, although it did appear qualitatively linear for ⌬P
ⱕ 10 mmHg.
The estimated marginal SDs of ␴ˆ ⌬P and ␴ˆ ⌬v, which characterize the magnitude of change induced in the pressure and
PPG signals by rapid pacing, averaged 9.2 mmHg and 5.5 mV,
respectively. Calculated over all data acquired in the study, the
average pressure and the PPG offset were 71.7 mmHg and
3.2 V, respectively, so that the relative change in these
signals with the onset of rapid pacing was 12.8% for
pressure and 0.17% for PPG.
The correlation coefficient r was large over the ensemble of
data sets, averaging 0.95. The correlation coefficients of all 13
data sets exceeded 0.9, and all had 95% CIs with lower bounds
that were ⱖ0.76.
The data were well described by the model of direct proportionality f2. For most data sets, its residual SE, equivalently,
H2563
Innovative Methodology
H2564
HEMODYNAMIC SENSING USING PHOTOPLETHYSMOGRAPHY
Table 1. Transient pressure and photoplethysmography changes during onset of rapid pacing ␴ˆ ⌬P and ␴ˆ ⌬v
Average
SD
Linear?
␴ˆ ⌬P, mmHg
␴ˆ ⌬v, mV
r
␴ˆ e2/␴ˆ e1
48
48
48
⫹
⫹
⫺
7.9
8.9
8.4
8.8
7.2
9.2
0.92
0.93
0.90
0.87
0.88
0.82
0.96
0.96
0.94
1.1
1.0
1.0
46
⫹
9.8
10.5
0.97
0.95
0.98
1.0
48
48
48
48
⫹
⫹
⫹
⫹
6.3
7.0
6.7
7.0
5.7
5.6
5.7
5.7
0.97
0.97
0.98
0.98
0.94
0.95
0.97
0.96
0.98
0.98
0.99
0.99
1.4
1.4
1.3
1.2
26
30
20
⫹
⫹
⫹
11.1
12.2
12.5
1.9
1.6
1.4
0.93
0.92
0.90
0.84
0.84
0.76
0.97
0.96
0.96
1.8
1.2
1.1
30
30
⫹
⫹
10.5
11.6
3.9
4.7
0.99
0.98
0.97
0.96
0.99
0.99
1.4
1.2
9.2
2.2
5.5
2.9
0.95
0.03
0.90
0.07
0.97
0.02
1.2
0.2
95% CI of r
n ⫽ Number of rapid pacing episodes; ⫹(⫺), meets (fails to meet) qualitative criteria for linearity; ␴ˆ ⌬P and ␴ˆ ⌬v, estimated SD of change in pressure and
photoplethysmography induced by rapid pacing, respectively; r ⫽ Pearson correlation coefficient; ␴ˆ e1 and ␴ˆ e2, residual SEs using best-fitting regression line and
best-fitting direct proportionality, respectively.
AVD pacing protocol. The AVD pacing protocol was conducted in five dogs. An example of the waveforms recorded
during a single AVD change from 180 to 90 ms is presented in
Fig. 4. AoP and PPG waveforms before and after digital
band-pass filtering are shown. The horizontal lines in the
filtered signals (Fig. 4, bottom traces) indicate the 5-s respiratory periods over which the pre- and post-AVD-change averages were calculated, and their vertical offsets represent the
calculated averages. Thus both AoP and PPG, on average,
increased after the AVD change shown in Fig. 4 by small but
measurable amounts. In this particular example, ⌬P ⫽ 0.83
mmHg and ⌬v ⫽ 0.14 mV.
Figure 5 presents a typical example of the scatter and
residual plots used to qualitatively examine the relationship
between the changes in AoP and PPG induced by all changes
in AVD in a single recording (dog 8, set 2). In Fig. 5A, the
best-fitting model of a linear transformation, f1(⌬v) ⫽ m̂1⌬v ⫹
b̂, is shown as a solid line, and the best-fitting model of direct
proportionality, f2(⌬v) ⫽ m̂2⌬v, is shown as a dotted line. The
data exhibit a moderately strong linear relationship with the
data points clustered about the line of best fit, although not as
tightly as in the plots obtained from rapid pacing data. The
ability of the two models f1 and f2 to describe the data
presented in Fig. 5A appears to be essentially identical, an
impression confirmed by the ratio of their residual SEs, which
is unity to three significant figures. As shown in Fig. 5B, to a
good approximation the residuals lack significant pattern or
trend away from e1 ⫽ 0 mmHg. The qualitative observations of
linearity are supported quantitatively by its large correlation
coefficient of r ⫽ 0.89 (95% CI: 0.78 – 0.94). This data set thus
meets our criteria for concluding that it exhibits a directly
proportional relationship between ⌬v and ⌬P.
The results for the ensemble of data recorded during the
AVD change protocol are presented in Table 2. Twenty-nine of
thirty-three data sets were qualitatively linear according to the
criteria described in MATERIALS AND METHODS, as noted by “⫹”
AJP-Heart Circ Physiol • VOL
under the heading “Linear?” in Table 2. The exceptions exhibited large residuals and a near-horizontal line of best fit. None
of the data sets had plots that suggested a nonlinear relationship. The estimated marginal SDs ␴ˆ ⌬P and ␴ˆ ⌬v were small,
averaging 0.32 mmHg and 71 ␮V, respectively. Given that the
average pressure and PPG offset were 71.7 mmHg and 3.2 V,
respectively, the relative change in these signals induced by
AVD change was quite small, 0.45% for pressure and 0.0022%
for PPG.
The correlation coefficient r was generally large over the
ensemble of data sets, averaging 0.72, with a 95% CI of
0.51– 0.85, on average. Of the 33 data sets, r exceeded 0.6 in
26 and 0.8 in 18. The lower confidence bound of r was ⱖ0.7
in 14 of 33 data sets and exceeded zero in 31.
The data were well described by the model of direct proportionality f2. For most data sets, its residual SE, equivalently
the SD of its residuals, was essentially equal to that of the
best-fitting line; the ratio ␴ˆ e2/␴ˆ e1 was ⱕ1.31 in all 33 data sets.
In summary, 14 of 33 data sets met our strict criteria for a
directly proportional relationship between ⌬P and ⌬v; the
remaining 19 had a lower confidence bound of r that was less
than the required value of 0.7; scatter plots suggest that this
was due to the effect of statistical variability, rather than
nonlinearity. If we had simply required that the 95% CI of r be
positive, then 29 of the 33 data sets could be considered
consistent with a direct proportionality.
Optimum AVD estimation. An example of the optimum
AVDs estimated from AoP and PPG is shown in Fig. 6 using
the same data presented in Fig. 5A. The maximum of the
best-fitting polynomial occurred at 74 ms for PPG and 68 ms
for AoP. Over the ensemble, the optima ranged between 62 and
143 for PPG and 53 and 129 for pressure, with an average
difference of 3 ⫾ 13 ms. The PPG- and AoP-derived optima
were within 35 ms of one another for all 33 data sets and were
within 15 ms for 26 of the 33 data sets.
295 • DECEMBER 2008 •
www.ajpheart.org
Downloaded from http://ajpheart.physiology.org/ by 10.220.33.3 on June 15, 2017
Dog 1
Set 1
Set 2
Set 3
Dog 2
Set 1
Dog 3
Set 1
Set 2
Set 3
Set 4
Dog 4
Set 1
Set 2
Set 3
Dog 5
Set 1
Set 2
n
Innovative Methodology
HEMODYNAMIC SENSING USING PHOTOPLETHYSMOGRAPHY
H2565
Signal amplitudes. Figure 7 presents estimates of the various
components of the AoP and PPG waveforms. Changes in arterial
pressure induced by rapid pacing ranged around 10 mmHg, falling
between pulse pressure and the variability associated with ventilation. The effect of AVD change on pressure was quite small, on
the order of a few tenths of a millimeter of mercury. The PPG
waveforms are dominated by a large offset, near 3 V for the gain
setting used in the study. All time-varying signal components
were much smaller than this. The effects of ventilation, cardiac
pulsation, and rapid pacing fell in the single-digit millivolt range
(i.e., 10⫺3 to 10⫺2 V), slightly above the 2.44-mV quantization
level of a 12-bit A/D converter. The change in detected light
induced by changes in AVD was miniscule compared with these
other signal components, falling in the tens of microvolt range
(i.e., 10⫺5 to 10⫺4 V), below the quantization level of a 16-bit
A/D converter (0.153 mV, dotted line at bottom) and almost two
orders of magnitude smaller than the effects of ventilation, cardiac
pulse, and rapid pacing.
DISCUSSION
This preliminary proof-of-concept study represents the first use of
PPG in a subcutaneous environment and the first demonstration of a
directly proportional relationship between acute changes in AoP and
AJP-Heart Circ Physiol • VOL
PPG waveforms induced by pacing maneuvers. The ability of PPG to
function subcutaneously and the existence of a direct proportionality
over short timescales suggest that PPG may provide hemodynamic
sensing capability to chronically implanted devices by providing a
surrogate for arterial pressure.
Theory. The correlation between changes in arterial pressure
and the PPG waveform observed in this study is consistent with
volume-dependent light absorption. Since hemoglobin is
strongly absorbing (44), an increase in blood volume increases
tissue photon absorption and results in less optical power
returning to the detector; conversely, a contraction of blood
volume results in an increase in detected optical power. This
interpretation of the dependence of detected light intensity on
volume change is intuitively appealing and consistent with the
observations of this study; however, other possible effects have
been suggested in the literature. The distribution of erythrocytes exhibits flow-dependent anisotropy, which may also
contribute to the observed modulation of detected light, independent of volume (11, 25). In addition, changes in erythrocyte
orientation induced by flow alter both the absorption by the cell
and the degree of scattering from its surface and account for the
modulation of light absorbed by pulsatile blood flowing
through a rigid tube of fixed volume (31). Finally, due to
295 • DECEMBER 2008 •
www.ajpheart.org
Downloaded from http://ajpheart.physiology.org/ by 10.220.33.3 on June 15, 2017
Fig. 4. Effect of atrioventricular delay (AVD) change on AoP and PPG waveforms. The AVD was changed from 180 to 90 ms at the location of the vertical
line in the ECG trace (top trace). AoP and PPG waveforms before (second and third traces) and after (bottom two traces) digital band-pass filtering are shown.
The horizontal lines in the filtered signals (lower two traces) indicate the 5-s respiratory periods over which the pre- and post-AVD-change averages were
calculated, and their vertical offsets are equal to the calculated averages. The changes in pressure and PPG were ⌬P ⫽ 0.83 mmHg and ⌬v ⫽ 0.14 mV,
respectively.
Innovative Methodology
H2566
HEMODYNAMIC SENSING USING PHOTOPLETHYSMOGRAPHY
reflection off its wall, the volume increase of a single macroscopic vessel can result in either an increase or a decrease in
the intensity of backscattered light, depending on the spatial
relationship of the PPG sensor and the vessel (43).
While the present study was not designed to elucidate the
mechanism by which the PPG waveform is modulated, we
favor a model in which vascular volume plays a dominant role.
Not only are the results interpretable purely in terms of a
simple volume-dependent model of light-absorption, but the
other mechanisms demonstrated in the laboratory are less
likely to be relevant in subcutaneous PPG. With respect to the
effects of macroscopic vasculature, we found that the PPG
AJP-Heart Circ Physiol • VOL
295 • DECEMBER 2008 •
www.ajpheart.org
Downloaded from http://ajpheart.physiology.org/ by 10.220.33.3 on June 15, 2017
Fig. 5. Scatter plot showing the relationship between the changes in average
AoP ⌬P and PPG ⌬v induced by all AVD changes of a single recording.
A: collection of ⌬P is plotted against the corresponding values of ⌬v (F), along
with the line of best fit (solid line) and best-fitting direct proportionality (dotted
line). The data exhibit a qualitatively linear relationship. The correlation
coefficient is large, 0.89 (95% CI: 0.78 – 0.94), and the line of best fit is
essentially identical to the best-fitting direct proportionality. B: plot of residuals. Points are distributed about e1 ⫽ 0 mmHg with no significant trends.
sensor did not require proximity to such vessels; indeed, none
were apparent in the subcutaneous pocket in the vicinity of the
optical components. Rather, the sensor only requires vascularized tissue containing capillary networks and their associated
arterioles and venules. Because of their vastly greater number,
these have a total cross-sectional area that dwarfs that of
macroscopic vessels (6), allowing robust arterial and venous
PPG signals to be generated. Furthermore, since the volume of
tissue illuminated by the sensor is much larger than the scale of
the microcirculation architecture, optical effects not related to
volume, e.g., erythrocyte orientation and reflection off moving
vessel walls, are likely minimized by spatial averaging.
While the raw PPG signal reflects changes in the total blood
volume in the region of illuminated tissue, signal processing
can extract changes associated with specific components of the
vasculature. For example, pulse oximetry operates on the pulse
amplitude, which is primarily an arteriolar phenomenon,
thereby allowing arterial oxygenation to be estimated. On the
other hand, low-pass filtering to eliminate cardiac pulsations
preserves the respiratory component, which primarily arises
from postcapillary venules due to modulation of venous return.
The specific vascular component that is being assessed thus
depends on the way the PPG signal is processed.
PPG sensor and signal processing. During data acquisition
the IR LED was driven with a constant current of 20 mA,
which would represent an unacceptably large current drain in a
chronically implanted device. However, in a practical embodiment the average current consumption can be made arbitrarily
small by driving the LED with sufficiently low duty cycle
current pulses. For example, operating the LED using a 1/2,000
duty cycle allows reconstruction of the cardiac pulsations at an
average current drain that is on the order of that consumed by
a functioning pacemaker. Lower duty cycles are readily
achievable with faster electronics.
Two signal processing techniques were used to improve the
ability to detect small changes in AoP and PPG waveforms.
One treated the cardiac pulses as noise and minimized their
effect by band-pass filtering. The other technique was to
perform a pair-wise comparison of data taken at corresponding
points of the respiratory cycle before and after the AVD change
or onset of rapid pacing. This is mathematically equivalent to
computing the difference of averages calculated over identical
phases of the respiratory cycle. Implementing this in a fully
automated implantable device would require detection of respiration, which could be accomplished with intrinsic analysis
of the PPG waveform, through an auxiliary sensing technique
such as thoracic impedance or through detection and analysis
of respiratory effects on the intracardiac electrogram.
Experimental results. The effects of respiration and cardiac
pulsation are clearly present in the subcutaneous PPG waveform. The appearance of systolic pulses demonstrates that the
sensor is responding to changes in microvasculature. Furthermore, because the fundamental physiological measurement in
pulse oximetry is the PPG pulse amplitude, it suggests that
pulse oximetry is possible in a subcutaneous location.
Ventilation modulates arterial and venous blood volumes
through changes in intrathoracic pressure, which influence
preload and afterload on the arterial side and the rate of
peripheral blood return on the venous side. While additional
work is necessary to definitively identify the origin of the
ventilatory component of the PPG signal, the relative ampli-
Innovative Methodology
H2567
HEMODYNAMIC SENSING USING PHOTOPLETHYSMOGRAPHY
Table 2. Transient pressure and photoplethysmography changes during AV delay change
Average
SD
Linear?
␴ˆ ⌬P, mmHg
␴ˆ ⌬v, uV
33
32
31
31
31
32
32
32
⫹
⫹
⫹
⫹
⫹
⫹
⫹
⫹
0.43
0.56
0.47
0.46
0.21
0.26
0.31
0.31
108
111
70
64
66
115
87
81
0.96
0.95
0.92
0.86
0.81
0.89
0.85
0.85
0.92
0.89
0.84
0.73
0.63
0.78
0.71
0.71
0.98
0.97
0.96
0.93
0.90
0.94
0.92
0.93
1.31
1.12
1.17
1.08
1.02
1.01
1.02
1.00
22
30
30
28
⫹
⫹
⫹
⫹
0.41
0.24
0.32
0.24
130
73
92
50
0.93
0.74
0.81
0.85
0.83
0.52
0.63
0.70
0.97
0.87
0.90
0.93
1.17
1.06
1.06
1.00
32
33
32
32
33
33
33
⫹
⫹
⫹
⫹
⫹
⫺
⫺
0.47
0.39
0.20
0.18
0.21
0.27
0.08
89
78
55
45
42
32
27
0.93
0.89
0.86
0.70
0.78
0.47
0.05
0.85
0.78
0.73
0.47
0.60
0.16
⫺0.30
0.96
0.94
0.93
0.85
0.89
0.70
0.39
1.00
1.00
1.02
1.01
1.00
1.00
1.01
24
28
20
29
27
24
27
⫹
⫹
⫹
⫹
⫹
⫹
⫹
0.40
0.40
0.18
0.18
0.31
0.42
0.16
53
36
32
42
90
52
64
0.92
0.81
0.71
0.60
0.88
0.53
0.80
0.83
0.63
0.39
0.30
0.75
0.16
0.61
0.97
0.91
0.88
0.79
0.94
0.77
0.91
1.04
1.21
1.00
1.00
1.01
1.03
1.00
32
32
30
32
24
30
28
⫹
⫹
⫹
⫹
⫺
⫹
⫺
0.22
0.49
0.46
0.26
0.33
0.39
0.42
80
170
61
48
50
35
124
0.38
0.45
0.60
0.57
0.69
0.64
⫺0.02
0.04
0.12
0.30
0.28
0.40
0.36
⫺0.39
0.64
0.69
0.79
0.77
0.86
0.81
0.36
1.09
1.06
1.06
1.00
1.02
1.18
1.03
0.32
0.12
71
33
0.72
0.24
0.51
0.33
0.85
0.15
1.05
0.08
r
␴ˆ e2/␴ˆ e1
95% CI of r
n ⫽ No. of atrioventricular (AV) delay changes per data set; otherwise, same definitions as in Table 1.
tudes of cardiac and ventilatory components of the arterial
pressure (⬃10⫻) and PPG waveforms (⬃1⫻) shown in Fig. 7
suggest that the modulation of venous blood volume is the
dominant factor. In addition, subtle motion of the sensor and
surrounding tissue may also contribute, although efforts were
made to minimize this effect.
The data collected during rapid pacing to simulate tachyarrhythmias strongly support the hypothesis that changes in MAP and
peripheral blood volume as measured by PPG are directly proportional: all scatter and residual plots were consistent with a strongly
linear relationship with one exception in which linearity was limited
to ⌬P mmHg; all had correlation coefficients with 95% CIs above
0.74; and all were well described by a direct proportionality, with
␴ˆ e2/␴ˆ e1 ⱕ. Data obtained during the AVD change protocol are also
consistent with a direct proportionality, though the correlation between pressure and PPG changes was weaker due to the much
smaller magnitude of pressure change (⬃30⫻) in the AVD protocol,
resulting in a larger relative statistical variability in the data.
A direct proportionality between small changes in the pressure and PPG signals is not surprising; indeed, it is a mathematical necessity arising from the fact that variables are related
AJP-Heart Circ Physiol • VOL
by a smoothly varying function. The proof of this is presented
in the APPENDIX. While a direct proportionality is expected for
small changes in pressure, the derivation does not predict the
limits of the linear range. The 13 data sets obtained during the
rapid pacing generally had values of ⌬P that ranged between
⫺20 and 15 mmHg. Only one set exhibited a significant trend
away from linearity, which occurred for large values of ⌬P.
Based on these results, we expect the direct proportionality to
generally hold for ⫺10 ⌬P ⱕ 10 mmHg in this preparation.
The demonstration of direct proportionality between transient
changes in average peripheral blood volume and mean arterial
pressure extends previous work in which changes in peripheral
volume pulse amplitude were correlated with aortic pulse
pressure (9, 27) and stroke volume (15).
The transient response of PPG and AoP to AVD changes
showed a dependence on AVD that is similar to that seen with
conventional optimization techniques: relatively large increases in
pressure and volume occurred at intermediate AVDs near the
expected physiological optimum, while progressively smaller increases occurred as the AVD approached extremely long or short
values. The estimated optima derived from PPG and AoP were
295 • DECEMBER 2008 •
www.ajpheart.org
Downloaded from http://ajpheart.physiology.org/ by 10.220.33.3 on June 15, 2017
Dog 6
Set 1
Set 2
Set 3
Set 4
Set 5
Set 6
Set 7
Set 8
Dog 7
Set 1
Set 2
Set 3
Set 4
Dog 8
Set 1
Set 2
Set 3
Set 4
Set 5
Set 6
Set 7
Dog 9
Set 1
Set 2
Set 3
Set 4
Set 5
Set 6
Set 7
Dog 10
Set 1
Set 2
Set 3
Set 4
Set 5
Set 6
Set 7
n
Innovative Methodology
H2568
HEMODYNAMIC SENSING USING PHOTOPLETHYSMOGRAPHY
concordant. Comparison of these techniques with clinically accepted optimization methods awaits further study.
Clinical applications. Incorporation of a PPG sensor into the
body or header of a pacemaker, implantable cardioverterdefibrillator (ICD), or hemodynamic monitor would avoid the
need for special leads and implant procedures that other approaches to hemodynamic sensing require, e.g., intravascular
pressure measurement (21). The potential clinical applications
of subcutaneous PPG are numerous, and many represent revolutionary changes in diseases management. For example,
since acute changes in arterial blood pressure reflect systemic
hemodynamics, a PPG sensor incorporated into an ICD may
allow therapy to be tailored to the perfusion status of an
arrhythmia. In contrast to the present paradigm, in which ICD
therapy is based exclusively on the electrical characteristics of
the rhythm, this approach would more closely emulate acute
arrhythmia management in the inpatient setting, where determination of the perfusion status of the arrhythmia is of paramount importance and dictates the subsequent management of
the rhythm (1). In particular, a tachyarrhythmia that results in
loss of pulse or a significant reduction in arterial blood pressure
is treated aggressively with a high-voltage defibrillation shock,
while those that are hemodynamically stable are treated less
aggressively with pharmacological interventions or synchronized cardioversion.
AJP-Heart Circ Physiol • VOL
295 • DECEMBER 2008 •
www.ajpheart.org
Downloaded from http://ajpheart.physiology.org/ by 10.220.33.3 on June 15, 2017
Fig. 6. Changes in PPG (top) and average AoP (bottom) plotted against the
AVD that induced the changes. The distribution of points show a dependence
on AVD that is similar to conventional measures used in AVD optimization,
with lower values at extremely long and short AVDs, and larger values in the
neighborhood of the expected physiological optimum. Solid curves represent
the best-fitting third-degree polynomials, with the locations of the maxima
serving as the estimated optimum AVD. PPG and AoP yielded similar results,
74 and 68 ms, respectively.
While ventricular fibrillation is appropriately diagnosed and
treated by ICDs using the electrical activity of the heart, other
rhythms, such as conducted atrial fibrillation and ventricular
tachycardia, may be hemodynamically stable or unstable even
at a moderate heart rate, depending on other factors such as
medications, degree of hydration, and underlying cardiac function. Thus even if the electrical rhythm were correctly diagnosed by the ICD, the hemodynamic consequences are in
general impossible to infer from electrical activity alone. Realtime hemodynamic sensing would allow optimum therapy to
be delivered in these cases. An example of the potential of PPG
to provide hemodynamic sensing is illustrated in Fig. 8, which
shows simultaneously recorded arterial pressure from an indwelling catheter and noninvasive PPG from a conventional
pulse oximeter finger sensor during a variety of ventricular
rhythms in a patient undergoing electrophysiology testing. The
strong similarity between PPG and pressure waveforms illustrates the ability for PPG to track hemodynamic changes
associated with arrhythmias and the potential for it to serve as
a surrogate for pressure in arrhythmia management.
Optimization of AV and interventricular pacing intervals
improves cardiac function acutely (4), and small, randomized
trials suggest it imparts a clinical benefit (29, 35). However,
optimization is infrequently performed because the present
techniques are time consuming, labor intensive, and dependent
on skilled operators (16). If successfully validated, an automatic, pacemaker-based approach to optimization using an
integrated subcutaneous PPG sensor would avoid the drawbacks of conventional methods. In addition it would allow
frequent optimization that could track changes in posture,
volume status, autonomic tone, and disease progression.
Robust detection of the arterial volume pulse by subcutaneous PPG makes subcutaneous pulse oximetry feasible. The
availability of a continuous record of arterial oxygen saturation
would be invaluable in a variety of clinical applications, one of
the most exciting of which is the monitoring, optimization, and
early detection of acute decompensation of chronic diseases,
such as heart failure, asthma, and chronic obstructive pulmonary disease. For example, the direct medical costs of heart
failure are projected to be $32 billion for 2008, of which 60%
will due to hospitalization (34). A modest, early intervention
typically restores the compensated state before inpatient care is
required, thus avoiding significant costs and patient risk. In
addition to disease management, continuous pulse oximetry
may allow the diagnosis of previously unrecognized problems,
such as sleep apnea, and would facilitate the evaluation of
acute illness in a patient with multiple comorbidities, e.g.,
pneumonia in the setting of chronic lung disease and heart
failure. The information provided by pulse oximetry would be
complimented by respiration parameters such as respiratory
rate, tidal volume, and respiratory effort derived from the
venous component of the subcutaneous PPG waveform. Other
vital signs, such as temperature and heart rate, would be
technically easy to obtain and would provide additional benefit
in these applications.
Other roles of PPG-enabled hemodynamic sensing include
automatic determination of the upper and lower pacing rate in
rate-responsive pacemakers, mechanical capture verification,
optimization of electrogram gain and sensing parameters, and
lead integrity validation.
Innovative Methodology
HEMODYNAMIC SENSING USING PHOTOPLETHYSMOGRAPHY
H2569
Limitations
This preliminary, proof-of-concept animal study was conducted with a small number of healthy dogs. The properties of
human subcutaneous and vascular tissue are different from
those of the subjects of this study. The encapsulation that forms
around chronically implanted medical devices may affect the
long-term performance of a subcutaneous PPG sensor and was
not tested in the present study. The acute nature of the prepa-
Fig. 8. ECG, arterial pressure, and PPG
waveforms during clinical electrophysiology
testing. The underlying rhythm varies among
ventricular tachycardia (VT), sinus rhythm
(SR), paced rhythm, and ultimately ventricular fibrillation (VF). The PPG waveform
recorded from a pulse oximeter finger sensor
closely replicates the features of the arterial
pressure recorded from an indwelling radial
artery catheter.
AJP-Heart Circ Physiol • VOL
295 • DECEMBER 2008 •
www.ajpheart.org
Downloaded from http://ajpheart.physiology.org/ by 10.220.33.3 on June 15, 2017
Fig. 7. AoP (left) and PPG (right) signal amplitudes presented as averages ⫾ 1 SD. Average
AoP, PPG offset, and signal components due to
ventilation, cardiac pulses, and changes induced
by the two pacing maneuvers are shown. Horizontal dotted lines indicate quantization levels
of the A/D converters. Signal processing techniques allowed detection of the small signal
transients induce by AVD changes, which for
pressure was a factor of 10 smaller than the
effects of ventilation and for PPG was less than
the resolution of the A/D converter and a factor
of 100 smaller than the effects of ventilation,
cardiac pulse, and rapid pacing.
Innovative Methodology
H2570
HEMODYNAMIC SENSING USING PHOTOPLETHYSMOGRAPHY
APPENDIX
For any two variables related by a smoothly varying function, there
exists a range of values for which, to any prespecified degree of
accuracy, the changes in dependent and independent variables are
directly proportional. The mathematical expression relating the two
variables need not be known. This mathematical fact is well appreciated and ubiquitously exploited in engineering and applied physics
fields; it forms the basis of engineered systems as diverse as aeronautical control systems and high-end audio components. We review the
mathematical proof of this statement after first discussing its content
and applicability to the physiological problem under study.
That a function is “smoothly varying” means that its derivatives
exists at each point. Consider a particular point A on the function and
the tangent defined at that point. The degree to which the tangent
approximates the function at points different from A is quantified as
the difference between the tangent and the function. This difference
converges to zero as one approaches A. Hence, for any prespecified
error threshold, there is a neighborhood of A for which the difference
between the function and tangent is less than the threshold. Thus the
dependent and independent variables are, to a good approximation,
linearly related within the neighborhood and their changes are consequently directly proportional.
We wish to obtain a mathematical expression relating the amount
of detected back-scattered light v to arterial blood pressure P. An
unknown mathematical model describes the influence of the central
arterial blood pressure P on the arteriolar volume V within the
AJP-Heart Circ Physiol • VOL
peripheral tissue sampled by a subcutaneous PPG sensor. Previously
described models (19) do not account for the difference in pressure
between the large arteries and the microvasculature nor do they
include the effects of time-dependent changes, such as those induced
by changes in vasoconstriction and autonomic tone (32). Distinct from
this relationship is the dependence of detected back-scattered light v
on the local arterial volume V. The Beer-Lambert law, which describes the absorption of light transmitted through a nonscattering
medium, does not apply in this setting because the tissue is highly
scattering and the detection is of backscattered rather than transmitted
light (17, 30). Some controversy surrounds the precise mechanism by
which the detected optical power in PPG is modulated. Most authors
view it as primarily due to changes in vascular volume (12, 17, 31),
while others present evidence implicating erythrocyte orientation (30,
31) or reflection off vessel walls (43). Despite the lack of mathematical expressions relating local arteriolar volume to central arterial
pressure and backscattered light to arteriolar volume, and despite the
uncertainty surrounding the precise mechanism by which the latter are
related, the linear relationship referred to above can be expected to
hold. The reason for this is that the requirements of the proof are
modest and apply to the problem under study: that a functional
dependence exists between the amount of backscattered light and
arterial pressure and that it is smoothly varying.
We present the proof here. Consider a generic functional dependence g of the backscattered light v on the arterial pressure P, i.e.,
v共P兲 ⫽ g共P兲.
(A1)
The Taylor-series expansion (24) allows the backscattered light v(P)
at an arbitrary pressure P ⫽ P0 ⫹ ⌬P to be expressed as the sum of
backscattered light at a reference pressure P0, a term proportional to
the difference in pressure from P0, i.e., ⌬P, and a higher-order
remainder term:
v共P 0 ⫹ ⌬P兲 ⫽ g共P0兲 ⫹
冏
⳵g共P兲
⳵P
⌬P ⫹
P⫽P0
冏
⳵2g共P兲
⳵P2
⌬P2
.
P⫽P1 2
(A2)
The first term on the right-hand side of the equation is simply v(P0),
the amount of backscattered light at reference pressure P0. The second
term, proportional to the change ⌬P in pressure from P0, contains the
first partial derivative of g with respect to pressure, evaluated at the
reference point P0. This factor depends only on the functional relationship between v and P and is constant for a fixed P0. Thus the
second term is a constant multiplied by the change in pressure ⌬P. The
final term similarly has a constant factor, the second partial derivative
of g with respect to pressure evaluated at P1, and a factor of ⌬P2. P1
is a value of pressure lying between P and P0 ⫹ ⌬P, which makes the
equality hold. For sufficiently small changes in pressure ⌬P, the
dependence of the final term on ⌬P2 forces the term to become
negligibly small and the expression reduces to
v共P 0 ⫹ ⌬P兲 ⬇ g共P0兲 ⫹
冏
⳵g共P兲
⳵P
⌬P.
(A3)
P⫽P0
Since v(P0) ⫽ g(P0), the change in detected light ⌬v can be expressed as
⌬v ⬅ v共P 0 ⫹ ⌬P兲 ⫺ v共P0兲 ⫽ v共P0 ⫹ ⌬P兲 ⫺ g共P0兲 ⬇
冏
⳵g共P兲
⳵P
⌬P.
P⫽P0
(A4)
Thus for small changes in pressure ⌬P, the change in backscattered
light ⌬v is approximately directly proportional to ⌬P and is given by
⌬v ⬇
The constant of proportionality
295 • DECEMBER 2008 •
冏
⳵g共P兲
⳵P
www.ajpheart.org
P⫽P0
⌬P ⫽ m⌬P.
(A5)
Downloaded from http://ajpheart.physiology.org/ by 10.220.33.3 on June 15, 2017
ration likely resulted in lower quality PPG signals than would
be expected in a chronic implant. Specifically, systemic effects
such as changes in fluid status, autonomic tone, depth of
anesthesia, and hemodynamics likely caused instability in the
measured relationship between and ⌬v and ⌬P over long
timescales, and local effects, such as blood extravasation,
edema, hyperemia, and mechanical motion of the sensor within
the pocket likely affected the relationship over both short and
long timescales. In a chronic setting, changes in autonomic
tone, volume status, and posture may degrade the direct proportionality observed here. A number of cardiovascular conditions may compromise the quality of the PPG signal, such as
aortic stenosis, in which the pulse amplitude is diminished, and
cardiogenic shock, in which elevated systemic vascular resistance reduces the peripheral blood volume. Like all mechanical
sensors, PPG is sensitive to motion artifact, a limitation that
may be partially ameliorated by activating the sensor in the
absence of motion (e.g., with pacing interval optimization),
applying motion-tolerant signal processing techniques (10),
or, for critical therapy as in the setting of arrhythmia management, validating the quality of the PPG signal before basing a
therapy decision on it.
In conclusion, this study is the first demonstration of subcutaneous PPG. Significant changes in both MAP and microvascular volume as measured by a subcutaneously placed PPG
sensor were detectable at the onset of tachyarrhythmias simulated by rapid pacing and during the changes in AV delay.
Regression analysis indicated that the changes were directly
proportional. The changes in PPG and AoP induced by changes
in AV delay yielded concordant estimates of optimum pacing
intervals. Subject to the limitations discussed above, these
preliminary results suggest that subcutaneous PPG may be a
useful surrogate for systemic blood pressure in chronically
implanted devices such as pacemakers, ICDs, and monitors
while avoiding the need for special leads and implant procedures.
Innovative Methodology
HEMODYNAMIC SENSING USING PHOTOPLETHYSMOGRAPHY
m⫽
⳵g共P兲
⳵P
冏
(A6)
P0⫽P0
is fixed for a given reference pressure P0, and the approximation
improves as ⌬P is made small. The definition of “small” depends on
the unknown mathematical expression g, however, the range of ⌬P for
which the approximation holds can be determined by empirically
testing for linear correlation between ⌬v and ⌬P.
In summary, for small changes in arterial pressure ⌬P that occur
over timescales that are shorter than those of the compensatory
feedback mechanisms, we expect directly proportional changes in the
amount of detected light ⌬v independent of the mechanism by which
pressure influences arteriolar blood dynamics and independent of the
precise mathematical relationship between blood dynamics and backscattered light.
We thank Dr. Euan A. Ashley for critical review and suggestions, Barathi
Sethuraman for valuable guidance with the statistical analysis, and Tim
Fayram and the St. Jude Medical Research Group for numerous insightful
discussions.
Present addresses: R. G. Turcott, Stanford University School of Medicine,
Division of Cardiovascular Medicine, Stanford, CA; T. J. Pavek, Center for
Animal Resources and Education, Cornell University, Ithaca, NY.
GRANTS
This work was supported by St. Jude Medical and conducted at its facilities.
DISCLOSURES
The authors were full-time employees of St. Jude Medical at the time this
work was conducted.
REFERENCES
1. American Heart Association. 2005 American Heart Association Guidelines for Cardiopulmonary Resuscitation and Emergency Cardiovascular
Care. Circulation 112: IV1–203, 2005.
2. Adamson PB, Magalski A, Braunschweig F, Bohm M, Reynolds D,
Steinhaus D, Luby A, Linde C, Ryden L, Cremers B, Takle T, Bennett
T. Ongoing right ventricular hemodynamics in heart failure: clinical value
of measurements derived from an implantable monitoring system. J Am
Coll Cardiol 41: 565–571, 2003.
3. Allen J. Photoplethysmography and its application in clinical physiological measurement. Physiol Meas 28: R1–R39, 2007.
4. Auricchio A, Stellbrink C, Block M, Sack S, Vogt J, Bakker P, Klein
H, Kramer A, Ding J, Salo R, Tockman B, Pochet T, Spinelli J. Effect
of pacing chamber and atrioventricular delay on acute systolic function of
paced patients with congestive heart failure. The Pacing Therapies for
Congestive Heart Failure Study Group. The Guidant Congestive Heart
Failure Research Group. Circulation 99: 2993–3001, 1999.
5. Awad AA, Ghobashy MA, Stout RG, Silverman DG, Shelley KH. How
does the plethysmogram derived from the pulse oximeter relate to arterial
blood pressure in coronary artery bypass graft patients? Anesth Analg 93:
1466 –1471, 2001.
6. Berne RM, Levy MN. Cardiovascular Physiology. St. Louis, MI: Mosby
Year Book, 1992.
7. Boersma L, Mont L, Sionis A, Garcia E, Brugada J. Value of the
implantable loop recorder for the management of patients with unexplained syncope. Europace 6: 70 –76, 2004.
8. Bradley DJ. Combining resynchronization and defibrillation therapies for
heart failure. JAMA 289: 2719 –2721, 2003.
9. Butter C, Stellbrink C, Belalcazar A, Villalta D, Schlegl M, Sinha A,
Cuesta F, Reister C. Cardiac resynchronization therapy optimization by
finger plethysmography. Heart Rhythm 1: 568 –575, 2004.
10. Comtois G, Mendelson Y, Ramuka P. A comparative evaluation of
adaptive noise cancellation algorithms for minimizing motion artifacts in
a forehead-mounted wearable pulse oximeter. Conf Proc IEEE Eng Med
Biol Soc 2007: 1528 –1531, 2007.
11. D’Agrosa LS, Hertzman AB. Opacity pulse of individual minute arteries.
J Appl Physiol 23: 613– 620, 1967.
AJP-Heart Circ Physiol • VOL
12. Dorlas JC, Nijboer JA. Photo-electric plethysmography as a monitoring
device in anaesthesia. Application and interpretation. Br J Anaesth 57:
524 –530, 1985.
13. East CE, Chan FY, Colditz PB, Begg LM. Fetal pulse oximetry for fetal
assessment in labour. Cochrane Database Syst Rev: CD004075, 2007.
14. Eliakim M, Sapoznikov D, Weinman J. Assessment of the atrial contribution to cardiac performance by a noninvasive photoplethysmographic
technique. Cardiology 58: 7–13, 1973.
15. Fagrell B, Lindvall K. Non-invasive beat-to-beat analysis of stroke
volume and digital pulse volume in patients with complete heart block and
artificial pacing. Acta Med Scand 205: 185–190, 1979.
16. Flaker G, Weachter R. In cardiac resynchronization therapy: should we
opt for AV optimization? J Cardiovasc Electrophysiol 18: 296 –297, 2007.
17. Flewelling R. Noninvasive optical monitoring. In: Biomedical Engineering Handbook, edited by Bronzino JD. Boca Raton, FL: CRC, 1995, p.
1346 –1356.
18. Goldenberg I, Moss AJ. Implantable device therapy. Prog Cardiovasc
Dis 50: 449 – 474, 2008.
19. Hardy HH, Collins RE. On the pressure-volume relationship in circulatory elements. Med Biol Eng Comput 20: 565–570, 1982.
20. Hare JM. Cardiac-resynchronization therapy for heart failure. N Engl
J Med 346: 1902–1905, 2002.
21. Hegbom F, Hoff PI, Oie B, Folling M, Zeijlemaker V, Lindemans F,
Ohm OJ. RV function in stable and unstable VT: is there a need for
hemodynamic monitoring in future defibrillators? Pacing Clin Electrophysiol 24: 172–182, 2001.
22. Kyriacou PA. Pulse oximetry in the oesophagus. Physiol Meas 27:
R1–R35, 2006.
23. Kyriacou PA, Powell S, Langford RM, Jones DP. Investigation of
oesophageal photoplethysmographic signals and blood oxygen saturation
measurements in cardiothoracic surgery patients. Physiol Meas 23: 533–
545, 2002.
24. Larson RE, Hostetler RP. Calculus. Lexington, MA: D. C. Heath, 1979.
25. Liebman FM, Pearl J, Bagno S. The electrical conductance properties of
blood in motion. Phys Med Biol 7: 177–194, 1962.
26. Machens HG, Mailaender P, Rieck B, Berger A. Techniques of postoperative blood flow monitoring after free tissue transfer: an overview.
Microsurgery 15: 778 –786, 1994.
27. Mest’an M, Kvasnicka J, Rejchrt P. A new method of estimation of the
optimal AV delay by using pulse oximetry in DDD paced patients. Acta
Medica (Hradec Kralove) 41: 135–139, 1998.
28. Molhoek GP, Wesseling KH, Settels JJ, van Vollenhoven E, Weeda
HW, de Wit B, Arntzenius AC. Evaluation of the Penaz servo-plethysmo-manometer for the continuous, non-invasive measurement of
finger blood pressure. Basic Res Cardiol 79: 598 – 609, 1984.
29. Morales MA, Startari U, Panchetti L, Rossi A, Piacenti M. Atrioventricular delay optimization by doppler-derived left ventricular dP/dt improves 6-month outcome of resynchronized patients. Pacing Clin Electrophysiol 29: 564 –568, 2006.
30. Moyle JTB, Hahn CEW, Adams AP. Pulse Oximetry. London, UK:
BMJ Books, 1998.
31. Nijboer JA, Dorlas JC, Mahieu HF. Photoelectric plethysmography–
some fundamental aspects of the reflection and transmission method. Clin
Phys Physiol Meas 2: 205–215, 1981.
32. Pagani M, Lombardi F, Guzzetti S, Rimoldi O, Furlan R, Pizzinelli P,
Sandrone G, Malfatto G, Dell’Orto S, Piccaluga E, Turiel M, Baselli
G, Cerutti S, Malliani A. Power spectral analysis of heart rate and arterial
pressure variabilities as a marker of sympatho-vagal interaction in man
and conscious dog. Circ Res 59: 178 –193, 1986.
33. Pappone C, Rosanio S, Oreto G, Tocchi M, Gulletta S, Salvati A,
Dicandia C, Santinelli V, Mazzone P, Veglia F, Ding J, Sallusti L,
Spinelli J, Vicedomini G. Cardiac pacing in heart failure patients with left
bundle branch block: impact of pacing site for optimizing left ventricular
resynchronization. Ital Heart J 1: 464 – 469, 2000.
34. Rosamond W, Flegal K, Furie K, Go A, Greenlund K, Haase N,
Hailpern SM, Ho M, Howard V, Kissela B, Kittner S, Lloyd-Jones D,
McDermott M, Meigs J, Moy C, Nichol G, O’Donnell C, Roger V,
Sorlie P, Steinberger J, Thom T, Wilson M, Hong Y. Heart disease and
stroke statistics–2008 update: a report from the American Heart Association Statistics Committee and Stroke Statistics Subcommittee. Circulation 117: e25– e146, 2008.
35. Sawhney NS, Waggoner AD, Garhwal S, Chawla MK, Osborn J,
Faddis MN. Randomized prospective trial of atrioventricular delay pro-
295 • DECEMBER 2008 •
www.ajpheart.org
Downloaded from http://ajpheart.physiology.org/ by 10.220.33.3 on June 15, 2017
ACKNOWLEDGMENTS
H2571
Innovative Methodology
H2572
36.
37.
38.
39.
40.
HEMODYNAMIC SENSING USING PHOTOPLETHYSMOGRAPHY
gramming for cardiac resynchronization therapy. Heart Rhythm 1: 562–
567, 2004.
Saxon LA, De Marco T. Cardiac resynchronization: a cornerstone in the
foundation of device therapy for heart failure. J Am Coll Cardiol 38:
1971–1973, 2001.
Severinghaus JW, Honda Y. History of blood gas analysis. VII. Pulse
oximetry. J Clin Monit 3: 135–138, 1987.
Shabetai R. Monitoring heart failure hemodynamics with an implanted device: its
potential to improve outcome. J Am Coll Cardiol 41: 572–573, 2003.
Shelley KH. Photoplethysmography: beyond the calculation of arterial
oxygen saturation and heart rate. Anesth Analg 105: S31–S36, 2007.
Sinex JE. Pulse oximetry: principles and limitations. Am J Emerg Med 17:
59 – 67, 1999.
41. Smith NT, Wesseling KH, de Wit B. Evaluation of two prototype devices
producing noninvasive, pulsatile, calibrated blood pressure measurement
from a finger. J Clin Monit 1: 17–29, 1985.
42. Teng XF, Zhang YT. An evaluation of a PTT-based method for noninvasive and cuffless estimation of arterial blood pressure. Conf Proc IEEE
Eng Med Biol Soc 1: 6049 – 6052, 2006.
43. Weinmann J, Hayat A, Raviv G. Reflection photoplethysmography of
arterial-blood-volume pulses. Med Biol Eng Comput 15: 22–31,
1977.
44. Zijlstra WG, Buursma A, Meeuwsen-van der Roest WP. Absorption
spectra of human fetal and adult oxyhemoglobin, de-oxyhemoglobin,
carboxyhemoglobin, and methemoglobin. Clin Chem 37: 1633–1638,
1991.
Downloaded from http://ajpheart.physiology.org/ by 10.220.33.3 on June 15, 2017
AJP-Heart Circ Physiol • VOL
295 • DECEMBER 2008 •
www.ajpheart.org