Pulse total-hemoglobinometer provides accurate noninvasive

Pulse total-hemoglobinometer provides accurate noninvasive
monitoring
Eisei Noiri, MD, PhD; Naoki Kobayashi; Yoshiaki Takamura; Takehiko Iijima, DDS, DMSc, PhD;
Toshiyuki Takagi, MD; Kent Doi, MD, PhD; Akihide Nakao, MD, PhD; Tokunori Yamamoto, MD, PhD;
Sunao Takeda, PhD; Toshiro Fujita, MD, PhD
Objective: Rapid noninvasive measurement of total hemoglobin
would be extremely useful for various clinical situations. This
study determined the clinical accuracy and utility for a pulse
total-hemoglobinometer using four wavelengths: 660 nm (reduced hemoglobin), 805 nm (isosbestic point), 940 nm (oxygenated hemoglobin), and 1300 nm (water density).
Design: Clinical trial.
Setting: University school of medicine.
Patients: Patients were 122 individuals (age, 18 – 82 yrs; 49.4
ⴞ 16.0 yrs [mean ⴞ SD]), including 71 healthy volunteers, 24
patients undergoing surgery, and 27 patients undergoing hemodialysis.
Interventions: The hemoglobinometer probe, which simultaneously indicated peripheral oxygen saturation, pulse rate, and hemoglobin, was placed on the fingertip similarly to a regular pulse
oximeter. The hemoglobin values were compared with those obtained by the co-oximeter or the sodium lauryl sulfate-methemoglobin method. Those hemoglobin values were assigned to either the
training set or the validation set for statistical evaluation.
Measurements and Main Results: Multiple regression analysis
including the ratio of the pulsatile optical density (⌽ij) derived
P
from the four wavelengths and other factors demonstrated that
the mean value of the normalized pulse wave obtained from the
photodiode at 805 nm (DC805) and the ratios of DC940 and
DC1300 (DC940/DC1300) were the pivotal factors in the hemoglobinometer’s increased accuracy in the clinically useful range. The
coefficient of determination between both methods was r2 ⴝ .81
(p < .0001) in the training set and r2 ⴝ .75 (p < .0001) in the
validation set. When the cutoff value of anemia was set at 10
g/dL, and anemia was defined as <10 g/dL, the respective
sensitivity and specificity of hemoglobinometer values to detect
anemia in intraoperative patients were 84.3% and 84.6% (n ⴝ 20).
Conclusions: The data demonstrated the necessity for consideration of light scattering in red blood cells for pulse-spectrophotometric hemoglobin measurement. This was accomplished with
additional factors, such as DC805 and DC940/DC1300. With these
improvements, the pulse hemoglobinometer provided noninvasive, clinically acceptable measurement of hemoglobin. The pulse
hemoglobinometer is a versatile tool that might be useful for
routine health checkups of neonates and young children, intraoperative monitoring of bleeding, and emergency care. (Crit Care
Med 2005; 33:E2831)
ulse spectrophotometry method
uses particular wavelengths and
computation to measure certain
components in circulating
blood. The peripheral oxygen saturation
monitor is used frequently in the clinical
setting worldwide. Pulse oximetry calcu-
lates the percentage of oxygenated hemoglobin (Hb) from the ratio between the
pulsatile optical density derived from two
wavelengths, where the reduced Hb level
comes from the wavelength of 660 nm and
the oxygenated Hb from the wavelength of
940 nm (1– 4). In addition, pulse oximetry
From the Department of Nephrology & Endocrinology (EN, KD, AN, TF), Center for Dialysis, Apheresis,
and Applied Medicine (EN, TF), University Hospital,
University of Tokyo, Tokyo, Japan; Research and Development Center (NK, YT, ST), Nihon-Kohden Corp.,
Tokyo, Japan; Department of Anesthesiology (TI, TT),
Kyorin University School of Medicine, Tokyo, Japan;
and Department of Urology (TY), Nagoya University,
Nagoya, Japan.
Supported, in part, by grants from the Cell Science
Research Foundation (Osaka, Japan; EN), Fukuda
Foundation for Medical Technology (Tokyo, Japan; EN),
and Foundation for Renal Anemia Therapy (Tokyo,
Japan; EN).
The pulse Hb meter is approved for receiving a
Japanese patent and is patent pending in USP and
DEP, which was conducted by investigators in NihonKohden Corp. (NK, YT, ST).
The first and second authors contributed equally to
this study.
Address requests for reprints to: Eisei Noiri, MD,
PhD, Departments of Nephrology and Endocrinology,
107 Laboratory, University of Tokyo, 7-3-1 Hongo,
Bunkyo, Tokyo. E-mail, [email protected]
Copyright © 2005 by the Society of Critical Care
Medicine and Lippincott Williams & Wilkins
Crit Care Med 2005 Vol. 33, No. 12
DOI: 10.1097/01.CCM.0000190430.96750.51
should be useful for measurement of red
blood cell (RBC) total Hb concentration. The
wavelength of 805 nm is also specific for Hb
and 1300 nm for water density. Therefore, we
attempted to measure Hb at clinical levels by
detecting these four wavelengths with a specific fingertip probe and then calculating Hb
levels using Lambert-Beer’s principle. Noninvasive measurement of Hb is particularly useful in the care of neonates and young children, intraoperative monitoring of bleeding,
ambulatory outpatient care, and emergency
care of victims in war and disaster areas. This
report describes the clinical accuracy and
utility of the noninvasive pulse total-hemoglobinometer (Hb-meter) that we developed
and tested in this study.
PATIENTS AND METHODS
Study Populations. The entire protocol of
this study was explained to all subjects. Their
informed consent was obtained before the ini-
E2831
centration is Cw ⫽ 100 (g/dL). The wavelength of 805 nm is the isosbestic point for
both oxygenated and reduced Hb. Therefore, it
is unaffected by the change of oxygen saturation and is specific for Hb. The wavelength of
1300 nm is specific for water density. The
wavelengths of 660 and 940 nm were added to
the calculation, particularly to minimize the
effect of oxygen saturation to increase the
measurement accuracy. The training set (Table 1) was also used for formulation of the
Hb-meter. The pulsatile components (AC) and
the mean value (DC) of normalized pulse wave
obtained from the photodiode were calculated
separately for each of the four wavelengths.
They further demonstrated the pulsatile component of optical density as
⌬A ⬵ AC/DC
[2]
for each wavelength. The ratio of ⌬A␭ between
two wavelengths (⌽ij) was calculated as
⌽ ij ⬅ ⌬A␭i/⌬A␭j ⬵ 共 AC␭i/DC␭i兲/共 AC␭j/DC␭j兲
[3]
(i, j ⫽ 1ⵑ4; where ␭1 ⫽ 1300, ␭2 ⫽ 940, ␭3 ⫽
805, ␭4 ⫽ 660 nm).
The ratio of the mean value of the current
obtained from the photodiode (⌿ij) was calculated using the following formula
⌿ ij ⬅ DC␭i/DC␭j
Figure 1. Flow chart for pulse-spectrophotometric computation of hemoglobin (tHbp).
tiation of the study, in accordance with the
protocol approved by the Human Study
Committee of Kyorin University. We measured Hb in 122 persons (age, 19 – 82 yrs;
49.4 ⫾ 16.0 yrs [mean ⫾ SD]) using the
pulse Hb-meter, including 71 healthy volunteers, 24 patients under operation, and 27
hemodialysis patients. Blood was drawn simultaneously for the reference Hb value
that was measured using a co-oximeter in
the operating room or by the sodium lauryl
sulfate-Hb method for hemodialysis patients
and a subgroup (n ⫽ 38) of the healthy
volunteers. The sodium lauryl sulfate-Hb
method, which engenders no toxic laboratory wastes, is a highly accurate laboratory
measurement of Hb that is used widely in
Japan as an alternative to the cyanmethemoglobin method (5, 6). Twenty intraoperative
patients were monitored independently using both the co-oximeter and the pulse Hbmeter to examine the sensitivity and specificity of the pulse Hb-meter.
Principle of the Pulse Hb-Meter. The pulsespectrophotometric Hb computation method is
summarized in Figure 1. The pulsatile optical
density (⌬A␭) indicates the change of total optical density of blood components at specific wavelengths (␭) according to the Lambert-Beer law:
Crit Care Med 2005 Vol. 33, No. 12
⌬A␭ ⫽ (Eh␭ ⫻ Ch ⫹ Ew␭ ⫻ Cw ⫻ ⌬l. In that
equation, Eh␭ (dL/g/cm) represents the extinction coefficient of Hb (linear combination of the
extinction coefficients of reduced Hb and oxygenated Hb according to the saturation oxygen);
Ch represents the Hb concentration in the
blood; Ew␭ (1/cm) represents the extinction coefficient of water (dL/g/cm); Cw represents the
water concentration in the blood (g/dL); and ⌬l
represents the change of light path length
aroused by pulsation of blood pressure (cm). The
ratio (⌽) of ⌬A␭ is expressed by two LambertBeer formulas in which ⌽ is mechanically detectable when the blood component is measured
at two wavelengths:
⌬A ␭1
⫽
⌽⬅
⌬A ␭2
共Eh␭1 䡠 Ch ⫹ Ew␭1 䡠 Cw兲 䡠 ⌬l
共Eh␭2 䡠 Ch ⫹ Ew␭2 䡠 Cw兲 䡠 ⌬l
⫽
Ch ⫽
Eh␭1 䡠 共Ch/Cw兲 ⫹ Ew␭1
Eh␭2 䡠 共Ch/Cw兲 ⫹ Ew␭2
Ew␭1 ⫺ ⌽12 䡠 Ew␭2
⫻ Cw关g/dl兴
⌽12 䡠 Eh␭2 ⫺ Eh␭1
[1]
Consequently, the Hb value is obtained
from the pulse Hb-meter when the water con-
[4]
Simple regression analysis was used to detect the respective optimum relations between
16 variables (⌽21, ⌽31, ⌽41, ⌽32, ⌽42, ⌽43,
⌿21, ⌿31, ⌿41, ⌿32, ⌿42, ⌿43, DC␭1, DC␭2,
DC␭3, and DC␭4) and the reference Hb value.
The respective coefficients of determination
were examined for the 16 variables and the
reference Hb value (Table 1 in supplemental
data online). A model to estimate the real Hb
value was subsequently established using multiple regression analysis (see detail in supplemental data online and Table 2 in supplemental data online). The following formula shows
the pulse-spectrophotometric estimation of
the Hb value (tHbp).
tHb p ⫽ a ⫻ ⌽ 41 ⫹ b ⫻ ⌽ 31 ⫹ c ⫻ ⌽ 21 ⫹ d
⫻ ⌿ 21 ⫹ e ⫻ DC ␭3 ⫹ f
[5]
In that equation, the values for a to f were
obtained using multiple regression analysis.
Fingertip Probe. Light-emitting diodes
were obtained from Epitex (Kyoto, Japan). We
obtained InGaAs PIN photodiodes from
Hamamatsu Photonics KK (Shizuoka, Japan).
We made the specific fingertip probe carrying
the four-wavelength light-emitting diode. The
actual monitor is 21 ⫻ 26 ⫻ 5 cm; the indicator also reports values for peripheral oxygen
saturation and pulse wave (Fig. 2).
Statistical Analysis. Sampled data were assigned to the training set and the validation
set to investigate bias and precision between
E2831
Table 1. Sampling numbers
Training set
Validation set
Total
Control
Surg
HD
Total
36
35
71
12
12
24
13
14
27
61
61
122
Control, healthy individuals; Surg, surgical patients; HD, chronic hemodialysis patients. Sampled
data were assigned to the training set and the validation set, alternately.
Table 2. Biases and precision values for respective groups in training and validation sets
tHbp
Training
set
Control
Surg
HD
All
Validation
set
Control
Surg
HD
All
tHbp-tHbs
No.
Mean
Max
Min
Mean
SD
36
12
13
61
14.4
11.1
9.8
12.8
17.0
14.6
12.2
17.0
10.3
7.3
6.8
6.8
⫺0.32
0.00
0.89
0.00
1.06
1.19
1.39
1.24
35
12
14
61
14.3
10.7
10.7
12.8
16.5
12.6
13.0
16.5
12.3
6.0
6.9
6.0
⫺0.57
⫺0.07
0.83
⫺0.16
0.99
1.04
1.65
1.30
tHbp, hemoglobin calculated pulse-spectrophotometrically; tHbs, reference hemoglobin value;
Control, healthy individuals; Surg, surgical patients; HD, chronic hemodialysis patients.
the tHbp and reference Hb values (Tables 1
and 2). The correlation between both values
was analyzed using Pearson’s rank test. The
distribution of the tHbp value was compared
with the reference Hb value using BlandAltman’s method. The sensitivity and specificity for the diagnosis of anemia were calculated
using a cross-table.
RESULTS
The Hb values obtained using Equation 5 (tHbp) were compared with those
of blood samples from study subjects as
determined by the standard Hb measurement using the co-oximeter or sodium
lauryl sulfate-Hb methods (reference
Hb). The range of tHbp in this study was
between 6.8 and 17.0 g/dL in the training
set and 6.0 and 16.5 g/dL in the validation
set. The training set data are plotted in
Figure 3, A and B. The tHbp and reference Hb values were highly correlated (r2
⫽ .81, p ⬍ .0001; Fig. 3A). Their variation
was 0.00 ⫾ 1.24 g/dL (mean ⫾ SD; Fig.
3B).
Next, the validation set (n ⫽ 61) was
used for evaluation of Equation 5, compared with the reference Hb value. Again,
a significant correlation between these
two values was found (r2 ⫽ .75, p ⬍
.0001; Fig. 3C). Bland-Altman plot deCrit Care Med 2005 Vol. 33, No. 12
picts the variation of this second set as
⫺0.16 ⫾ 1.30 g/dL (Fig. 3D). The difference between tHbp and the reference Hb
value was examined in each group of the
training and validation sets. It is summarized in Table 2, which shows that the
larger bias was more pronounced in the
hemodialysis group.
The fingertip probe of the pulse Hbmeter was attached to 20 intraoperative
patients, and the reference Hb was measured using a co-oximeter. The maximum value was 15.8 g/dL in tHbp and
14.1 g/dL in the reference Hb. The minimum value was 4.1 g/dL in tHbp and 4.3
g/dL in the reference Hb; the standard
error was 0.098. The sensitivity to detect
anemia during operation using the pulse
total-hemoglobinometer was 84.3% and
the specificity was 84.6% when the anemia was defined as Hb of ⬍10.0 g/dL. A
representative pulse Hb monitoring was
recorded during an emergency operation
of ectopic pregnancy; it is shown in Figure 4. Both Hb values simultaneously increased when bleeding was stopped during surgery.
DISCUSSION
The pulse Hb-meter’s utilization of
four wavelengths (660, 805, 940, and
Figure 2. Photo and display of pulse hemoglobinometer.
1300 nm) and other factors, instead of
only the two wavelengths (660 and 940
nm) used by the oximeter, allows measurement of Hb up to its most clinically
useful level. This fact is convinced by
Figure 5, which portrays the relation between the reference Hb concentration
and the ratio of ⌬A␭ by two wavelengths,
where variables in Equation 1 are assigned respectively as ␭1 ⫽ 805 and ␭2 ⫽
1300. Although these variables show linearity, Lambert-Beer’s two-wavelength
principle is unsuitable for clinical Hb
measurement as is (r2 ⫽ .40). In addition,
the application of the four wavelengths
and their ⌽ij were insufficient to increase
the accuracy (see supplemental Fig. 1)
because the coefficient of determination between the spectrophotometrically estimated Hb and the reference Hb
improved only to r2 ⫽ .51 in the training set and r2 ⫽ .54 in the validation
set, which were not clinically appreciable levels.
Figure 3 shows other variables in
addition to the ⌽ij (Table 1 in supplemental data online) examined in multivariate regression analysis. Addition of
these variables to the model further improved and achieved the accuracy of
formula (2). This improved accuracy is
derived from the offset effect of the
subtle changes in absorbance that are
presumably related to RBC light scattering. In other words, DC805 and
E2831
T
he pulse hemoglobinometer is a versatile tool that
might be useful for routine
health checkups of neonates
and young children, intraoperative monitoring of bleeding, and emergency care.
Figure 3. Correlation and discrepancy between pulse-spectrophotometric computation of hemoglobin
(tHbp) and reference hemoglobin values (tHbs). A, The mutual correlation of the methods was
examined in the training set and found to be significant (r2 ⫽ .81, p ⬍ .0001). B, difference between
tHbp and tHbs value was plotted in the training set. Mean difference and SD were 0.00 ⫾ 1.24 g/dL.
C, The mutual correlation of the methods was further examined in the validation set and was
confirmed as similarly significant to that of the training set (r2 ⫽ .75, p ⬍ .0001). D, the difference
between tHbp and tHbs was plotted in the validation set. The mean difference and SD were ⫺0.16 ⫾
1.30 g/dL.
Figure 4. Monitoring of a clinical case. Anemia
caused by ectopic pregnancy was monitored, and
both pulse-spectrophotometric computation of
hemoglobin (tHbp; pulse Hb-meter) and the reference hemoglobin (Hb) value were plotted chronologically. The anemia improved during surgery.
DC940/DC1300 in Equation 2 are able
to handle the information about tissue
blood volume in addition to arterial
blood and thickness of the blood layer,
which allows the calculation of the RBC
scattering coefficient depending on the
site-specific blood thickness.
Nonetheless, we found maximally 10%
bias in tHbp compared with the reference
Crit Care Med 2005 Vol. 33, No. 12
Figure 5. Relation between ratio (⌽) of pulsatile
optical density and reference Hb value (tHbs).
The line with black circles is derived from the
Lambert-Beer principle using two wavelengths;
the simple line shows the actual correlation of
both values.
Hb value. This bias is comparable to the
difference between venous and arterial
blood in terms of Hb, which can be as
much as a 10% difference depending on the
vascular bed served and on the ratio between oxygenated and reduced Hb. The ref-
erence Hb value in our study was a mix of
venous and/or arterial sampling, which induced an approximately 10% difference between tHbp and the reference Hb value. In
addition, the larger bias was more pronounced in the hemodialysis group.
Hemodialysis patients are apt to suffer with microcytic hypochromic renal
anemia despite the administration of
erythropoietin. In addition, they have a
wide range of hemoglobin levels based
on their bone marrow response to
erythropoietin therapy. Because the
mean corpuscular volume of RBC directly affects the scattering coefficient
level, and because it is slightly larger in
diabetics than in nondiabetics because
of the higher osmolarity attributable to
hyperglycemia, it is conceivable that
these pathophysiologic conditions contributed to the increased bias in the
current investigation. These circumstances also affected both the sensitivity
and specificity in our study. Nonetheless, the Hb-meter would provide a useful Hb level for screening purposes in
widely various clinical settings. In addition, it can detect anemic conditions
with sensitivity of 84.3% and specificity
of 84.6%; it is capable of providing an
alert for bleeding during surgery. Moreover, tHbp can also noninvasively monitor the cessation of a bleeding site during surgery (Fig. 4).
Although we experienced no problems
during measurement, general concerns
related to pulse spectrophotometry, such
as noise related to body motion and low
signals from patients with insufficient peripheral circulation, remain a practical
concern with the use of the Hb-meter.
Substantial issues related to troubleshooting in the pulse spectrophotometry
method must be considered for smoke
victims with suspected carbon monoxide
E2831
inhalation and for methemoglobinemia in
neonates (7). Whereas the ratio of fetal hemoglobin (Hb-F) to Hb is almost 50% in
neonatal Hb, the difference in the spectrophotometric curve of Hb-F compared with
that of Hb is maximally 0.41% under oxygen saturation of 100% and 1.12% under
50% (8). These differences are virtually
negligible for neonates, in view of the accuracy of the pulse spectrophotometry
method, although it is necessary to shield
the probe from light effects during phototherapy for neonatal jaundice.
CONCLUSIONS
The pulse Hb-meter provided accurate
noninvasive blood monitoring. In addition to
its use in surgical patients, it also promises to
be useful for the routine care of neonates and
younger children, from whom it is difficult to
take blood for screening purposes, at blood
donation sites, and in the emergency care of
victims in war and disaster areas. In a hospital
setting, the considerable time savings would
benefit patient care; the cost savings would
benefit both hospital management and patients’ insurance providers. The findings of
our study demonstrate that pulse spectrophotometry for Hb measurement is a valuable
clinical tool. Its full practical use remains to
be further defined.
REFERENCES
1. Aoyagi T, Kishi M, Yamaguchi K: Improvement of earpiece oximeter. Jpn Soc Med Biol
Eng 1974; 12:S90 –S91
2. Steinke JM, Shepherd AP: Role of in whole
blood oximetry. IEEE Trans Biomed Eng
1986; 33:294 –301
3. Aoyagi T, Miyasaka K: The theory and applications of pulse spectrophotometry. Anesth
Analg 2002; 94:S93–S95
4. Wahr JA, Tremper KK, Samra S, et al: Nearinfrared spectroscopy: Theory and applications.
J Cardiothorac Vasc Anesth 1996; 10:406 – 418
5. Gamperling NM: Evaluation and comparison
of the Sysmex NE-1500™ and automated hematology analyzer with the Sysmex NE8000™. Sysmex J Int 1992; 2:26 – 43
6. Karsan A, MacLaren I, Conn D, et al: An eval-
Crit Care Med 2005 Vol. 33, No. 12
uation of hemoglobin determination using sodium lauryl sulfate. Am J Clin Pathol 1993;
100:123–126
7. Mendelson Y, Kent JC: Variations in optical
absorption spectra of adult and fetal hemoglobins and its effect on pulse oximetry. IEEE
Trans Biomed Eng 1989; 36:844 – 848
8. Pologe JA, Raley DM: Effect of fetal hemoglobin on pulse oximetry. J Perinatol 1987;
7:324 –326
SUPPLEMENTAL ONLINE DATA
Regression Analysis for
Calculation of Equation 5
Simple regression analysis was used to
detect the respective optimum relations
Online Table 1. Coefficients of determination between the calculated 16 variables and the reference hemoglobin values
Coefficient of
Determination
Variables
⌽41
⌽31
⌽21
⌿21
DC␭3
⌿31
DC␭2
⌿41
DC␭4
⌿42
⌿43
⌽32
⌽42
⌽43
⌿32
DC␭1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
0.707
0.646
0.626
⫺0.603
⫺0.566
⫺0.541
⫺0.530
⫺0.429
⫺0.420
0.399
0.392
0.361
0.321
0.242
0.210
⫺0.162
Variables are shown in the order of larger
absolute correlation coefficient.
between 16 variables (⌽21, ⌽31, ⌽41, ⌽32,
⌽42, ⌽43, ⌿21, ⌿31, ⌿41, ⌿32, ⌿42, ⌿43,
DC␭1, DC␭2, DC␭3, and DC␭4) and the reference hemoglobin (Hb) value. The respective correlation coefficients were
determined between each of 16 variables and the reference Hb value (Online Table 1). The model to estimate the
actual Hb value was finally established
by multiple regression analysis (Online
Table 2) assigning the reference Hb
value of the training set as subordinate
variables and the previously calculated
variables as dependent variables, where
eight patterns of eight independent
variables (m) were chosen from the
larger order of the correlation coefficient. Akaike’s information criterion
(AIC) was computed, and the minimum
AIC model was considered the optimum.
AIC ⫽ n ⫻ log共DEVSQ/n兲 ⫹ 2 ⫻ 共m
⫹ 1兲
DEVSQ ⫽
[Online 1]
冘共 x ⫺ x៮ 兲
2
[Online 2]
In those equations, x is the difference
between the spectrophotometrically estimated Hb and the reference Hb value.
The residual sum of squares (DEVSQ) directed to each variable coincides with the
AIC (Online Table 2). Therefore, m ⫽ 5
was ultimately chosen for the optimum
model. Equation 5 in the text was finally
obtained for the pulse-spectrophotometric estimation of the Hb value (tHbp).
Online Table 2. Residual sum of squares (DEVSQ) and Akaike’s information criterion (AIC)
m
1
2
3
4
5
6
7
8
DEVSQ
AIC
244
40.7
241
42.4
240
44.3
144
32.8
92
23.0
91
24.6
91
26.6
88
27.8
m, independent values in each model.
E2831
Online Figure. Correlation and discrepancy between the pulse spectrophotometrically estimated hemoglobin
(tHbp) and referential hemoglobin values (tHbs). Values of tHbp were calculated from the ratio of the pulsatile
optical density (⌽41, ⌽31, ⌽21) obtained from four wavelengths. Therefore, the formula for tHbp in this case is
specified as follows:
tHB p ⫽ a ⫻ ⌽ 41 ⫹ b ⫻ ⌽ 31 ⫹ c ⫻ ⌽ 21 ⫹ f
A, correlation of both methods was examined in the training set (r2 ⫽ .51, p ⬍ .0001). B, correlation
of both methods was further examined in the validation set; the coefficient of determination was found
to be similar to that of the trainingset (r2 ⫽ .54, p ⬍ .0001).
Crit Care Med 2005 Vol. 33, No. 12
E2831