ELSEVIER Clinica Chimica Acta 251 (1996) 187-200

Use of an enzyme thermistor for semi-continuous

blood glucose measurements

Thomas Carlsson *a'b, Ulf Adamson a, Per-Eric Lins a,

Bengt Danielsson b
aDepartment of Medicine, Karolinska lnstitutet, Danderyd Hospital, Stockholm, Sweden
bDeoartment of Pure and Applied Biochemistry, University of Lund, Lund, Sweden

;Received 10 July 1995; revised 22 January 1996;accepted 5 February 1996

Abstract

A method to monitor glucose in whole blood is presented. The aim of the project was
to develop a prototype for a bedside monitor system for semi-continuous monitoring of
the blood glucose concentration, requiring only one calibration. This was made possible
by using the special advantage of the thermal sensor technique in combination with the
adjustment of flow. The glucose concentration was determined from the difference between
the sensor response and an estimated background signal. Using standard addition
technique, ~zalibration factors for background and sensitivity were set and remained
unchanged ,during the monitoring. The background signal was 45 ___8 mV (mean + S.D.,
n = 8) and 'the sensitivity was 28 + 1 mV/mmol (mean + S.D., n = 4). Recovery in whole
blood was o0-98% (mean 94%, n = 12). With an injection interval of 3 min the precision
with the sensor was < 3 % over more than 100 blood samples. Response time was about
60 s. The calculated glucose values correlated, r = 0.98, with the values obtained with an
YSI glucose: analyser (Yellow Springs Instruments, Yellow Springs, OH, USA), over the
range 2-20 mmol/l.

Keywords: Enzyme thermistor (ET); Glucose monitoring; Semi-continuous; Bedside

1. Introduction

Knowledge of the actual glucose concentration in the blood is import-

ant in many clinical situations. In some situations the continuous or

*Corresponding author.

0098-8981/96/$15.00 © 1996 Elsevier Science B.V. All rights reserved

SSDI S0009-8981(96)06306-1
188 T. Carlsson et al. I Clinica Chimica Acta 251 (1996) 187-200

semi-continuous monitoring of the blood glucose concentration would be

of considerable value for the treatment of the patients, such as in diabetic
keto-acidosis. One way of improving the situation would be to use a
sensor which more or less continuously monitors the glucose level. Ex vivo
monitoring of glucose can be performed by sampling intravenously or in
the interstitial fluid in the subcutaneous tissue. Although many interesting
attempts have been presented using implantable and external monitoring
systems, no such sensor system with sufficient precision and stability yet
exists.
A number of different principles for the measurement of glucose exist
today [1]. Almost all of them are based on an enzymatic reaction. Routine
analysis in hospitals often uses the enzyme hexokinase together with
spectrophotometric detection of a UV-absorbing or colored product [1].
Home monitoring is usually based on a colorimetric detection principle.
These monitors often produce a satisfactory result; however, their reliabil-
ity may not be sufficient [2]. Most of the sensors used so far are based on
enzyme electrodes with hydrogen peroxide or oxygen detection techniques.
However, in these, interference causing unpredictable drift is still not a
fully solved problem. Work is also going on with new concepts such as the
use of mediator techniques or direct oxidation without any enzyme present
[3]. Non-invasive monitoring with near infrared light (NIR) looks very
attractive but suffers from problems with background interference.
The major problem with the available enzyme electrodes is the internal
drift which is attributable to the construction of the sensor per se, changes
in the transducing layer, fouling of membranes and/or interference from
electrolytes [4]. To compensate for this drift, a recalibration procedure has
to be arranged at regular intervals [5,6]. With the thermal detection
principle on the other hand, the sensor part (a pair of thermistors) is
physically separated from the sample stream. Fouling is not likely to take
place in this system due to the continuous cleaning by the axial flow. A
rather thick layer of deposited material would probably be needed to affect
the signal. We have, however, never seen this in practical work [7].

2. Materials and methods

2.1. Equipment
The main components of the bedside monitor system was an enzyme
thermistor together with a buffer pump and an injection valve connected
to a computer for control, analysis and presentation of the results. For
connecting the sensor to the patient we used a prototype to a sampling
device, composed of a double-lumen catheter glued to standard pump
 T. Carlsson et al. I Clinica Chimica Acta 251 (1996) 187-200 189

tubing. The catheter was a commercially available double-lumen catheter

from Gambro (Lund, Sweden) and pump tubing (PVC) was from Alitea
(Stockholm, Sweden). The sampling device was carefully sterilized before
use. Before sterilization, it had to be controlled for the eventual presence
of any bacterial contamination (pre-sterile value). This was done through
a conventional bacterial counting test. Under sterile conditions, aliquots
of a peptone solution were flushed through the fine lumen, collected on
Petri plates and incubated at 37°C for 2 days. In the event of growth, the
device was discarded or the sterilization was repeated before an additional
bacterial test could be performed. Sterilization was effected with formalin
at 70°C.
A detailed description of the enzyme thermistor together with other
applications has been published elsewhere [8]. In the present trials, the
buffer flow was 1.2 ml/min and the sample volume 7.5 #1. The buffer was
composed of 0.1 mol/1 phosphate, 0.004 mol/1 EDTA, 0.154 mol/1 NaC1
and Tw,een-80 at 0.5% (v/v) concentration. The pH of the buffer was
adjusted to 7.4 with 5.0 mol/1 NaOH. As anticoagulant we used a solution
with a concentration of 0.008 mol/1 EDTA and 0.154 mol/1 NaC1 in water.
The enzyme columns were made from polyacetal tubes, with an approx.
size of 20 x 4 mm. Each column was filled with 0.1 g of an activated
bead-shaped support, controlled pore glass (CPG, Schuller GmbH,
Steinach, Germany). The particle size distribution was 100-160 #m with
a mean pore diameter of 50 nm. The beads were kept in the tube by small
prefilter discs made of sintered polyethylene (Vyon®).
Glucose oxidase (EC 1.1.3.4, from Aspergillus niger) and catalase (E.C.
1.11.1.6, from bovine liver) were from Sigma Chemical (St. Louis, MO,
USA). (Sin the support, were immobilized 5 mg of glucose oxidase (approx.
100 U) and 13000 U of catalase. The immobilization procedure was
performed according to a standard method [9].

2.2. Sampling and measurement

Fresh blood was collected from the clinical laboratory at the hospital in
3 ml wacutainer vials (Becton Dickinson Systems, France). To each
container 0.15 ml of NaF solution (sat.) was added and stored at +4°C
before use. The samples were used within 4 h. No preservative was used
in the serum samples. The sensor system was controlled by a computer
through a 12-bit interface, PCL 812 (Advantech USA), equipped with the
data acquisition program, Labtech Notebook (Labtech, Wilmington, MA,
USA). The interactive software program was designed to calculate the
peak heights during the monitoring and transfer these data into mmol/1 of
glucose by the use of the preset calibration factors for sensitivity and
background. Fig. 1 shows the principal experimental setup.
190 T. Carlsson et al. I Clinica Chimica Acta 251 (1996) 187-200

amplifier
I 15.5 m m ~

PC
in~

/~' ~ oolurnn
~zyn~tl-~mslm-

w~pains
liquidflow

d~ ln~f~

Fig. 1. The experimental setup used in the project was composed of an enzyme thermistor,
a pump for buffer and anticoagulant, an injection valve, and an amplifier. The PC was
connected through an interface card with A/D and D/A conversion facilities (12 bit). The
sampling device was connected to the injection valve by standard plastic fittings.

In our experiments we have used manually packed columns and, to test

the variation between them, the sensitivity coefficients for the columns
were determined. The thermal sensor measures the sum of all exothermie
and endothermic events taking place in the column. Interference from
nonspecific reactions (background) could be removed with a reference
column (the sample flow is divided into two separate streams entering one
active and one inactive column, respectively [10]). This is, however, a
somewhat more complicated solution, since it m a y be difficult to arrange
two equal flow paths. In this project we have instead estimated a value for
the background signal and used this in the calculation of the glucose
concentration. The background was determined with internal standard
addition technique by extrapolating the calibration lines to zero glucose
concentration where the y-intercept gave an estimation of the background.
 T, Carlsson et al. I Clinica Chirnica Acta 251 (1996) 187-200 191

0,5

0.4

~0.3

~0.2

0.1
J

0 --
2 4 6 8 10 12 14
Glucosecovxentralion,mmoFI
Fig. 2. The background signal was determined through a standard addition technique. In
this, increasingamounts of glucose were added to aliquots of whole blood samples in steps
of 2, 4 and 10 retool/l, respectively,giving samples of increasing concentrations of glucose.
The response was plotted against the glucoseconcentration, the slope showing the column
sensitivity. By extrapolating the response lines to zero glucose concentration, the y-
intercept represented an estimation of the background signal (non-glucose-dependent
signal). In the figure, results from three differentsamples are shown.

In order to determine the true response, the actual glucose concentration

in the sample must be known with acceptable accuracy. Since there is no
absolute method available (with high accuracy and precision), for the
direct determination of glucose in whole blood [2-1, we have constructed
the calibration lines by using glucose values taken with an YSI glucose
analyser (Yellow Springs Instruments, Yellow Springs, O H USA). Fig. 2
describes the basis for the background signal determination. In order to
evaluate the lifetime of a column, repetitive injections at 3-min intervals
were performed, the glucose signal was measured and any drift in the
baseline monitored.
On-line monitoring in diabetic patients was performed at the hospital
during daytime with volunteer patients placed in a leaning position and
lasted 2--5 h. In these trials (approved by the Local Ethical Committee at
Karolinska Institutet) an i.v. cannula Venflon (Viggo, Sweden) was
inserted into the vein of the forearm of the patient to guide the tip of the
double hlmen catheter. The other end of the catheter was connected via
the pump tubing at the site of the injection valve. During sampling, the
blood flow from the vein (approx 3 ml/h) was immediately mixed with an
equal flow of anticoagulant and a fixed sample volume of this mixture was
192 T. Carlsson et al. I Clinica Chimica Acta 251 (1996) 187-200

injected to the sensor. For comparison, samples were also taken from an
indwelling catheter in the left forearm and analyzed off-line with the Y S I
glucose analyser.

2.3. Biocompatibility
At the start of the experiment we used a sampling device without any
pretreatment; however, we soon encountered clogging problems with this.
These clots were mostly located at the site of the injection valve, in some
cases at the connection between the double-lumen catheter and the p u m p
tubing, and sometimes at the tip of the catheter. For improved biocom-
patibility some parts (the injection valve and some critical connections) of
the sampling device were therefore coated with covalently coupled heparin
(Carmeda Stockholm, Sweden) [11-].

3. Results

From the standard addition trials we have found that the column
sensitivity (mV/mmol) with a sample volume of 10 #1 was 46 m V / m m o l
(Table 1) in standard solution and 45 mV/mmol (Table 2) in whole blood.
With a 7.5 pl loop the value was 28 mV/mmol in blood, significantly
smaller than the sensitivity, 36 mV/mmol, found in standard solution
(from Fig. 3). Our explanation of this is that, at 10 pl, the dispersion of
the substrate in the blood sample was as effective as in the standard
sample. Decreasing the sample size by shortening the sample loop led to
a reduced dispersion in the blood sample. This change in sensitivity was
more or less identical to the size reduction. We speculate that the possible
explanation of this is related to the high viscosity of the blood. The
decrease in standard sensitivity was, on the other hand, directly related to
the smaller sample volume (46/36 ~ 1.3; 10/7.5 ,-~ 1.3) Thus substrate
transport in the standard sample was not restricted by decreased disper-
sion in this case. The effect of sample loop diameter (I.D.) and dilution of
the sample can be seen in Fig. 3. At a fixed volume of 10 pl, I.D. = 0.15
mm and 1 + 1 dilution gives about two to three times higher sensitivity
compared to I.D. = 0.3 m m (1 + 1). Extrapolating the line for 0.15 mm
(1 + 1) gives a sensitivity of more than 100 mV/mmol at 22 pl, superior
to all tested cases. The sensitivity was at 7.5 #1 sample volume 36
mV/mmol, nearly the same as 20 #1, I.D. = 0.5 m m (1 + 2). The turnover
of the substrate with I.D. = 0.15 m m (1 + 1) was thus much better, by a
factor of nearly 2. Therefore, we were able to decrease the total amount of
blood necessary for every measurement by using a sample loop with a
smaller I.D. Moreover, the sensitivity in the 7.5 #1 case was only slightly
lower than what was achieved with a sample volume of 20/fl, I.D. = 0.3
 T. Carlsson et al. I Clinica Chimica Acta 251 (1996) 187-200 193

Table 1
Sensitivity coefficients for manually packed standard columns

Column number Sensitivity(slope mV/mmol)

1 48.8
2 45.8
3 45.9
4 48.4
5 47.6
6 45.2
7 47.1
8 44
9 47.4
10 46.8
11 42.8
12 48.1
13 43.7
14 48.9
15 45.9
16 46.8
Mean -I- S.D. 46.5 + 1.8

The sensitivity coefficients were determined from the slope of the response lines
using standard glucose solutions of varying concentrations. Sample size was 10/A and flow
rate 1.2 ml/min. The samples were diluted 1 + 1 with anticoagulant solution
before injection. The intervariation among the columns was below 4%.

mm (1 + 1). This was most probably due to greater dispersion leading to

improved diffusion of substrate and product with the smaller I.D. Since
the slope in this case was rather steep, decreasing the sample volume below
7.5 #1 would have reduced the sensitivity too much. F o r the studies on
whole blood we therefore choose the sample loop of 7.5 #1 with an I.D. of
0.15 m m
The variation in background, from the nonspecific reactions was found
to be small regardless of whether 10 or 7.5 #1 sample loops were used.
F r o m two independent experiments, shown in Table 2, the average value
was found to be 45 _+ 8 mV (mean _+ S.D., n = 8) and we thus estimated
the contribution from the background to be approximate 0.05 V in whole
blood. In comparison, the b a c k g r o u n d with aqueous standard was
11 _+ 23 mV, n -- 16. We speculate that the high background with whole
b l o o d could be related to the physical properties of the blood, e.g.
viscosity. The shortest possible injection interval was 3 min and under this
condition a CV of 2.3% was achieved during more than 100 measurements
in blood. The coefficient of variation corresponds to an S.D. of 0.15
194 T Carlsson et al. I Clinica Chimica Acta 251 (1996) 187-200

Table 2
Determination of the value of the background signal and the sensitivity coefficient in whole
blood

Intercept (mY) Slope (mV/mmol)

Experiment 1
1 40 45
2 38 46
3 50 46
4 53 44
Mean 45 45 + 1
Experiment 2
1 52 29
2 31 29
3 54 28
4 53 27
Mean 48 28 + ]

Linear regression analysis was performed after the addition of known amounts of a
glucose standard to aliquots of the samples. The background signal was determined from
the y-intercept and the sensitivity coefficient from the slope. In the first experiment a
sample loop of 10 pl volume was used while in the second experiment the sample volume
used was 7.5 pl. Two separate columns were used. As shown, the intercept values in the
experiments were not significantly different.

80
/
70 /
0.5 mm (1+2)//"
~ 60

5O
0.15 ~ (1+1/)//" + ~

///~/ 0.3 mm (1

~ 20

10

0 6 12 18 24 30
Samplevolume
Fig. 3. The sensitivity, expressed as mV/mmol, is a function of sample loop volume, sample
loop diameter, and dilution. For the overall best performance, an optimization between
sensitivity and column lifetime was done. As can be seen from the figure, sample tubing
with I.D. = 0.15 ram, 1 + 1 dilution, gave the same specific sensitivity (slope) as tubing of
0.5 mm (1 + 2). Thus, we could decrease the amount of blood in the sample by a factor
of two while preserving sensitivity.
 T. Carlsson et al. I Clinica Chimica Acta 251 (1996) 187-200 195

0.8

0.7

0.6

_- 0.5

~'0.4 .,,-.-.,.,--

~0.3
/
1).2

0.1

0
20 40 60 80 100 120
Number of whole bleed samples
Fig. 4. The glucose signal from two different whole blood samples (squares, triangles) was
determined. The thermal baseline drift (dashed line) did not interfere with the glucose
specific signal in the sample for over 100 whole blood samples. Injection interval was 3 min
and the overall variation coefficient was 2.3%. The glucose concentration, as determined
with our method, was 6.04 + 0.13 mmol/l, n = 96 (squares) and 7.78 ___0.18 mmol/l,
n = 20 (triangles).

mmol/l and would probably be without any clinical relevance. The special
advantage with the methodology is demonstrated in Fig. 4. This figure
shows that a drift in the thermal baseline (dashed line) did not have any
effect on the true measured glucose signal. This is because the glucose
value was always calculated from the difference between a peak maximum
signal and a peak start signal (relative measurement). Therefore, any drift
in the baseline will always be compensated for. Fig. 5 presents results from
measurements in sera and whole blood compared with values determined
with an YSI instrument. The analytical recovery (through standard
addition) in sera was found to be in the range of 95 to 106% (mean 101%,
n = 19), while in whole blood the range was 90 to 98% (mean 94%, n %
12). In comparison, the recovery with the YSI analysis was in the range
93 to 107% (mean 97%, n = 11) in blood.
In the patient monitoring trials, problems with clogging in the sampling
part were in some cases so troublesome to overcome that the trial had to
be cancelled. Dismantling and visual examination of the device showed the
presence: of blood clots. The clots were located mostly in the fine lumen of
the catheter and at the interface between the sample loop tubing and the
inlet site of the injection valve head (narrow passage). In some few cases,
clogging at the tip of the catheter also was observed. We noticed a
196 T. Carlsson et al. I Clinica Chimica Acta 251 (1996) 187-200

20

18

[]
[]
14
[] D>~x

.~ ]0
~EU
s

8 6

~ 2

0 2 4 6 8 10 12 14 16 18 20
Ghicm¢ concentration retool/l, ET-mcthod

Fig. 5. Glucose concentration in sera (square) and whole blood (cross) determined with the
ET method in comparison with the YSI method. Results in sera are mean of three, while
in whole blood we used the mean of two determinations. Regression analysis for whole
blood gave YSI -- 0.96ET+0.35 mmol/l and r = 0.99.

difference between the patients in that the coagulation cascade seemed to

be easily triggered in some cases, while in others the opposite condition
existed. Some results from patient trials are found in Fig. 6.

4. D i s c u s s i o n

The purpose of this study was to investigate whether the enzyme

thermistor could be used as the sensor part in a bedside monitor system.
Such a system m a y be characterized by, (a) precision and stability with
CVs < 5%, (b) 3 - 4 readings/h up to 24 h, and (c) one calibration at the
start. We have shown that with the prototype system presented, semi-
continuous monitoring of glucose in whole blood overnight could be
possible and calibration could be reduced to once before start. With this
methodology, the magnitude of the nonspecific, i.e. non-glucose-depend-
ent, signal can be related to factors such as flow, viscosity, buffer
composition, support and column packing. Since we have used manually
packed columns, we cannot exclude the possibility that flow characteristics
for individual columns could be different which in turn could have an
effect upon the background signal.
The small a m o u n t of blood used in each measurement will eventually
cause the enzyme column to clog if no special arrangements are made to
prevent this. Therefore, minimizing the amount of blood absorbed in the
 T. Carlsson et aL I Clinica Chimica Acta 251 (1996) 187-200 197

A
12

10
| x x
X

r7 0 O
X

O 0
×

O 0
[]
X

OD D
o []
f $ o
x
0 o
o
o
o
6 x o
x 0 0

J o o o o o
o []

O 20 40 60 80 I00 120 140

Time, mm
B
24
22
×
2O

0 0 .
o
16 o ×
×0 ×
14
0 ° o o o o
~× x o o
12 o 0
[] 0 0 O o
0 []

$
6
4
2
50 100 150 200 2JO 3OO
Tia~ m
Fig. 6. On-line monitoring in two diabetic patients with the prototype bedside monitor.
Injection interval was 4 min (A) and 10 min (B), respectively. Values with reference method
shown as crosses. The regression equation was ET = 0.76YS1 + 1.28 mmol/1, r = 0.97.

enzyme column is a task of major importance [12]. However, there is no

single solution to this problem. The selection of the most appropriate flow
condition could be one important factor in order to minimize adsorption
onto surfaces [13]. Dilution also will probably improve column perform-
ance and lifetime by decreasing the viscosity and thereby decreasing the
amount of blood components adsorbed [12]. Also, the choice of support
material in the column may be of importance 114,157. The lifetime is also
dependent on the injection intervals (washing-out effect). For long-time
198 T. Carlsson et al. I Clinica Chimica Acta 251 (1996) 187-200

monitoring (>overnight), injection intervals of 15-20 min may be more

realistic which in turn will improve column lifetime.
In comparison with the reference values, our method seemed to give a
somewhat lower reading. This may to some extent be explained by the fact
that in the YSI method the sample is diluted 25 times; these values being
more comparable to plasma glucose, contrary to the one time dilution of
the sample in our method. Heterogenous sample due to insufficient mixing
of blood and anticoagulant could also be partly responsible for this.
Development of a fluidic component (sampling device) as a critical part
of a bedside monitor is a challenge due to the coagulation complex in the
blood. In this work we have, to some extent, addressed the problem of
biocompatibility among materials and equipment and we have identified
some of the hurdles for a continuous sampling device for whole blood in
small channels (<0.2 mm). Coating of the injection valve head with
heparin reduced the clogging problems at that site; however, the major
part of the sampling device, i.e. lumen catheter and sample loop, could not
be treated with the available coating process.
Subcutaneous monitoring, because of its less invasive mode of oper-
ation, could be an interesting alternative to blood monitoring. The small
sample volumes available in the extracellular space should not be a
restriction to the thermal sensor, since in the present study it is already
working in that range (7.5 pl). With a miniaturized thermal sensor, sample
volumes in the sub-microliter range are possible [16]. Relating the
subcutaneous glucose value to the true value in the blood might not be so
straightforward, especially under non-steady state conditions [17]. Even
though some reports state that the interstitial glucose concentration is
nearly identical to that in blood [18,19], other reports [17,20,21] show
evidence that the difference can be significant. As an example, the actual
rise in the blood glucose level following a glucose load can be delayed up
to 20 min, when measured in the subcutaneous tissue [17,22,23]. More-
over, with a subcutaneously implanted sensor, a lag time of 45 min has
been reported [24]. Sampling from whole blood as an alternative to
subcutaneous sampling (e.g. microdialysis, ultrafiltration) has the advan-
tage of showing the actual blood glucose concentration in real time,
without any unnecessary delay. Though more invasive, the risks with
sampling from a vein (infection, thrombosis, etc.) do not present any
additional problems in the clinic. Since it would be possible, also, to
monitor glucose in the subcutaneous tissue with our sensor, simultaneous
comparisons of the glucose level in blood and extracellular fluid would be
possible. This prototype represents an important step in the development
of a clinical bedside monitor that in the future would be a valuable tool
in intensive care and metabolic research.
 T. Carlsson et al. I Clinica Chimica Acta 251 (1996) 187-200 199

Acknowledgments

T h i s w o r k w a s s u p p o r t e d in p a r t b y g r a n t s f r o m the N a t i o n a l S w e d i s h
B o a r d for T e c h n i c a l D e v e l o p m e n t ( N U T E K ) , the Swedish M e d i c a l R e -
s e a r c h C o u n c i l ( N o . 19x-6589) a n d the Bert y o n K a n t z o w F o u n d a t i o n .

References

[1] Sacks D. Determination of glucose in body fluids. In: Burtis C, Ashwood E, eds. Tietz
Textbook of Clinical Chemistry, 2nd edn. Philadelphia: W B Saunders Company
1994;958-967.
1"2] Wiener K. Whole blood glucose: what are we actually measuring? Ann Clin Biochem
1995;2;2:1-8.
[3] Preidel W, yon Lucadou I, Lager W. In vivo experiments with the electrocatalytic
glucose sensor in sheep. Biosens Bioelectron 1993;8:299-306.
I-4] Moatti-Sirat D, Poitout V, Thom6 V et al. Reduction of acetaminophen interference
in glucose sensors by a composite Nation membrane: demonstration in rats and man.
Diabetologia 1994;37:610-616.
1-5] Poitout V, Moaffi-Sirat D, Reach G. Calibration in dogs of a subcutaneous
minial:urized glucose sensor using a glucose meter for blood glucose determination.
Biosens Bioelectron 1992;7:587-592.
[6] Velho G, Froguel Ph, Th6venot D, Reach G. Strategies for calibrating a subcu-
taneous glucose sensor. Biochim Biomed Acta 1989;48:957-964.
[7] Harborn U, Bin X, Danielsson B. Evaluation of a miniaturized thermal biosensor for
the determination of glucose in whole blood. (in press).
[8] Danielsson B. Calorimetric biosensors. J Biotechnol 1990;15:187-200.
[9] Weetall HH. Preparation of immobilized proteins covalently coupled through silane-
coupling agents to inorganic supports. Appl Biochem Biotechnol 1993;9:157-188.
[10] Mattiason B, Danielsson B, Mosbach K. A split-flow enzyme thermistor. Anal Lett
1976;9:867-889.
[11] Olsson P, Larm O. Biologically active heparin-coating in medical devices. Int J Artif
Organs 1991;14(8):453-456.
[12] Wilding P, Pfahler J, Haim B, Zemel J, Kricka L. Manipulation and flow of
biolo~;ical fluids in straight channels micromachined in silicon. Clin Chem 1994;40/
1:43-47.
[13] Diqu61ou A. Lemozy S, Dupouy D, Bernard B, Sakariassen K, Cadroy Y. Effect of
blood flow on thrombin generation is dependent on the nature of the thrombogenic
surface. Blood 1994;84:2206-22t3.
[14] Nobu'ioshi K, Motohisa F. Flow enthalpimetric determination of glucose. Methods
Enzyraol 1988;137:225-232.
[15] Harborn U, Bin X, Danielsson B. Determination of glucose in diluted blood with a
thermal flow injection analysis biosensor. Anal Lett 1994;27(14):2639-2645.
[16] Bin X, Hedberg U, Mecklenburg M, Danielsson B. Fast determination of whole
blood glucose with a calorimetric micro-biosensor. Sens Actuators B 1993;15-
16:14!-144.
[17] Schmidt F, Slutter W, Schoonen A. Glucose concentration in subcutaneous extracel-
lular space. Diabetes Care 1993;5:695-700.
200 T. Carlsson et aL I Clinica Chimica Acta 251 (1996) 187-200

1-181 Jansson PA, Fowelin J, Smith U, Lonnroth P. Characterization by microdialysis of

intercellular glucose level in subcutaneous tissue in humans. Am J Physiol
1988;255:E218-E220.
[19] Bolinder J, Hagstrom E, Ungerstedt U, Arner P. Microdialysis of subcutaneous
adipose tissue in vivo for continuous glucose monitoring in man. Scand J Clin Lab
Invest 1989;49:465-474.
1-20] Schmidt F, Aalders A, Schoonen A, Doorenbois H. Calibration of a wearable glucose
sensor. Int J Artif Organs 1992;15:55-61.
[21] Poitout V, Moatti-Sirat D, Reach G e t al. A glucose monitoring system for on line
estimation in man of blood glucose concentration using a miniaturized glucose sensor
implanted in the subcutaneous tissue and wearable control unit. Diabetologia
1993;36:658-663.
[22] Hashiguchi Y, Sakakida M, Nishida K, Uemura T, Kaijiwara K, Schichiri M.
Development of a miniatuized glucose monitoring system by combining a needle-
type glucose sensor with microdialysis sampling method. Diabetes Care 1994;5:387-
396.
[23] Rebrin K, Fischer U, von Woedtke T, Abel P. Clinical relevance of subcutaneous
glucose concentration in diabetes monitoring: an experimental study. Horm Metab
Res 1993;25:65.
[24] Claremont DJ, Shaw GW, Pickup JC. In vivo molecular sensing in diabetes mellitus:
an implantable glucose sensor with direct electron transfer. Diabetologia
1989;32:213-217.