Glucose Oxidase Method for ContinuousAutomated Blood Glucose Determination

John W. Rosevear,* Kenneth J. Pfaff, Frederick J. Service,George D. Molnar, and Eugene Ackermanf

A glucoseoxidase-peroxidasemethod for continuous automated monitoring of bloodglucosehas been developed.The responseis linear over the range 0-800 mg/100 ml.Sensitivity can be maintained for 24 hr or longer and can be restored by rinsing theanalytic system with sulfuric acid to permit studies of longer than 48 hr in duration.A precisionof ± 1% can be maintained between rinsesfor samplescontaining 100-600 mg of glucoseper 100 ml. This method is satisfactorily specific for glucose: Theresponse with other sugars is less than 1% of the response obtained with thesame concentration of glucose. Ascorbic acid causes no significant inhibition ofthe response to glucose. The inhibition by uric acid has been reduced fifty-foldcompared to that in other methods. Transit through the sampling catheter andanalytic system requires 15 mm. Timed from the first detectable responseto a change in concentration, 25% of total responseis achieved in 30 sec and 90%in 80 sec. Fifty percent of an oscillation with a half-period of 45 sec can be detected;no oscillations this short were observedin records of human blood glucose. Applica-bility and feasibility of this method have been demonstrated in over 2000 hr ofrepeated blood glucose recordingsin 12 diabetic and 6 normal subjects.

HAVE BEEN STUDYING how blood glucose behavior in unstable

diabetics differs from that in stable diabetics and in nondiabetic sub-jects. Even under highly standardized conditions, blood glucose levels

can change rapidly and extensively, as well as variably, over successive

From the Mayo Clinic and Mayo Foundation, Sections of Biochemistry, of Medicine, andof Biophysics, Mayo Graduate School of Medicine, University of Minnesota, Rochester, Minn55901.

Supported in part by Research Grants AM.10152, FR.5530, and FR-0007 from the NIH, US

Public Health Service.The technical assistance of E. Stuart Eickelberg, Elton E. Erp, Richard T. Jones, Helen L.

Sievers,and Orlando P. Wadel is gratefully acknowledged. Mrs. La#{235}lC. Gatewood contributed

to the numerical analysis methods and digital computer programs for calculating and plottingthe blood glucose concentrations from the absorbanee values taken from the strip record.

Received for publication Aug 24, 1968; accepted for publication Jan 20, 1969.*Address for reprint requests: Dr. John W. Rosevear, 200 First St Southwest, Rochester,

Minn 55901.tPresent address: College of Medical Sciences, University of Minnesota, Minneapolis.

680

Vol. 15, No. 8, 1969 CONTINUOUS BLOOD GLUCOSE 681

nyctohemeral cycles, especially in the most unstable (brittle) diabetics(1). Our methods were designed so that data could be obtained re-garding these rapid changes in blood glucose while an experiment wasin progress for periods of 48 hr. For the determination of blood glucose,we selected a continuous monitoring technic in which the patient wasattached directly to the analytic equipment by a catheter long enough

to permit freedom of movement.

An automate(l method for continuous determination of blood sugarin hulnall subjects was first described by Weller et al (2) in 1960. The

development of tile continuous monitoring technic provided a morerapid and convenient method of following a subject’s current bloodsugar level by taking blood directly from the subject and introducingit continuously into the analyzer. This is in contrast to the conventionalmethod of frequent analysis of discrete specimens. More recently, otherinvestigators (3-8) have described methods for continuous monitoring

of blood sugar. These methods all used an AutoAnalyzer ferricyanidereduction method for tile determination. The limitations of the Auto-Analyzer ferricyanide method with respect to precision, range, andspecificity are well recognized (7-12). For a given range, the precisionfor low glucose concentrations is limited by the minimal transmittance

change which can be read from the stri1) record. Since the determinationdepends on the disappearance, rather than tile appearance, of color,

equal increments of glucose result in progressively smaller incrementsof transmittance as the glucose concentration of tile analyzed sampledecreases. The precision improves as the range of the method is de-creased. Readability errors equivalent to ±0.2% transmittance orgreater appear likely from the published data for all ranges (2-8).

For a range of 0-500 mg/100 ml, the error equivalent to ±0.2% trans-mittance would be ±5% of the true value at 50 mg/100 ml, ±2% at

100, ±1%, at 200,and ±3% at 500 mg/100 ml. For a range of 0-250 mg/100 ml (40-100% transmittance), the error equivalent to ±0.2% trans-mittallce would be at least ±2% at 50 mg/100 ml and ±1% at 100 mg/

100 ml. (Reducing the readabilityerror to ±0.1% transmittance would

reduce the equivalent error to one half of that for ±0.2% trans-

mittance.) Furthermore, the ferricyanide method is not specific for

glucose. Nonglucose reducing substances of plasma contribute a smallbut variable increment to the reading on the strip record.

As the ferricyanide method is used in tile routine clinical laboratory,its limitations either are minimized by restricting the range of the

samples (usually to 50-250 mg/lOt) ml) and standardizing frequentlyor are considered unimportant for a routine chemical determination.

682 ROSEVEAR ET AL Clinical Chemistry

For certain applications, these limitations are excessively restrictive.in an attempt to avoid these limitations, modifications of the glucoseoxidase-peroxidase method for glucose have been developed for theAntoAnalyzer (13-17). These have the advantages of combining agreater precision at low values with an extended range and of specifi-

cally measuring glucose. Recently, two variations of the glucose oxidasemethod for the continuous monitoring of blood glucose have beendescribed. One method (18) depends upon the measurement of oxygendepletion as a result of glucose oxidase activity on glucose. The othermethod (19) uses glucose oxidase and o-toluidine for discontinuous orcontinuous application. Compared to the ferricyanide methods, theglucose oxidase-peroxidase methods offer many advantages.

To carry out our research plans, we needed a method for monitoringthe variations of the blood glucose level over a wide range of values inambulatory human subjects for periods of 48 hr or longer. The methodwas expected to meet tile following requirements: (1) be convenient

and suitablefor use in the proximity of the subject; (2) produce preciseand frequent (essentially continuous) measurements over long periods;

and (3) provide the data with as short a time delay as possible.Byimproving tile range, precision, stability, specificity, dynamic response,

and convenience of glucose oxidase-peroxidase methods, these goalshave been approached. The following report describes our method,evaluates its performance, and illustrates its application.

Materials and MethodsAnalytic System

The analyzing system consists of AutoAnalyzer* modules and suitablecontainers for reagents and waste solutions, all mounted on a mobilecart (Fig 1) for convenient use in a hospital room. The components of

the analyzing system and flow scheme are diagrammed in Fig 2 and 3.

The blood is pumped directlyfrom the patient,by means of a Teflon

catheter (internal diameter, 0.015 in.),at a rate of 0.1 mi/mm, suc-cessively through two proportioning pumps. The analytic system has

too many lines for one pump (Technicon proportioning pump I) tohandle with consistent, noise-free performance. Furthermore, the use

of two pumps has a safety feature in that if the roller chain of one isaccidentally released, the other will prevent flow of reagents into thesubject. This is a possibility in systems with only one pump because

occasional obstructions have been observed in the reagent lines and

could result in a higher pressure in the analytic system than in thesampling catlleterin the subject.In addition,the long sampling catheter

*Technicon Corp, Tarrytown, NY.

Fig 1. Mobile analyzing unit.

Vol. 15, No. 8, 1969 CONTINUOUS BLOOD GLUCOSE 683

used (up to 10 ft) allows the subject enough freedom of movement tooccasionally be below the level of the analytic system. In this case thesubject’s venous pressure might be lower than the head of pressurethrough the analytic system.

The blood is diluted with magnesium sulfate (diluent) solution andis passed to the dialyzer. Tile glucose dialyzes through a Type Cmembrane into a sodium sulfate (recipient) solution. The recipient

solution passes tllrough a pulse suppressor (Technicon pump manifoldtubing, 0.025-in, inside diameter-color code orange and white) and is

then combined with the enzyme-dye solution from the proportioning

pump. Teflon tubing (lightweight spaghetti tubing, AWU No. 16 LWT/natural-ID, 0.053 in.; wall, 0.006 in.*) conveys tile reaction mixture

to the 50#{176}heating bath. (Because the Teflon tubing has very little

tendency to adsorb the oxidized dye, it is much superior to Tygon

tubing for transporting the reactionmixture.)

In the heating bath, the Inixtureflows through 40 ft of glass coil; the

bath contains a special thermostat (Bronwill therrnoregiilatort), whichpermits maintaining the temperature within ±0.10 to reduce variationsin the color development cause(l by variations in temperature. After a4-mm passage through the heating bath, tile reaction mixture is

combined with sulfuric acid (49% w/w) in two double-mixing coilsand is conveyed to the colorimeter in Teflon tubing. The sulfuric acidis added to intensify and stabilize the oxidized dye. The colorimeteruses a 528 nm interference filter and standard 15-mm flow cell. Seg-ments of air are introduced into 1)0th diluent and recipient lines and

*pemmmssvlvani,t Fluorocarbon Co., Immc., Clifton Heights, Pa.

tOatalog No. 65437-23; Mathesozi ScientificCo., Chicago, Ill.

PROPORTIONING

684 ROSEVEAR fT AL Clinical Chemistry

subsequently are removed before the reaction mixture enters the cuvet(standard AutoAnalyzer technic). In the enzyme-dye line, there is apulse suppressor consisting of about 3 ft of transmission tubing coiledaround a glass rod, followed by a short length of small-diameter Tygontubing. Other pulse suppressors are placed in the air lines to thediluent and recipient lines and in the 112S04 line (see Fig 2 for details).

The enzyme-dye reagent is kept at 00 on top of the cart in an

insulated bucket filled with crushed ice. A 30-mm coarse sintered-glass

mi/mm

AIR 1.20

AIR 0.60

SAMPLE 0.10AIR 1.20

Na2504 2.00

ENZYME 2.50DYE

ENZYME 2.50DYE

2.00

2.00

1.20

0.10

1.60

1.60

Fig 2. AutoAnalyzer modules and flow scheme for glucose oxidase-peroxidase method of

continuous determination of blood glucose. In the following definitions of parts, catalog letters

and numbers refer to Technicon products. Pulse suppressors: (a) P8-4; (b) PS-2; (c) Tygon

0.063-in.ID x 0.031-in,wall x 3 ft; (d) orange-white Tygon manifold tubing 0.025-in.ID;

(e) P8-3. Glass connectors: DO; Dl; H3. Transmission tubing: (T) Teflon 0.053-in. ID x0.006-in,wall; (TY) Tygoa 0.063-in.ID X 0.031.in. wall; (AC) acid flex 0.062-in.ID. Mixing

coils: (MC,) No. 116-103-7; (MC,) No. 116-103-7 plus 116-103-9. Special connectors: (f) for

details see Fig 3.

IS NS N4

I

lU!IHHffl’

Fig. 3. Teflon tubing (T) is connected to glass tubing (C) by means of an N5 nipplesecured at the glass end by a Tygon sleeve (TS) (#{188}-In.ID X h-in. wall) and at the Teflon

end by an N4 nipple applied as a collar.

Vol. 15, No. 8, 1969 CONTINUOUS BLOOD GLUCOSE 685

filter (such as Cornilig 39580) is used to filter the enzyme-dye reagentas it is drawn fronl the stock bottle (4-liter glass bottle). The upper shelfof the cart houses the dialyzer, water bath, and colorimeter, and pro-vides storage for the containers of diluent, recipient, and sulfuric acid(10-liter polyethylene wide-mouth bottles; Bel-Art Products, F 10906).

The lower shelf of the cart houses a container (Lucite, 13 in. wide by10 in. deep by 21 in. long) for the waste solution. A drain spigot isprovided underneath the cart for draining the waste container. Two

fans are mounted inside the cart to prevent heat accumulation bydrawing air in at one end and exhausting it through the opposite end,which is open.

Reagents

Diluent Dissolve 40 g of analytic reagent grade MgSO4 . 7H00 inabout 800 ml of distilled water. Add 5 ml of a 20% (w/v) solution ofpoiyoxyethylene (23) lauryl ether (BRIJ 35 SP; Atlas Chenlical Indus-

tries, inc.). Add 10,000 units of heparin and dilute to 1 liter withdistilled water.

Recipient (sodium sulfate) T)issolve 16.1 g of analytic reagentgrade Na2SO4 in about 800 ml of distilled water. Add 10.0 ml of iso-octyl phenyl etiler of polyethylene glycol (Triton X-i00; Rohm andHaas) and dilute to 1 liter with distilled water.

Buffer solution 1)issolve 21.0g of citricacid mo7nohydrate crystal(Baker Analyzed Reagent) in 800 nil of distilled water. Adjust the pIT

to 5.0 with 50% (w/v) NaOH. Dilute to 1 liter with distilled water.Enzyme-dye reagent Dissolve 800 mg of o-dianisidine dihydro-

chloride (Fermco Laboratories, Inc.) in 50 ml of water. Add 400 ml ofanalytic reagent grade glycerol (Fisher) to the dye solution. Add500 ml of the buffer solution and 13.4 ml (10,000 units) of glucose oxi-

dase-peroxidase solution (Fermcozyme 952-DM; Fernico Laboratories,Inc.) and dilute to 1 liter with distilled water. Store tile reagent at 40

until used. (Storage for as long as 10 days has not adversely affectedthe sensitivity or the linearity of the response obtained with thereagent.)

Sulfuric acid To 684 ml of distilled water, add 380 ml of 96.5%H2S04 slowly while stirring and cooling.

Standard glucose solution For routine calibration, prepare glu-cose standards by diluting a 2000-mg/100 ml stock solution. All stand-ard solutions of glucose are made up with 0.2% henzoic acid in double-deionized water as a diluent and allowed to stand for 24 hr or longerat room temperature to allow the equilibrium between the and $ formsof glucose to be attained (16, 20).

686 ROSEVEAR fiT AL Clinical Chemistry

Results and DiscussionLinearity

The response of the method is linear from 0 to 800 mg/100 ml to

within ±0.2% transmittance under optimal operating conditions. Therecorded and calculated linear absorbances for such a calibration aredocumented in Table 1. With respect to absorbance, the relative stand-ard deviation was less than ±0.4%. To evaluate the linearity underroutine operating conditions, 21 calibrations, each with nine samples,were carried out over a period of 4 months. Regression lineswere used

to test the linearity of the response and nature of the errors involvedin the calibrations. The mean relative standard deviation from theresponse predicted by Beer’s law was less than ±1%, as determined

from linear regression linesbased on minimizing the sums of squares

of fractional (relative)absorbance deviations.The nine samples were

made up by gravimetric methods to 50, 100, 200, 300, 400, 500, 600,

700, and 800 mg/100 ml with 0.2% benzoic acid in double-deionized

water as a diluent. Each solution was aged at least 1 month at roomtemperature before being used. The mean deviation of the blanks of the21 calibrations from the extrapolated values was 0.006 absorbance units.

Range

The usable range of the present method is 0-800 mg/100 ml. The

upper limit gives an absorbance of about 1.2. Above this value, theeffective sensitivity error increases rapidly owing to the limitation in

reading the strip record and, in addition,the response becomes some-what unpredictable. Blood glucose values beyond this range have not

been encountered even in the most severely diabetic patients in ourstudies to the present. To achieve a linear response to 800 mg of glucoseper 100 ml, an o-dianisidine concentration of 800 mg/liter in the enzyme-

Table 1. L1NERITY OF RESPONSE

(fluco.e cone (mg/100 ml) Recorded aboorbance Calculaled linear absorbance

0 0.0120 0.0120

50 0.0850 0.0855

100 0.159 0.159

200 0.307 0.306

300 0.455 0.453

400 0.601 0.600

500 0.748 0.747

600 0.895 0.894

700 1.040 1.041

800 1.180 1.188

---4---H-

4py--H --1I -

-1-H- -

-i --

jJ:ii:

A B

irregularitiesin tileefficiencyof the pump. The effect of the pulse

suppressor in the air line to the recipient solution is shown in Fig 4.The quality of the recording was further improved by replacing the 7X

detergent (0.5 mi/liter) used by Robin and Saifer (17) in the recipient

*Tris (hydroxyinethyl )aminomethane.

fol. 15, No. 8, 1969 CONTINUOUS BLOOD GLUCOSE 687

dye reagent was used. Slightly greater color response (about 3%) canbe obtained by reducing the o-dianisidine to 600 mg/liter, but this oc-casionally causes some loss of linearity.

With the use of Tris*ma1eate buffer (pH 7.0),we observed precipita-

tion of dye in the stored enzyme-dye solution, which resulted in a lossof sensitivityabove 400 mg/100 ml and, consequently, a lossof linearity.

Since substitutingcitricacidlbuffer (pH 5.0),in which tileo-dianisidine

dye is more soluble, we have not encountered this problem. In fact, theenzyme-dye-citric acid buffer preparation gives linear results up to

800 mg/100 ml, even after storage at 4#{176}for as long as 10 days.

Sensitivity

An improvement in sensitivity was obtained by increasing the con-

centration of the sulfuric acid in the final mixture to 19% (w/w),

similar to that (18%, w/w) recommended by Kingsley and Getchell(10, 21). Robin and Saifer (17) used 16% (w/w) sulfuric acid in thefinalmixture.

To achieve optimal readabilityof the striprecord, the quality of the

recording was improved by several means. Placemellt of pulse sup-pressors in air and solution lines ensured a smooth and consistent pro-portioning of the solutions by compensating for the effect of the

EIiI:I± T1

Fig 4. Effect of positive-

pressure pulse suppressor

in air line to recipient

solution on “noise” in strip

record with glucose stand-

ard of 200 mg/100 ml. A,with pulse suppressor. B,

without pulse suppressor.

E:i

i:

I I I

C

L \

0

80

60

40

20

10

8

6

4

0.5rnI/L Triton X-lOO

10 mI/L Triton X-100

10 20 30 40 50 60 70 80 90 100

Tmme , secon ds

Fig 6. Effect of concentration

of Triton X-100 on trailing.Con.

centration of glucose in solutionbeing sampled was changed

abruptly from 500 to 0 mg/100

ml. Time is measured from first

detectable response to new con-

centration; percent lack of at-

tainment of steady state refers

to the change from complete

response at high level to com-

plete response at low (0 mg/lOG

ml) level.

688 ROSEVEAR ET AL Clinical Chemistr)

solution with Triton X-100 at a concentration of 10 ml/liter (Fig 5).However, slight loss of sensitivity (7%) occurs at 10 ml/liter as com-pared to 5 mi/liter. One effect of the higher concentration of Triton

X-100 isdecreased trailingof the oxidized dye (Fig 6).

A

0

0

0

0

C

C

C

C

C

Fig 5. Effect of detergent in

recipient solution on CCflj5

the strip record with glucosestandard of 50 mg/100 ml. A,

7X detergent, 0.5 mI/liter. B,Triton X-100, 1 mI/liter. C,Triton X-100, 5 mI/liter. D,

Triton X-100, 10 ml/liter.

The concentration of glucose oxidase-peroxidase in the enzyme-dye

reagent has some effect on the sensitivity. A slight increase in sensi-tivity (6%) can be obtained by increasing the enzyme concentration

from 10 to 15 units/ml of enzyme-dye reagent. With 10 units of enzymeper milliliter of reagent, the cost of the enzyme preparation is about $1

Vol. 15, No. 8, 1969 CONTINUOUS BLOOD GLUCOSE 689

per hour. The slight increase in sensitivity gamed with 15 units/mi isnot considered to be worth an additional $0.50 per hour at the present

time.

With the present method, the errors due to readability, as well asthose due to noise, appearing in the strip record result in effective

sensitivity errors no greater than ±0.1% transmittance. The errorequivalent to ±0.1% transmittance varies from ±0.3 mg/100 ml at 50mg/100 ml to ±4.5 mg/100 ml at 800 mg/100 ml. This effective sensi-tivity error of the system is one-third that of a typical ferricyanide sys-tem, with the same readability, at 50 mg/100 ml, which, furthermore,could only accommodate a range of 0-250 mg/100 ml. Increasing therange of such a ferricyanide system could not be accomplished withouta further increase in effective sensitivity error at 50 mg/100 ml. Theglucose oxidase-peroxidase methods have the advantage of inherently

better sensitivities at low glucose concentrations. The published dataof other investigators regarding their glucose oxidase-peroxidasemethods suggest that the higher noise level of their methods results ina lower effective sensitivity (higher effective sensitivity error) thanthat of the present method in spite of the smaller ranges of theirmethods (15-19). The higher effective sensitivity and range of thepresent method were necessary for a system which was to be used for

monitoring the unusually high and low levels of blood glucose that wehave observed in ulistable diabetics (Fig 9).

Specificity

The response of the analytic system is less than 1% of its responseto glucose when tested at 100 mg/100 ml with the following sugars:D-galactose, D-fructose, n-mannose, n-ribose, D-xylose, L-xyloSe, sucrose,and lactose. With maltose the response is 1.3%. The only knownexception to the specificity of glucose oxidase for D-glUcoSe is 2-deoxy-n-glucose. This is not a usual constituent of plasma and even if present

gives a response of only 12% of that for rI-glucose (21, 22).

The effectof uric acid was tested on solutions of glucose (100 mg/

100 ml) in 0.1 M pH 8.0 Tris and sodium phosphate buffers. Theresponse to glucose in the same solutionbut without uric acid was usedas a control. No effect could be detected with uric acid concentrationsof 0.5 and 5 mg/100 ml. At 50 mg/100 ml, 10 times the normal value forserum, there was 1.2 and 0.7% inhibition in the Tris and I)hoSPhate

buffers,respectively.This inhibition is equivalent to only 0.02% for

each 1 mg of uric acid per 100 ml of solution. Other investigators (13,

15, 16, 19) have reported inhibitions ranging from approximately 0.5 to

690 ROSEVEAR ET AL Clinical Chemistry

1% for each 1 mg of uric acid per 100 ml of solution when testingsolutions containing about 100 mg of glucose and 10-20 mg of uric acid

per 100 ml. These investigators used either o-dianisidine at pH 7.0 oro-toluidine at pH 5.0. We also noted a similar level of inhibition by uricacid (1% per milligram uric acid per 100 ml) when our system wasused with pH 7.0 enzyme-dye reagent. The use of pH 5.0 enzyme-dyereagent decreased the inhibition fifty-foid but resulted in a decrease(16%) in color response, compared to the pH 7.0 reagent. In spite of

this decrease ill color response, the effective sensitivity of the presentmethod, even with the pH 5.0 enzyme-dye reagent, is greater than thatof other methods (as discussed above under Sensitivity).

The effect of ascorbic acid was tested 011 solutions of glucose (100mg/100 ml) in 0.1 M pH 6.0 sodium phosphate buffer. The responseto glucose in the same solution but without ascorbic acid was used asa control. No effect could be detected with 0.06 and 0.6 mg/100 ml. At6 mg/100 ml, 10 times the normal value for plasma, there was a 7%inhibition, about 1%/mg/100 ml. Faulkner (16) reported a 20% inhibi-tion at 20 mg/100 ml (also equivalent to about 1%/mg/100 ml). At

physiologic levels (1 mg/100 ml) or less, ascorbic acid should produceno significant inhibition.

Glucose oxidase methods measure predominantly the $ form ofglucose (20). At the increased temperature (50#{176})of the present method,an appreciable part of the form wouid be converted to tile $ form asthe $ form is oxidized (20). Furthermore, the glucose of plasma isbelieved to be a thermodynamic equilibrium of the and fi forms (23).

Consequently, the glucose solutions used for standardizing the presentmethod were allowed to stand at least 24 hr at room temperature beforethey were analyzed to ensure that they contailled an equilibrium mix-

ture of the and $ forms comparable to that of plasma.

Stability

The stabilityof the analytic system can be illustratedby data from

a typical continuous monitoring which ran for 24 hr. During this time,the blank, which started at 0.010 absorbance units, increased smoothlyand progressively to 0.027 absorbance units. The reading for a glucose

standard (after correction for blank) of 100 mg/100 ml showed notrend or drift during this period but remained 0.136 with a standarddeviation of ±0.001 absorbance units. The variation reported for theother methods, for no more than half the range and for periods nolonger than several hours, is at least twice that obtained for the present

method.

Vol. IS, No. 8, 1969 CONTINUOUS BLOOD GLUCOSE 691

One of the reasons for the low coefficient of variation obtained forthe present method is undoubtedly the completeness of the responsepermitted by the sampling time of 6 mm. A sampling time of 3 mmwould have given a coefficient of variation almost as low, because, asdiscussed below under Dynamic Response, a response which is 99%

complete can be obtained with a sampling time of 135-180 sec. Samplingfor periods which are appreciably shorter than that required for a

complete response would have resulted in a greater variation. Theresponse in this case would be influenced also by variations in the rate

of approach to complete response. The present method would be subjectto this additional variation if it were used to analyze discrete specimensin a manner that did not permit 180 sec or longer for the aspiration ofeach sample.

The present method, with its high degree of linearity, sensitivity, andstability, requires a sampling rate of only 0.1 ml/min to obtain theillustrated precision. The low sampling rate was specifically in-corporated into the present method to avoid the undesirable loss ofblood that would occur with higher sampling rates during continuousmonitoring of human subjects for periods of 48 hr. A sampling rate offour times that of the present method is required for the most suitable

of the previously published methods (13, 17, 19).

Usually, there is some loss of sensitivity and linearity after a periodof 20-24 hr. When sensitivity decreases by 10%, the dialyzer and alllines are rinsed with 1% 112S04, which restores sensitivity and linearity.Occasionally, the sensitivity and linearity are maintained adequatelyfor a period of 48 hr of continuous operation. In the present method, aglass filter in the enzyme-dye line allowed sensitivity and linearity to

be maintained for longer periods and prevented escape of particulatematter into the manifold. (Prior to use of the filter, the reaction mixtureline occasionally became plugged.)

Sensitivity and linearity are monitored by running standardizationsamples at convenient intervals. When possible, standardization isperformed at two concentrations (0 and 100 mg/100 ml) every hour,and at these plus an additional two concentrations (300 and 500 mg/100 ml) every 6 hr. Each standard solution is sampled for at least180 sec. Standard solutions are introduced into the sample line be-

tween the two pumps, permitting collection of blood to continue duringstandardizing sequences. Standard solutions (50, 100, 200, 300, 400, and500 mg/100 ml) are analyzed before and after each experiment andafter each rinse to provide initial and final calibrations. At thebe - inning and end of each experiment, the analytic system was cali-

692 ROSEVEAR ET AL Clinical Chemistry

brated by introducing standard solutions into the sample line before thefirst pump and by the more usual procedure of introducing standardsolution into the sample line between the two pumps. With carefulselection of the pump manifold tubing, the calibration by the twomethods usually agreed within 2%. For applications in which thecalibration was especially critical, both the standard solutions and thesamples were introduced into the sample line before the first pump.

Both interpolation between the times of standardization and betweenthe levels of standardization are performed by means of a digital com-puter to facilitate processing the large number of values obtained from

the continuous monitoring records. First, the values read from thestrip record are converted to blood glucose values. Then, Newton’smethod of divided differences is used to interpolate between standardconcentrations to take into account the nonlinearity which, while small,may exceed experimental error. A linear interpolation scheme is usedto cofllbille tile glucose concentrations calculated from the initial andfinal calibrations. Any number of standard points (two or greater) can

be used for the standard calibrations. Intermediate calibrations of ablank and one standard solution are combined with the initial and finalcalibrations in performing the time interpolation. Even though thesensitivity of the analytic system is usually very stable, significant andunexpected changes in sensitivity can occur. The errors resulting fromsuch changes in sensitivity are greatly minimized by the interpolation

scheme.

Dynamic Response

The delay time from a change in blood glucose level in the subject tothe first detectable change in the recording apparatus is 15 mm. It

takes 5 mm for blood to traverse 10 ft of Teflon catheter and 10 mmfor transit through the analytic equipment. The delay time of 10 mm intile recording apparatus was quite permissible for the intended ap-plication of the present method. Reduction of the delay time may beessential for other applications. This may be difficult to accomplishwithout decreasing the stability and sensitivity, and consequently theprecision, of the present method.

Figure 7 shows the percentage lack of attainment of steady state dueto tile delayed response of the system when the glucose concentrationof tile solution being sampled through an 18-in, line (Tygon transmis-SiOfi tubing, 0.063-in. ID, 0.031-in. wall) is changed abruptly from oneconstant level to another. Tile time required for reducing the lack of

attainment of steady state by half can be referred to as the half-wash

40

Time , seconds

60 80 100 20 40 60 80

OAscending Limb l/2wI7sec Lag Time23secX Descending Limb I/2w I6sec Log Time 23 s.c

Square Wove /2w. 23sec Log Time 22 sec

Vol. IS. No. 8, 1969 CONTINUOUS BLOOD GLUCOSE 693

time (24). The half-wash times are 17 and 16 see, respectively, for theascending and descending (step function) responses. All ascending anddescending responses of glucose standards varying in concentrationfrom 50 to 500 mg/100 ml also have half-wash times between 16 and

00

60

40‘a

3o

20

10

IiFig 7. Lack of attainment of steady state due to delayed response of analytic system whea

glucose concentration of solution being sampled is abruptly changed from one constant level

to another (with an 18-in,sample line of 0.063-in. ID Tygon tuhing). Represented arc: (1) log

scale percent lack of attainment of steady state for ascending and descending limbs of (step

function) changes from complete response at one level to complete response at new level

plotted against elapsed time from first detectable response and (2) similar values calculated

from differences between maximal and minimal readings resulting from repetitive (square

wave) changes plotted against half-period times for these changes.

17 sec.* Note that, after a lag time of 23 see, the log percentage lack ofattainment of steady state is approximately linear. Twenty-five percentof the new level is achieved in 30 sec and 90% in 80 sec. Decreasing theTriton X-100 concentration in the recipient solution from 10 to 0.5 ml/liter further delays the attainment of the final 20% of the steady state,

as shown in Fig 6. With the higher level of detergent, there is a similardelay in the attainment of the final3-5% of the steady state (Fig 7).The dynamic response of the system is not altered appreciably by theaddition of a 10-ft catheter (Teflon, 0.015-in. ID) to the 18-in, sample

*FOr the ferricyanide method, used in the routine clinical chemistry laboratory at the Mayo

Clinic,a half-wash time of 11 sec has been obtained. The published methods for continuousmonitoring (2-8, 18, 19) appear to have somewhat longer half-wash times, although the dataare not detailed enough for definitivecomparison.

694 ROSEVEAR fiT AL Clinical Chemistry

line; the half-wash times are increased from 17 and 16 sec to 22 and 22see, respectively, for the ascending and descending responses.

With the method of evaluating the dynamic response described

above, the time from the initialintroduction of one constant level of

sample to the introduction of a new constant level of sample is long

enough to allow a complete response to the first level before theresponse to the second level begins. This separates the response to eachlevel into separate lag, linear, and trailing phases. If the time werenot long enough to allow a complete response at the first level, thetrailing phase of the first level would begin to overlap the lag phaseof the second level. Further decreases would lead to progressive over-lapping of all the phases until the responses to the first and secondlevels would be indistinguishable. A desirable characteristic of anycontinuous monitoring system would be the ability to faithfully rep-resent fluctuating, as well as constant, levels. The effect of such over-

lapping of responses was tested by alternately sampling a 50 mg/tOO mlsample and a 200 mg/100 ml sample (with the 18-in, sample line) for aseries of constant time intervals. The time each solution was sampled

can be considered as a half-period of a square wave. The process wascontinued with the same time interval until consistent maximal andminimal responses were obtained. Time intervals of 30, 45, 60, and 120see were used. The difference between the maximal and minimal

responses were measured and expressed as the percentage of thatobtained when the time interval was long enough (6 mm) for a completeresponse at each level. The results are included in Fig 7. The complexoverlapping of the square-wave responses has resulted in slower attain-ment of a complete steady state response than would have been pre-dicted from the response to a simple step function. As illustrated, we

can detect 25% of the magnitude of any square-wave oscillation with ahalf-period of 30 sec and 50% of the magnitude of any similar oscilla-

tion with a half-period of 45 sec. In our records of continuous bloodglucose measurements, we have not seen oscillations with half-periods

as short as 30 sec. 1Vithout the innovations of the present method,especially the Teflon transmission tubing and the high level of detergent,oscillations with half-periods as short as 30 sec could not have beendetected.

Application

Over the past 2 years, we have studied 12 diabetic and 6 normalsubjects by continuous blood glucose analysis during brief, as well asprolonged, experiments. Our total experience now exceeds 2000 hr ofmonitoring. All subjects were heparinized to avoid obstruction of the

Vol. 15, No. 8, 1969 CONTINUOUS BLOOD GLUCOSE 695

catheters by clots.* The catheter (Teflon, 0.015-ill. ID) was inserted,under sterile conditions, into the antecubital vein and was advanced sothat it could be expected to lodge in a major intrathoracic vein. Thelength (up to 10 ft) of the catheter prolonged (30 sec/ft at a samplingrate of 0.1 ml/min) the time necessary for the patient’s blood to reach

the analyzer. This delay was overshadowed by the advantages ofpermitting near-normal activities such as walking about the room,sitting in a chair, eating, drinking, and performing ablutions.

The advantages of continuous monitoring of blood glucose are thedetection of changes in blood glucose concentration while an experi-ment is in progress and the production of a permanent record foranalysis later. During the studies, symptoms and signs related tohypoglycemia or to rapidly decreasing or increasing blood glucoselevels can be observed. The most suitable times for taking additional

discrete specimens of blood or urine for measurement of other param-eters can be determined from the continuous blood glucose record.

Continuous recording also facilitates controlled decreasing or in-creasing of blood glucose concentratioll with accuracy and safety. Forexample, as illustrated in Fig 8, a hyperglycemic diabetic patient canbe brought to a sufficient degree of hypoglycemia, by intravenousinfusion of insulin, to demonstrate his potential for augmented secre-tion of growth hormone. Uniform end pomts of hypoglycemia may beso obtained in experimental subjects without risk of excessive decreasesin blood glucose.

The method for continuous automated blood glucose determination

described in this article has proved to he particularly suitable forstudying diabetic and normal human subjects. The method is con-venient. It is suitable for use in the subject’s room. It has producedprecise and essentially continuous measurements over periods as long as50 hr. It has provided the investigators with the subject’s blood glucoseconcentration, while an experiment was in progress, in a manner thatcontributed to the satisfactory completion of the experiment.

Examples of data provided for analysis of blood glucose dynamics

during waking, sleeping, ingestion of various foods, and exercise are

Systemie administration of heparin has been avoided by pumping a solution of heparin

into the stream of blood being aspirated by way of the intravascularly placed catheter(f, 6, 12). We, as well as others (3), have noted interruptions in blood flow due to clotformation in the catheter tip when blood is heparinized extraeorporeally in this manner.

Another problem encountered has been the variable and unpredictable dilution of the aspiratedblood with the heparin solution (6, 12). Both problems were avoided by administering theheparin to the patient. Nonetheless, we recognize that for some studies the problems associatedwith the effects of heparin may exceed those associated with occasional clot formation andloss of precision due to extracorporeal heparinization.

0700 1100 1500 1900 2300

Time - hours

696 ROSEVEAR ET AL ClinicalChemistry

5964 7255

500 50 D2 III 11-9-66 Insulin0.1 Ii / kg / hr #{149}

45045

40040

35035 I

#{149}2345’6789 10Hours

Fig 8. Continuous record of blood glucose (solid line) with intermittent determinations of

plasma growth hormone (HGH) concentrations (broken line) in diabetic subject during

insulin-induced hypoglycemia test of ability to augment HGH secretion.All recorded values

and events are coordinated with time scale (elapsed time, in hours). Glucose curve is based on

values taken at 5-mm intervals from continuous recording of optimal densities. Start and endof insulininfusion, 0.1 unit/kg/hr, are indicated by arrows.

a

N

Fig 9. Nyctohemeral blood glucose patterns of a brittle (dotted line) and a stable (solid

line) diabetic and a normal control subject (broken line) studied under matching conditions.

Time scale is in 24-hr clock time. Curves are based on values taken at 5-mm intervals from

continuous recording of absorbance. B represents 1 hr of exercise; B, breakfast; L, lunch;S, afternoon snack (1600 hr) or bedtime feeding (2300 lir); D, dinner. Diabetics receivedinsulin once daily, at 0745 hr.

Vol. 15, No. 8, 1969 CONTINUOUS BLOOD GLUCOSE 697

illustrated in Fig 9. The nyctollemeral blood glucose patterns of abrittle diabetic, a stable diabetic, and a normal control subject, allstudied under identical conditions, are reproduced in this figure. Inter-mittent sampling of blood to obtain the same information would notonly be much more costly in terms of the patient’s blood and in theinvestigator’s efforts but also would interfere physically with the sub-ject’s freedom of movement. Last, but not least, the continuous record

obviously minimizes the omission of brief but potentially importantportions of the blood glucose pattern.

References

1. Molnar, G. D., Gastineau, C. F., Rosevear, J. W., and Moxuess, K. E., Quantitative aspectsof labile diabetes. Diabetes 14, 279 (1965).

2. Weller, C., Linder, M., Macaulay, A., Ferrari, A., and Kessler, G., Continuous in vivodetermination of blood glucose in human subjects. Ann. N.Y. Acad. Sci. 87, 658 (1960).

3. Mirouze, J., Jafilol,C., and Sany, C., Enregistrement glycomique nycthmemeral continudans Ic diabete instable.#{163}ev.Franc Endocrinol. Clin. 3, 337 (1962).

4. Kadish, A. H., Physiologic monitoring of blood glucose. Calif. Med. 98, 325 (1963).5. Butterfield, W. .1.H., Sargeant, B. M., amid Whichelow, M. J., The metabolism of humnami

forearm tissues after ingestion of glucose, fructose, sucrose, or liquid glucose: A studyby continuous in-vivo autoanalysis.Lanect 1, 574 (1964).

6. Burns, T. W., Bregant, R., Van Peenan, H. J., and Hood, T. E., Evaluation of the oral

glucose tolerance test by a continuous snmnplmg technique. J. Lab. Clin. Mcd. 65, 927(1965).

7. Brown, G. M., Zachwieja, A., and Stancer, H. C., An improved technique for continuousin vivo analysis of glucose. Clin. Chim. Acta 14, 386 (1966).

8. Azerad, E., and Duprey, J., Etude des tests de tolerance an glucose par In m#{233}thode

d’enregistrement autonoatique continu. I. Note prCliminaire: Critique de la mCthode.Path. Biol. (Paris) 15, 485 (1967).

9. Middleton, J. E., and Griffithus,W. J., Rapid colorimactric micro-method for estimatingglucose in blood amid C.S.F. using glucose oxidase. Brit. Med. J. 2, 1525 (1957).

10. Kingsley, G. B., and Getchell, G., 1)irect ultramicro glucose oxidase umethod for determina-tion of glucose in biologic fiui(ls. Clin.. Cheni. 6, 466 (1960).

11. Fingerhut, B., Ferzola, B., amid Marshi, W. H., Application of a ferrocyanide-phospho-molybdate reaction to an automated determnimmation of serum glucose. Cliii. Chim. Acta.8, 953 (1963).

12. Burns, T. W., Bregant, R., Van Peenan, H. J., and hood, T. E., Observations on bloodglucose concentration of human subjects during continuous sampling. Diabetes 14, 186(1965).

3. 11111, J. B., and Kessler, G., An automated determination of glucose utilizing a glucoseoxidase-peroxidase system. J. Lab. Clin. Med. 57, 970 (1961).

14. Discombo, G., An inexpensive method for the estimation of true glucose in blood amidother fluids by the AutoAnalyzer. J. CUe. Path. 16, 170 (1963).

15. Getchell, G., Kingsley, G. B., and Schaffert, R. B., Direct autonmated determination of

glucose by a glucose oxidase-peroxidase system. Clin. Chc,n. 10, 540 (1964).

16. Faulkner, D. E., An automated micro determination of blood-glucose with the Auto-

Analyzer. Analyst 90, 736 (1965).

17. Robin, M., and Saifer, A., Determination of glucose in biologic thuds with an automated

enzymatic procedure. Clin. Cheni. 11, 840 (1965).IS. Kadish, A. H., and Hall, D. A., A new method for the continuous monitoring of blood

glucose by measurement of dissolved oxygen. CUn. Chem. 11, 869 (1965).19. Kawerau, E., Die enzymatische Blutzucker-Bestimmung in vitro und in vivo malt dem

Auto-Analyzer. Z. Kim. Chem. 4, 224 (1966).

698 ROSEVEAR ET AL Clinical Chemistry

20. Keilin, D., and Hartree, E. F., Specificityof glucose oxidase (Notatin). Biochem. J. 50,331 (1952).

21. McComb, B. B., and Yushok, W. D., Colorimetric estimation of D-glucose and 2-deoxy-D-glucose with glucose oxidase. J. Franklin Inst. 265, 417 (1958).

22. MeComb, B. B., Yushok, w. D., and Batt, W. G., 2-Deoxy-D-glucose, a new substrate for

glucose oxidase (glucose aerodehydrogenase). J. Franklin Inst. 263, 161 (1957).23. Power, M. H., and Greene, C. H., The nature of the blood sugar as shown by a comparison

of the optical rotation and the reducing power of the in vivo dialysate.J. Biol. Chem.94, 295 (1931).

24. Thiers, B. E., Cole, B. B., and Kirsch, W. J., Kinetic parameters of continuous flow

analysis. Clin. Chem. 13, 451 (1967).