| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Articles |
1
Department of Clinical Chemistry, Vejle County Central Hospital, DK-7100 Vejle, Denmark.
2 Department of Clinical Chemistry, Fredericia
Hospital, 7000 Fredericia, Denmark.
3 General practice, 6000 Kolding, Denmark.
a Author for correspondence. Fax + 45 75 82 18 14; e-mail bbo1497{at}vip.cybercity.dk
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
At present, there are no recommendations in Denmark concerning relevant use of B-glucose analysis and no quality demands, in contrast to US practice (1). The introduction of a variety of different small instruments for performing near-patient testing in primary healthcare has revealed problems regarding quality assessment, quality assurance, and documentation.
Previous proficiency testing schemes for B-glucose have been hampered by the instability of glucose in whole blood. The stabilizers used interfere with some bedside test principles, according to instruction manuals for One Touch (LifeScan, Johnson & Johnson), Elite (Bayer Diagnostica), and Reflolux 2 (Boehringer Mannheim). Serum, which is more stable because of the absence of cells, is also unsuitable for some test methods (instruction manual for Companion Meter; MediSense). Therefore, in considering a new procedure for external glucose assessment, we formulated the following requirements and specifications for a control material and procedures:
1. Whole blood.
2. Anticoagulation without test interference.
3. No addition of antiglucolytic agents.
4. Three glucose concentrations in control material, from 2 to 20 mmol/L.
5. No need for stable glucose or for knowing glucose value prior to analysis.
6. Control material capable of yielding realistic and reliable estimates of performance with patients' samples, and reflecting the whole technical procedure as carried out in the office (including sampling of specimens, an important source of imprecision).
7. Results expressed in relation to a comparison method by difference plot (there being no known "true" value determined).
From these specifications the materials and procedure were chosen and were tested for fulfillment of the criteria. Initial results with the system led to use of the method over a 5-year period—to assess both the quality of glucose measurements in general practice and the effect of the quality-assurance program so as to justify the expenses incurred.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
control materials
Whole blood from a volunteer donor was anticoagulated with 12–30
IU/mL Li-Heparin (Monovette; Sarstedt) and was partially depleted of
glucose, to ~2 mmol/L, by standing overnight at 22 °C. The next
morning we added glucose to 2 of the 3 aliquots of this sample to yield
3 different glucose concentrations (~2, 7, and 15 mmol/L), and all 3
samples were taken to the GP offices by a technologist.
patients' samples
Capillary samples were taken from 2 to 4 visiting patients in each
clinic, the same day, in the usual way, for the purpose of documenting
the validity of this approach and also for determination of glucose
content by both the method used in that clinic and the laboratory
method. According to our procedure there was no need for the glucose
concentration to be known in advance; we stabilized the concentration
by adding hemolyzing agent at the same time that B-glucose
concentration was being measured in the GP office.
gp office procedures
Control samples.
The three aliquots prepared were
carried by teaching laboratory technologists to the GPs and given to
the GPs' staff, who determined B-glucose concentrations on these in
duplicate, using their own instruments. At exactly the same time, the
laboratory technologist mixed 20 µL of each control blood in 1 mL of
hemolyzing solution (cat. no.13888; Merck), also in duplicate. These
test samples represented the "true" glucose concentration exactly
at the time of analysis, because glycolysis was stopped and the
hemolysates were stable for at least 24 h (2).
Moreover, the results obtained by the laboratory method were regarded
as "true" values because results by this method in national and
international external proficiency programs showed a bias range of only
between -2% and +2% (see Table 2
).
|
Patients' samples.
At exactly the same time, and from
the same drop of blood from the patient, the laboratory technologist
added 20 µL of blood sample to 1 mL of hemolyzing solution, also in
duplicate.
hospital laboratory glucose method
On the same day, the glucose concentration in hemolysates from the
control samples and from the patients' samples from each GP was
measured with a highly precise and highly accurate method in the
Department of Clinical Chemistry by the glucose dehydrogenase method
(3). The reagent used was Unimate 7 Gluc GDH (Roche
Diagnostica), analyzed with the Cobas Fara or Cobas Mira instruments
(Roche). The method was calibrated against NIST 909a Standard Reference
Material (4).
definitions (5) and data analysis
Imprecision (SD) was determined from duplicate determinations
of the control samples and patients' samples by each GP office method.
The percentage bias of a method was defined as the systematic deviation
from the "true" value (hospital laboratory value), expressed as the
difference between many measurements with the investigated GP office
method and the "true" value.
The difference between the values obtained for a sample with the method under investigation and by the comparison method was calculated as the % difference between the first bedside test result and the mean of the duplicate determination in the laboratory. The first GP office determination was used because that is how results are produced in daily work there. The total error of a method, as a descriptive, statistical parameter, was expressed as bias ± 1 SD, which includes the method's systematic deviation from the true value and the imprecision of the result obtained with that method.
Previously published criteria for evaluating tests (6) cannot be applied, because the control materials do not have assigned values but instead are measured with a comparison method/instrument. Hence % differences are calculated from the difference plot (7), for the control materials as well as the patients' samples.
instruments
The bedside instruments used in 1992 were: Accutrend, Hypocount,
and Reflolux (all from Boehringer Mannheim); Glucometer (Bayer Denmark
A/S); HemoCue (HemoCue Denmark A/S); and MediSense. Since then, the
technology has been extended with the Elite, One Touch, and Reflotron
instruments. The measurement range of these is generally from 2 to
~25 mmol/L glucose. The HemoCue is mostly sold as a laboratory
instrument; the other systems are primarily sold to patients and used
for self-monitoring.
In all, during 1992, 338 single and 265 duplicate determinations were performed by the usual staff for 171 doctors in 98 offices, with use of 98 instruments of 6 different types. In 1994, 103 GP offices participated and in 1996, 120. The results of glucose measurements on control samples and capillary patients' samples by each instrument type were pooled and results by the different instruments were evaluated separately.
The validity of the control materials for use in this concept of performing external quality assurance was evaluated against actual patients' specimens by comparing bias and precision results for both types of blood for each of the different types of office instruments. In this part of the investigation we used the mean of duplicate determinations of both doctors' and laboratory results.
economy
Proficiency testing for B-glucose is a part of a general program
for quality assurance of laboratory analysis in general practice in
Vejle County. Four laboratory technologists and one chemist are
involved in the project, an equivalent of one full-time employee being
allocated to the program. The longest distance between a GP office and
the nearest department of clinical chemistry is ~45 km. Again, all
expenses are paid by our County's health authorities and no extra
expenses are incurred by the GP offices.
ethics
The procedures followed in this study were in accordance with the
Second Helsinki Declaration (amended in 1989), Danish law on biomedical
research, and the Scientific Ethical Committee system of October 1,
1992. According to this, approval by a scientific ethical committee is
not needed if only a part of the material is used for technical or
quality-assurance purpose and no additional blood samples are taken.
The local scientific ethical committee was informed about the study.
All involved patients gave their informed consent before participation.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The average % bias for both types of samples and for each instrument
type is shown in Table 1
. Use of the control material was valid for proficiency testing
for 4 instruments, giving differences that were not significant. For
Accutrend and Exactech/MediSense, the possibility of systematic error
caused by a matrix effect with the control material should be
considered. (We observed that Exactech/MediSense is very sensitive for
oxygen in the sample; this is not a problem when capillary samples are
measured, but can be a source of error when venous blood is tested. An
effective way to avoid this error would be to oxygenate samples just
before measurement by extended mixing.) Nonetheless, the numerical
differences were small (Table 1
) and, though statistically significant,
were considered to be of minor importance for the evaluation program.
Consequently, the results from capillary blood and venous control
samples were pooled.
|
Comparison of the results from the capillary and control material in 1994 and 1996, when new instruments types were tested, also showed no significant difference between these two types of control materials, and the results for both by each new instrument type were also pooled.
evaluation of instruments
To evaluate the instruments, we used the difference between the
first result in the doctor's office and the mean of duplicate
determinations with the laboratory method for both control materials
and capillary samples for calculations of total error (which also
includes imprecision of the method because the determinations reflect
realistic circumstances of a single measurement in the doctor's
office). The imprecision itself, however, was also evaluated on the
basis of true duplicate determinations.
The imprecision of the different instruments is stated in Table 2
. In the reference interval for glucose, HemoCue gives results
comparable with those by the laboratory method; i.e., 1 SD = 0.11
mmol/L. The other instruments have imprecision between 0.26 and 0.48
mmol/L—values that are of course influenced by the limited number of
determinations.
Table 3
shows the total error results at all three control
concentrations. Table 4
shows the frequency of results falling outside the acceptance
range (i.e., ±20% from the comparison method value; see
Discussion). In 1992 there was no significant difference
between the different instruments, but none could fulfill the goal of a
maximum of ±20% as acceptable error.
|
|
effect of the quality program on performance
In the second round, after presentation of results from the first
round to the GPs, many offices replaced instruments and improved
procedures, assisted by laboratory technologists. Results in the second
and third rounds (1994 and 1996; Table 4
) showed considerable
improvement.
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Because we deemed both the control material depleted in glucose and supplemented with glucose to suitable concentrations to be reasonably comparable with capillary blood taken from patients, the material could be used for quality evaluation of instruments of different types. The design of the control procedure involves the entire testing procedure in GP office; thus this quality assessment monitors not only instrument performance but also variations in practical handling, as the process was designed to closely resemble daily practice. (Only the blood simultaneously taken for determination by the comparison method was drawn by a professional laboratory technologist.)
An advantage of this procedure is that it can be performed over a span of time. The use of concentration-independent materials makes timing flexible, and the process can be arranged at the convenience of the individual GP office and the laboratory. Furthermore, a GP showing unsatisfactory results can be assisted, corrected, and instructed, and the external quality assessment can easily be repeated just for that one.
The American Diabetes Association has recommended a maximum error of 10% (from a reference measurement value) for future instruments and 15% for existing instruments (8). The NCCLS recommends that the discrepancy between the bedside result and a reference measurement should be <20% (9). Recently, the ±20% limit has also been used by the FDA in evaluation of over-the-counter B-glucose meters (10). The general opinion among Danish medical professionals is that a ±10% deviation from a reference value is good, between 10% and 20% is acceptable, and greater than this is unacceptable. In other words, it is an absolute demand that no single result is allowed to exceed the ±20% limit.
For monitoring the treatment of diabetes, therefore, a CV <10%,
corresponding to SD <0.7 mmol/L, seems reasonable (11).
All meters can comply with this (Table 2
). However, this only refers to
the single instrument/practice, and as can be deduced from SD on total
error, shown in Table 3
, the variation from practice to practice for
use of the same instrument type far exceeds the single instrument's
imprecision in reproducing its own results. This is important if a
patient is monitored with different meters.
According to these criteria, the first proficiency testing round in
1992 showed unsatisfactory performance. Even though in our opinion the
range of the area for acceptance (±20% as a total error) is extremely
large, in 1992 no instrument type could comply (Table 4
), and hence at
that time we could not recommend diagnosis of diabetes in GP offices on
the basis of results by such methods.
All results were anonymously reported back to the GPs, who evaluated their own results. Instrument vendors were also informed about results from the first round. They were very responsive, which in some cases resulted in changing calibrations, developing and standardizing the glucose test sticks, and improving manuals and instructions. We were very pleased to note this active interest and participation from commercial companies to improve the technical quality.
All doctors were visited by a laboratory technologist. As a
consequence, many bought new instruments and asked for instruction from
the central laboratory. After this, it was agreed to repeat proficiency
testing. Results of the second proficiency testing round in 1994 showed
considerable improvement (Table 4
). We take this as an indication of
the effect of a formalized quality-assurance program, which prompted
doctors to improve procedures for handling patients, specimens, and
testing.
The improvement can be seen clearly in the low frequency of results
exceeding ±20% in the 1994 survey (Table 4
). Several instrument types
(Accutrend, HemoCue, One Touch) have no values outside this range,
despite the greater number of determinations made. The quality level
was maintained through the 1996 survey.
We attribute this considerable improvement for instrument–practice combinations to both better and more- standardized calibration of the machines, the mean total error being much reduced for Accutrend and HemoCue, and improved precision, probably induced by standardized sample handling after professional instruction. The fact that many GP offices switched to the HemoCue, which has better precision, also may account for part of the improvement.
Table 4
shows a reduction in the SD of the mean for total error, both
on the single machine and on mean bias %. Duplicate determinations
confirm this: In 1996, the imprecision for Accutrend,
Exactech/MediSense, and HemoCue was 0.25, 0.38, and 0.26 mmol/L,
respectively (Table 2
). One instrument, Hypocount, was no longer in use
in 1996.
According to WHO (12), the lower limit for fasting
B-glucose in diabetes is 6.7 mmol/L. The broadly accepted upper
reference limit for fasting B-glucose in nondiabetes is 6.0 mmol/L, a
difference of 0.7 mmol/L. Thus, if a patient who has a true blood
glucose value of 6 mmol/L should not falsely be classified as a
diabetic, the glucose test should never yield a value >6.7 mmol/L for
that patient's blood sample. From this and previous publications
(8)(9)(11)(13), and
considering this difference to be clinical significant, we have
formulated the following local performance goal for quality in glucose
measurements to meet clinical needs: For diagnostic purposes, maximum
allowable total error (e.g., numerical value of bias + 2 SD) for the
single instrument at 6 mmol/L should not exceed 0.7 mmol/L (12%).
Generally, as it can be seen from Table 4
, bedside instruments cannot
comply with this demand. Most will produce false-positive or
false-negative results, leading to incorrect diagnosis for the presence
or absence of diabetes.
The problem of diagnosing hypoglycemia is outside the scope of this
paper; but given that the lower acceptable glucose value for diabetics
in treatment is 5 mmol/L and a further decrease to 4 mmol/L is expected
to be important, an instrument or method should be capable of detecting
this difference of 1.0 mmol/L (13)(14). The
critical difference between two consecutive results, i.e., the smallest
change that makes them significantly different (P <0.05),
has been defined as 2.77 x CV (15). For many GP
office instruments, the CV at 4–5 mmol/L is 10% (Table 2
). Hence,
they can detect a change of 2.77 x 10% = 27.7%, corresponding
to 1.4 mmol/L (from 5 to 3.6 mmol/L), but not a decrease from 5 to 4
mmol/L.
In fact, glucose measurements on diabetes patients are used extensively, though not recommended by any Danish authority. Historically, GPs have handled diabetic patients for many years, and self-monitoring has been only slowly adopted by patients. We agree with American Diabetes Association (1) (and so do Danish diabetes specialists) that B-glucose should not be used by GPs to assess glycemic control in diabetes patients, but the tradition is difficult to change. Rather, the use of the described instruments should be restricted to screening for severe hypo- or hyperglycemia only and to monitoring glucose >5 mmol/L in already diagnosed diabetics.
Our experience with these small instruments is limited to GP offices only. Many diabetics use the same type of devices (except HemoCue, which is not handy enough and is too expensive for home use). Precision of all these instruments is satisfactory for self-monitoring. However, there is a big difference between use of an instrument by the same patient who uses it in the same (even if not quite correct) way every time, or by different staff at GP offices. And accuracy of home measurement is still a problem.
In GPs' offices, these instruments may be useful for educating diabetes patients in the understanding of their diseases, but their lack of accuracy and precision makes it impossible for the GP to check a patient's own reading on the patient's own instrument, and the combined error can produce severe differences in many instances.
With regard as to whether this type of proficiency testing program can be used generally, 98 GP offices (later 120, almost all) in Vejle County have participated voluntarily in a quality-assurance program involving visits by specially trained laboratory technologists. In connection with these annual visits, where we try to help the GPs' offices and solve their problems, we prepare the control samples and take them to the GPs' offices. Because this is also a part of our own quality-assessment program, it is difficult to calculate exact expenses concerning just this part of proficiency testing, but the cost could be prohibitive in sparsely populated areas because of the transportation expenses involved.
In conclusion, we have shown that this program for external quality assurance of GP office analyses is possible and also effective, in 5 years having contributed considerably to quality improvement in measurements of blood glucose in those offices. We propose that this system could be generally useful also in the hospital setting with decentralized satellite laboratories.
|
Acknowledgments |
|---|
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The following articles in journals at HighWire Press have cited this article:
|
|
 |
|
  R. Haeckel, U. Brinck, D. Colic, H.-U. Janka, I. Puntmann, J. Schneider, and C. Viebrock Comparability of Blood Glucose Concentrations Measured in Different Sample Systems for Detecting Glucose Intolerance Clin. Chem., June 1, 2002; 48(6): 936 - 939. [Full Text] [PDF]
|
|
|
|
|
|
M. du Plessis, J. B. Ubbink, and W.J. H. Vermaak Analytical Quality of Near-Patient Blood Cholesterol and Glucose Determinations Clin. Chem., August 1, 2000; 46(8): 1085 - 1090. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. C Delaney, C. J Hyde, R. J McManus, S. Wilson, D. A Fitzmaurice, S. Jowett, R. Tobias, G. H Thorpe, and F D R. Hobbs Systematic review of near patient test evaluations in primary care BMJ, September 25, 1999; 319(7213): 824 - 827. [Abstract] [Full Text]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
