| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Articles |
a Address for correspondence: GSC Consulting, 4913 Bruce Ave., Edina, MN 55424. Fax 612-915-1061, e-mail cembr001{at}gold.tc.umn.edu
 |
Abstract |
|---|
 Top Abstract IntroductionOptimization of Reference Sample...Quality Control Using Patient...Point-of-Care Quality-Control...External Quality Assessment and...Current Quality...References |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionOptimization of Reference Sample...Quality Control Using Patient...Point-of-Care Quality-Control...External Quality Assessment and...Current Quality...References |
|---|
This laboratory metamorphosis has been expensive in terms of the
dislocated and (or) unemployed technologist and laboratory scientist.
The transformation tends to reduce the quality of services of both the
clinical laboratory and the clinical laboratory industry. Table 1
shows the how some of the factors that enabled the
transformation can potentially compromise laboratory quality. The
nationwide drive to reduce costs has resulted in large- scale
replacement of medical technologists by less-trained medical laboratory
technicians, increasing workloads, a higher reliance on general-purpose
float and temporary personnel, and a reduction of supporting technical
staff with the resulting deemphasis of training. The widespread trend
of integrating the high-volume chemistry, hematology, and coagulation
laboratory into a core laboratory can compromise quality as the pool of
highly skilled technologists is reduced. Even the pathologist's role
in the clinical laboratory continues to diminish given that anatomic
pathology, the pathologist's primary responsibility, requires more
attention as it increases in complexity while its reimbursement remains
constant or decreases.
|
In these highly competitive times, the laboratory is more dependent on
the laboratory industry than ever before. The laboratory not only needs
well-designed, efficient instruments but also requires the manufacturer
to continually improve those assays already in use and to provide
robust, transferable reference intervals. Unfortunately, the
manufacturer has been experiencing the changes associated with
downsizing and mergers longer than the clinical laboratory and also may
be delivering less than optimal assistance to its clinical customer
(see Table 1
). To complicate issues further, the perceived need to
reduce costs has increased the power of various purchasing groups,
which in turn has diminished the strength of the relationship between
the clinical laboratory purchaser and the clinical laboratory
manufacturer. Decisions to acquire instrumentation are thus based less
on total quality and more on costs or even the procurement of
additional instrumentation for other laboratory disciplines.
Table 2
lists the clinical laboratory's determinants of high-quality
testing. When well-trained and motivated technologists use high-quality
assays according to standard operating procedures, the end result is
usually highly accurate and precise testing. Because unacceptable
results are sometimes produced even by the best systems, procedures
must be devised to detect and correct the error situation and amend any
erroneous patient results. This article deals with quality control and
the detection and reduction of analytical error. I remind
the reader that most laboratory mistakes occur not during the
analytical phase but before or after testing. For example, Ross and
Boone (1) reviewed 363 incidents that occurred in a
tertiary-care hospital in 1987. In the 337 medical records
investigated, preanalytical (missed or incorrectly interpreted
laboratory orders, improper patient preparation, incorrect patient
identification, wrong specimen container, and mislabeled or mishandled
specimens) and postanalytical (delayed, unavailable, or incomplete
results) mistakes accounted for 46% and 47% of the total incidents,
respectively. Nonlaboratory personnel were responsible for 29% of the
mistakes. Most of these errors were interdepartmental.
|
Quality control may be defined as the control of the testing process to ensure that test results meet their quality requirements (2)(3). Quality control may be practiced prospectively and provide information about the acceptability of the most recent analytical run(s) or may be practiced retrospectively and provide information about past performance. Prospective quality control can involve the statistical analysis of reference samples, the review of patient data, and even instrument-based electronic checks. Retrospective quality control includes external quality assessment or proficiency testing, the use of summary quality-control data provided by a regional or manufacturer-supplied quality-control program, calibration checking of unlike analyzers, and even the follow-up of clinician inquiries.
|
Optimization of Reference Sample Quality Control |
|---|
|
TopAbstractIntroductionOptimization of Reference Sample...Quality Control Using Patient...Point-of-Care Quality-Control...External Quality Assessment and...Current Quality...References |
|---|
Westgard and others soon realized that quality control should be specified by instrument or even by analyte. The application of the six-rule multirule control procedure to highly precise analyses tended to detect small, sometimes insignificant analytical error. Westgard derived formulae to calculate the maximum analytical error that could be tolerated in a measurement process. Assuming an analyte with a maximum medically allowable error (MAE), an analytical process with an imprecision of s, and a maximum error prevalence of 5%, the largest systematic error that could be tolerated was: MAE/s - 1.66. The largest tolerable random error was (MAE/s)/2. Westgard used these maximum systematic and random errors and the power function diagrams to construct quality-control selection grids (8). These grids permitted the selection of optimal quality-control rules for various amounts of tolerable error and error frequency. Westgard is now marketing a computer program (9) that allows the user to specify an analyte, its imprecision, and the MAE (usually the proficiency test limits as specified by the Clinical Laboratory Improvement Amendments of 1988). The program then proposes optimal quality-control procedures for that analyte.
The laboratorian is more motivated to use the optimized quality-control procedure if it reduces the frequency of quality control. Koch et al. (10) showed that the application of optimized analyte-specific quality-control practices reduced the frequency of falsely rejected runs, reduced quality-control expenses, and increased the efficiency of their high-volume chemistry analyzer, the Hitachi 737. Koch et al. replaced a quality-control procedure of the 13s, 22s, and the R3.6s rules applied to 2 controls run for every 18 patients. This new control procedure consisted of the 13.5s rule for sodium, potassium, glucose, and blood urea nitrogen; the 12.5s rule for albumin, chloride, and CO2; and the 12.5s rule for calcium, which was run in duplicate and averaged. Considerable reprograming of the LIS quality-control program and retraining of the analysts were needed for their analyte-specific quality- control procedures to succeed.
The laboratorian is far less motivated to use the optimized quality-control procedure if it increases the frequency of quality-control testing or detects more occurrences of analytical error. Such optimized quality-control systems detect problems with which the manufacturer may not be able to help, thus causing more difficulty in convincing administrators and technologists of the added value of more sensitive procedures. It may be more effective to modify or even replace the analytical procedure for one more stable and requiring less sensitive quality control.
|
Quality Control Using Patient Results |
|---|
|
TopAbstractIntroductionOptimization of Reference Sample...Quality Control Using Patient...Point-of-Care Quality-Control...External Quality Assessment and...Current Quality...References |
|---|
Patient data can be evaluated on an individual basis or grouped to
provide meaningful information about the analytical run. Markedly
abnormal values, those defined as critical or panic, are followed with
reanalysis, checking of previous values, or with expedited reporting to
the clinician (11)(12). Arithmetic checks may
be done within a group of analytes, e.g., anion gap, to determine the
acceptability of the constituent measurements or between a calculated
parameter and the one actually measured (e.g., calculated bicarbonate
from a blood gas measurement and the total serum CO2). If a
patient's previous results are available during the testing of the
current specimen, the LIS can calculate the difference (
) between
the current and previous measurements and indicate a significantly
large . Houwen and Duffin (13) have proposed a
unorthodox approach to the calculation of that results in only one
for a set of current and previous test values:
 | (1) |
In the typical calculation of , the magnitude of the is greatly
affected by the direction of the change. By the calculation of Houwen
and Duffin, the is the same whether the current value increases or
decreases. I believe it is easier to apply than the more prevalent
calculation.
Delta check violations should be investigated before reporting the
current value. Such prospective investigations will prevent the
reporting of erroneous results associated with the current
determination. Although more efficient, investigations that occur after
result reporting will lead to aberrant data reviewed by the clinical
staff. If no errors are detected, the large is usually attributed
to either an analytical error in the first determination, mixup of the
first specimen, or greater than expected intraindividual variation.
Because of improved instrument reliability and the increasing use of
bar-code identification of specimens, the prevalence of analytical
errors and specimen mixups has been decreasing. Large values will
more often indicate a real change in a patient's test values rather
than an error (14). For this reason, check limits
should be considered for review on a per analyte basis whenever a
technologist suggests that the check limits are too narrow and are
resulting in too frequent investigations.
Only a minority of laboratorians, usually hematologists, advocate the use of patient data over reference sample results to determine run acceptability. Most microcomputer-driven hematology analyzers are programed to average consecutive patient erythrocyte indices with a unique smoothing algorithm, named after Brian Bull who first described its use for quality control. As patient erythrocyte indices are symmetrically distributed and usually exhibit few extreme outlying values, the averages of as little as 20 observations are quite stable. Changes in serial average erythrocyte indices thus can indicate errors in the component measurements hemoglobin, erythrocyte count, and mean corpuscular volume. Simulations of Bull's averaging technique indicate that a minimum of 40 to 60 specimens must be analyzed daily for the method to have any error detection capabilities (15)(16). As such, the averaging technique is not useful at start-up or after maintenance nor is it useful with instruments requiring frequent calibration. Smith and Kroft (17) have provided more general simulations of similar averaging techniques.
Although the averages of patient clinical chemistry data were
originally described for quality control over 30 years ago
(18), a series of investigators found that the technique
lacked the error-detection capabilities of reference sample quality
control. The error-detection capabilities of patient averages depend on
multiple factors (19) with the most important being the
number of patient results averaged (Np) and the
ratio of the standard deviation of the patient population
(sp) to the standard deviation of the analytical
method (sa). Other important factors included the
limits for evaluating the mean (control limits), the limits for
determining which patient data are averaged (truncation limits), and
the magnitude of the population lying outside the truncation limits.
Douville et al. (20) have provided a formula for
determining the number of patient results that must be averaged to
provide the error-detection capabilities of
Nc controls.
 | (2) |
 | (3) |
compares Westgard's results to those of Douville et al.
Westgard's estimates for Np are often reasonably
close to those of Douville et al.
|
While the averages of patient results have been shown to be useful for
the quality control of electrolyte [22] and endocrine test
analyzers (20), only a few laboratories, usually
large-volume reference laboratories, use patient averages for either
prospective or retrospective quality control (23). The
many reasons for this low usage rate include the lack of sophisticated
computer programs to smooth the patient data and then present
meaningful graphic summaries of the both the patient and control data;
limited experience in interpreting these summaries; and the fact that
the patient mean can indicate changes in mix of patients rather than
analytical error. Fig. 1
shows the daily means of patient potassium, sodium, calcium,
and glucose for ~45 days in the summer of 1996 at Hotel Dieu, a
tertiary care hospital in Quebec City, Canada. The day of the week is
indicated on the abscissa. Also shown is the number of patients
averaged to obtain the daily patient mean.
|
On Saturdays and Sundays ~50 to 100 fewer specimens are analyzed, and on these days the mean patient calcium, potassium, and sodium values drop but glucose values increase. Healthier patients tend to be discharged on weekends. Proportionately more sick patients are hospitalized on weekends and cause these changes. In the future, as more moderately ill patients are treated on an outpatient basis rather than being admitted to the hospital, the increased proportion of sick inpatients will cause even greater swings on weekends and holidays. The reader should be reminded that results outside of reference limits (the truncation limits) are not being averaged, and these temporal variations are occurring in the truncated data. Patient data averages are also sensitive to outlying groups of patient results, e.g., those from nephrology and dialysis clinics. As such, it would be advantageous to identify such results and exclude them from averaging.
The greatest usefulness of patient data averages for prospective quality control will be in reference and outpatient laboratories that receive large proportions of specimens with few abnormal tests. If multiple analyzers are operated simultaneously, it would be useful to randomize the specimens among the analyzers, thus decreasing the probability of a single instrument analyzing large groups of specimens from a single source such as a dialysis center. Douville et al. provided guidelines for exponential smoothing of the patient data (20), thus simplifying the calculation of patient averages for any desired run length. In an interesting use of patient averages, Miller (24) periodically calculates patient averages to determine the need for recalibration of multiple chemistry analyzers at the Medical College of Virginia.
|
Point-of-Care Quality-Control Practices |
|---|
|
TopAbstractIntroductionOptimization of Reference Sample...Quality Control Using Patient...Point-of-Care Quality-Control...External Quality Assessment and...Current Quality...References |
|---|
Because many point-of-care analyzers are more precise and accurate than whole-blood glucose meters, their manufacturers recommend extremely limited reference sample quality-control testing or, alternatively, daily electronic quality control. I recommend that the laboratory community and regulators be conservative in approving such limited reference sample quality control. Most point-of-care analyzers are relatively new, and their evaluations have been short-term and performed under ideal conditions. The error-detection capabilities of point-of-care analyzers are not well characterized. For each point-of-care analyzer, the manufacturer must compile a database of instrument malfunctions, error indicators, and the results of running the electronic control and reference specimens. Eventually, the manufacturer should be able to gather enough data to convince the regulator (and laboratorian) of the soundness of the limited reference sample approach.
|
External Quality Assessment and Quality Control |
|---|
|
TopAbstractIntroductionOptimization of Reference Sample...Quality Control Using Patient...Point-of-Care Quality-Control...External Quality Assessment and...Current Quality...References |
|---|
Cembrowski et al. (28) have proposed a multirule system to evaluate the HCFA-mandated proficiency test results: (a) a screening rule; (b) a rule for the detection of systematic error; and (c) two rules for the detection of random error. The rules are use multiples of the standard deviation index (SDI) which is calculated from: SDI = (laboratory value - group mean)/group standard deviation. The rules follow:
screening rule: 2/51sdi
If two or more observations are outside the same +1 SDI limit or
-1 SDI limit, the screening rule is violated and the data are then
tested with rules specific for systematic and random error. This rule
is usually violated in the presence of shifts exceeding 1.0 SDI and
increases in random error >200%.
mean rule: x1.5sdi
If the average of the five observations exceeds 1.5 SDI or is less
than -1.5 SDI, substantial systematic error is present. The magnitude
of the systematic error is equal to the magnitude of the average.
13sdi rule
If one or more observations is outside either the +3SDI or the
-3SDI limits, a high probability of random error exists.
r4sdi rule
If the range or the difference between the largest and the
smallest proficiency testing result exceeds 4SDI, a high probability of
random error exists.
Because of the high specificity of the follow-up rules, their violation should be followed by review of the laboratory records including the internal quality-control results. Mixups of proficiency specimens or of proficiency and clinical specimens should be excluded. Whenever possible, an aliquot of the survey specimen should be saved and reassayed for the analytes that yield erroneous results. Results that still deviate significantly after retesting indicate a long-term bias. If the deviations are variable in magnitude and direction, there may be a problem with random error. In the event that repeat analysis yields satisfactory results, the error probably represented a random error or transient bias encountered during the testing period.
The multirule system for inspecting proficiency testing data is simple and easy to teach. Its use results in a uniform style of proficiency test evaluation by supervisor, doctoral, or pathologist director. We have been using this multirule approach in several laboratories for the last 4 years (29) and have formally evaluated its application to 16 months of immunoassay testing in two laboratories (30). Significant sources of both random and systematic error have been discovered and corrected by this technique. Many proficiency test programs do not provide SDI summaries of the participant's data. Lack of SDI summaries indicates that the proficiency test provider is marketing a suboptimal product.
|
Current Quality Control Climbing the Tower of Babel
|
|---|
|
TopAbstractIntroductionOptimization of Reference Sample...Quality Control Using Patient...Point-of-Care Quality-Control...External Quality Assessment and...Current Quality...References |
|---|
Twenty years after more-efficient quality-control practices were introduced to the laboratory, their application is still uneven. It is simple to ascribe the lack of uniformity to numeric agnosia. The most important reason, however, for such heterogeneous quality-control practices is that many of the current quality-control procedures were developed locally more than a decade ago and have not changed significantly since their original implementation. Laboratorians are reluctant to change systems if they are perceived to be working satisfactorily. A recent survey of quality-control practices in over 370 US hospital laboratories (32) indicates that 50% of the respondents believe that their staff who make daily decisions about run reliability and accuracy have at least an average understanding of statistical process control rules; another 36% believe that their staff have an above-average understanding; another 4% put their staff in the "expert" category. Such high levels of understanding imply a high satisfaction with their quality-control knowledge and indirectly with their quality-control systems.
Change to more-efficient, analyte-specific quality-control procedures will thus not arise de novo in a laboratory. There must a precipitating event including the acquisition of a multichannel analyzer or laboratory information system with the facility for analyte-specific quality control. Second, there must be a change agent in the laboratory who believes that the usage of newer quality-control practices will decrease the probability of false rejections. The change agent or her/his delegate must be skilled in laboratory quality control and either manually calculate MAE and then use Westgard's quality-control selection grids (8) or else use Westgard's quality-control microcomputer program (9) or a related manual (33). The quality-control procedure must be set up in the analyzer software or LIS and tested. Finally, the quality-control procedure must be written, and all of the laboratory staff must be trained. Unfortunately, such conversions are major undertakings in today's down-sized laboratory.
|
Footnotes |
|---|
1 Nonstandard abbreviations: LIS, laboratory information system(s); MAE, medically allowable error; SDI, standard deviation index. 
|
References |
|---|
|
TopAbstractIntroductionOptimization of Reference Sample...Quality Control Using Patient...Point-of-Care Quality-Control...External Quality Assessment and...Current Quality...References |
|---|
The following articles in journals at HighWire Press have cited this article:
|
|
 |
|
  E. Bakker Is the DNA sequence the gold standard in genetic testing? Quality of molecular genetic tests assessed. Clin. Chem., April 1, 2006; 52(4): 557 - 558. [Full Text] [PDF]
|
|
|
|
|
|
R. A.R. Bowen, Y. Chan, J. Cohen, N. N. Rehak, G. L. Hortin, G. Csako, and A. T. Remaley Effect of Blood Collection Tubes on Total Triiodothyronine and Other Laboratory Assays Clin. Chem., February 1, 2005; 51(2): 424 - 433. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. A. Parvin and A. M. Gronowski Effect of analytical run length on quality-control (QC) performance and the QC planning process Clin. Chem., November 1, 1997; 43(11): 2149 - 2154. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
