Risk Management and Healthcare Policy Dovepress

open access to scientific and medical research

Open Access Full Text Article

ORIGINAL RESEARCH

Adjusting Quality Control Chart Limits for WBC,

RBC, Hb, and PLT Counts to Reduce Daily
Control Risks in Hospital Laboratory
This article was published in the following Dove Press journal:
Risk Management and Healthcare Policy

Chen-Mao Liao 1, * Background: To continuously improve medical quality and provide clinicians with more
Chih-Ming Lin 2, * accurate blood test reports, this study collected blood quality control data in 2017 from
Chin-Chia Kuo 1 a medical examination laboratory in a teaching level hospital located in Taoyuan City,
Ming-Shu Chen 3 Taiwan.
Material and Methods: The quality control data were arranged and analyzed from daily
Chun-Yang Huang 4
complete blood count (CBC), including white blood cells (WBC), red blood cells (RBC),
Ching-Yuan Lin 4
hemoglobin (Hb), and platelets (PLT) recorded by a laboratory blood analyzer. Using the
1
Department of Applied Statistics and empirical Bayesian method, we estimated the variation of concentrations of the last and
Information Science, Ming Chuan
University, Taoyuan City 33352, Taiwan; current batches to establish a novel control chart with adjusted upper and lower limits for the
R.O.C.; 2Department of Healthcare current batch, and then compared results with the traditional Shewhart method. The average
Information and Management, Ming run length (ARL) and sensitivity of the empirical Bayesian method were explored.
Chuan University, Taoyuan City 33352,
Taiwan; R.O.C.; 3Department of Results: The study found that ARL showed a qualified capability for the four blood routine
Healthcare Administration, College of tests when using the empirical Bayesian method. Compared to the Levey–Jennings control
Management and Healthcare, Oriental
chart, the novel control chart presents an alert earlier when a deviation occurs and shows
Institute of Technology, New Taipei City
22061, Taiwan; R.O.C.; 4Department of a fake alert later when there is no deviation.
Laboratory Medicine, Ten Chan General Conclusion: The parallel tests showed that the longer the time is, the better the test’s
Hospital, Chung-Li, Taoyuan City 32043,
Taiwan; R.O.C. proficiency. We concluded that the empirical Bayesian method could be applied effectively to
improve the capability of daily control in CBC laboratory tests.
*These authors contributed equally to Keywords: empirical Bayesian, EB, average run length, ARL, complete blood count, CBC,
this work
Levey–Jennings control chart

Introduction
In an era of increasing global attention to evidence-based medicine, Taiwan’s
health-care system is growing stronger, and the hospital system is changing rapidly.
Coupled with a comprehensive national health insurance payment system, the
number of people visiting hospitals is increasing. To accommodate the huge tide
of patients, automated medical equipment and artificial intelligence systems have
been introduced to ensure medical quality. Hospitals also gradually updated auto­
Correspondence: Ming-Shu Chen matic medical laboratory equipment to enhance clinical blood testing and routine
Department of Healthcare
Administration, College of Management inspection of the work. Medical laboratory equipment can process clinical speci­
and Healthcare, Oriental Institute of mens, and the resulting test results can be used by clinical specialists as a reference
Technology, No58, Sec. 2, Sichuan Road,
Pan-Chiao Dist., New Taipei City 22061, for diagnosis, screening, treatment, and prognosis. The test results become an
Taiwan; R.O.C. important basis for medical diagnosis, so the importance of the quality of inspection
Tel +886-2-77388000 ext. 6223
Email tree1013@gmail.com reports is self-evident.

submit your manuscript | www.dovepress.com Risk Management and Healthcare Policy 2020:13 3039–3049 3039
DovePress © 2020 Liao et al. This work is published and licensed by Dove Medical Press Limited. The full terms of this license are available at https://www.dovepress.com/terms.php
http://doi.org/10.2147/RMHP.S285180
and incorporate the Creative Commons Attribution – Non Commercial (unported, v3.0) License (http://creativecommons.org/licenses/by-nc/3.0/). By accessing the work
you hereby accept the Terms. Non-commercial uses of the work are permitted without any further permission from Dove Medical Press Limited, provided the work is properly attributed. For
permission for commercial use of this work, please see paragraphs 4.2 and 5 of our Terms (https://www.dovepress.com/terms.php).
Liao et al Dovepress

Monitoring and quality control of laboratory instru­ once a day to do two concentrations of the QC reagent be
ments help reduce medical misjudgments and avoids the allowed. Currently, domestic hospitals mostly follow the
additional cost of medical errors, thereby reducing the once-per-day QC rules for clinical laboratories, whereas
waste of medical resources, which is an important element large hospitals employ QC at least twice a day.
of clinical examination. A routine blood test (complete The clinical laboratory CBC QC reagent is normally
blood count, CBC) is the most common test in medical made of sheep’s blood, with a short half-life and expiration
institutions: general outpatient or medical examination period. QC reagents are replaced after about 48 days to
patients require essential routine blood tests such as ensure the batch is stable. Ezzelle et al (2008) recom­
a CBC blood test with white blood cells (WBC), red mended that a new batch number containing different
blood cells (RBC), hemoglobin (Hb) and platelets (PLT) lots of QC reagent should be implemented between the
being four of the most common. At present, the major old and new batch number of the reagent between the
hospital medical inspection units use the existing various parallel test.5 In the medical laboratory or when conduct­
quality control techniques to monitor and analyze. ing quantitative tests, parallel testing should be performed
The concept of control charts was proposed by by assaying the same samples or reference materials with
Shewhart in 1924 and published in 1931.1 Shewhart both the old and new lot numbers to assess comparability.
argued that in a stable process, there would be some For most QC materials that are handled in a medical
differences in each product after manufacturing, which laboratory, the general rule for computing the center line
would present a normal distribution of those variances, and limits is 20 daily runs with two replicates per run.
which was statistically assigned arithmetic mean (average) However, complete blood count analyses are an exception
of the data, with a three-fold standard deviation (SD) as due to their short validity period, and the new and old lots
a control line, and a control chart to monitor the produc­ of QC reagents are usually used concurrently in detection
tion process. A sample of the pipeline in the clinical for 5 to 10 working days. When the new lot of QC
laboratory is similar to a factory-produced item. The pro­ reagent’s test value reaches the standard of admission,
cess is to obtain raw materials (patient specimens), add the QC technologist is to replace the QC reagent, and
other formulations (different lots of control material), send a parallel test is conducted with the new lot of QC reagent
them to the production line (laboratory instruments), and detection average as the new batch number of the control
finally, produce the product (medical laboratory results). mean. The SD of an old batch of QC reagent detection
Levey and Jennings (1950) used the Shewhart control values is used to construct a control chart of the new batch
chart to monitor samples from the quality control of daily number for detection quality. However, there may be dif­
clinical trials. Known as the Levey–Jennings control chart, ferences in the SDs of the new and old batch numbers of
the method opened a new direction for medical quality the QC reagent, as well as the risk of lack of representa­
management in the hope of increasing the stability of the tives of centerline and limits. These issues can contribute
instrument.2 Later, Westgard et al (1981) applied multi- to the overestimation or miscalculation of probabilities of
rule quality control from industrial engineering to the field error detection and false rejection, that is, the risk of “false
of medical laboratory examination, which allowed alarms“ or ”lack of alarm” is high. We should therefore
a quality-control (QC) medical technologist to explore recognize these limitations and refrain from recommend­
the random and systemic errors of the sample.3 ing such practices to minimize the effect of QC material
Currently, many domestic and international scholars use data heterogeneity on the QC process.
the Levey–Jennings control chart and the Westgard rules In the traditional CBC daily quality control process, the
in practical research to explore random and systemic errors QC technologist is responsible for making the control
in laboratory specimen testing.4 chart easy to control, and sometimes when the deviation
Using various lots of QC materials or reagents in is large, the QC technologist will contact and get the
hospital laboratory QC, mostly in accordance with the equipment supplier to modify and define the upper and
Westgard rules, the instrument is QC-measured twice lower control limits (UCL and LCL). However, this is not
daily and executed with each QC concentration reagent the standard practice and the QC process should be con­
(level 1 and level 2). Due to the high cost of QC solutions, trolled by the lab staff and not the equipment-resident.
the Clinical Laboratory Improvement Amendments Otherwise, the largest SD of the past five batch numbers
(CLIA) Final Rule in 2003 proposed that a minimum of will be used as the new SD to calculate the UCL and LCL

3040 submit your manuscript | www.dovepress.com Risk Management and Healthcare Policy 2020:13
DovePress
Dovepress Liao et al

of the new batch number detection value. This prompted Table 1 shows that regardless of any concentration of QC
the hospital to set the control limit, usually too loose and, data in the count detection test of PLT, the change in its
therefore, unable to accurately produce a clear sense of the detection value is significant. This study will use seven lot
information in the clinical report. numbers of platelet counts in a low-concentration (Level-1)
Therefore, this study uses the empirical Bayesian detection value series. Figure 1 indicates that the degree of
method6–9 to integrate the past batch number quality con­ variation is different, while Table 2 shows results of a Levene’s
trol data SD and the new batch number quality control data test to determine the number of the seven lot PLT QC count
standard difference information to carry out the adjustment variations in the same degree of variation analysis. Results
of the boundary of the new batch number QC reagent show that the variation of the PLT value of the seven numbers
control chart. This method should draw out the most varies. In the medical laboratory where our study was con­
suitable data for the new batch number QC reagent solu­ ducted, we used the Westgard multi-rules for CBC daily QC.
tion control chart. There are three limits, namely 1:3s, 2:2s, and R4s, which serve
This paper first introduces the example of the routine as rejection rules (an action plan was required when a rejection
hospital blood CBC test, explaining the challenges occurred). Also, the 1:2s was a warning rule. Therefore, this
encountered in the clinical examination of the teaching work maintains that for the control chart of the detection value
hospital. The application of the empirical Bayes method of the QC reagent, the control limit should be adjusted using
is offered, and the adjustment steps of the empirical the different lot numbers. UCL and LCL can be adjusted using
Bayesian method control chart are introduced. The use of the empirical Bayesian method, which can increase the effi­
hospital CBC test examples explains how UCL and LCL ciency of the control chart.
are adjusted, then computer simulations and CBC daily
QC test data are compared with traditional control charts. Bayes’ Theorem of Normally Distributed
Finally, the conclusion and discussion are offered. Variances
The Wilcoxon signed-rank test was used to test differences in
the paired data in this study, with the S. Consider the detec­
Materials and Methods
tion values of a new QC reagent as
This study collected CBC analyzer (Sysmex XN-2000; �
2
Sysmex Corporation©, Hyogo, Japan) test data used in X1 ; X2 ; � � � ; Xn ,N μ0 ; σ , in which μ0 is the mean detection
a teaching hospital in northern Taiwan. QC reagent of the reagent given by the manufacturer and θ ¼ σ2 . Given
(Sysmex XN CHECK; Sysmex Corporation©, Hyogo, that the variance Θ ¼ θ, the conditional probability distribu­
Japan) data included a low concentration (Level-1), med­ tion of X1 ; X2 ; � � � ; Xn is:
� �
ium concentration (Level-2), and a high concentration 1 n n=2 ∑ni¼1 ðxi μ0 Þ2
f ðx1 ; x2 ; � � � ; xn jμ0 ; θÞ ¼ pffiffiffiffiffi θ e 2θ ; θ>0
(Level-3). QC reagent data were collected for one year 2π
from January to December 2017, with a daily QC in
Bayesian analysis can be performed using conjugate
a morning and evening test, during which the WBC
prior distributions.6 Assuming that the prior distribution of
count, RBC count, Hb, and PLT count were collected.
the variance Θ is an inverse gamma distribution, that is,
There are four test items in the period, with a total of
Θ,IGðα; βÞ, then its probability density function is:
seven lot numbers of QC reagent data. Table 1 shows
1
descriptive statistics of CBC test data. For simplicity, this e βθ
πðθjα; βÞ ¼ ; θ>0; α>0; β>0
study selected four common CBC test items in which the ΓðαÞβα θαþ1
QC test values included level-1, level-2, and level-3 vari­
The mean and variance of the inverse gamma distribu­
able coefficients for an elevated PLT count for analysis and
tion are, respectively:
explanation. The empirical Bayesian method could be
1
applied effectively to improve daily quality control for μΘ ¼ EðΘÞ ¼
βðα 1Þ
CBC laboratory tests. Through the empirical Bayesian (1)
1
method, the data of the CBC test items were integrated σ2Θ ¼ VarðΘÞ ¼
β2 ðα 1Þ2 ðα 2Þ
and tested by the new and old batch number of the QC
reagent, then imported into the chart for analysis and Then, the joint probability density function of
explanation. X1 ; X2 ; � � � ; Xn ; Θ is

submit your manuscript | www.dovepress.com

Risk Management and Healthcare Policy 2020:13 3041
DovePress
Liao et al Dovepress

Table 1 Descriptive Statistics of CBC Test Items

BN WBC RBC Hb PLT

Level 1 Level 2 Level 3 Level 1 Level 2 Level 3 Level 1 Level 2 Level 3 Level 1 Level 2 Level 3

1 n 23 23 23 23 23 23 23 23 23 23 23 23
Mean 3.06 7.16 16.59 2.32 4.44 5.40 6.09 12.55 17.03 83.78 244.74 556.52
var 0.005 0.010 0.015 0.001 0.001 0.002 0.004 0.006 0.008 13.178 23.656 104.625
cv 0.167 0.136 0.088 0.019 0.027 0.029 0.064 0.050 0.046 15.729 9.666 18.800

2 n 69 69 69 69 69 69 69 69 69 69 69 69
Mean 2.95 6.98 16.35 2.41 4.46 5.42 6.29 12.77 17.33 79.45 245.64 554.23
var 0.003 0.011 0.036 0.001 0.002 0.002 0.003 0.006 0.008 52.192 21.205 65.063
cv 0.117 0.151 0.218 0.024 0.038 0.045 0.046 0.045 0.046 65.693 8.633 11.739

3 n 71 72 72 71 72 72 71 72 72 71 72 72
Mean 2.82 6.79 16.03 2.38 4.50 5.58 5.95 12.07 17.11 85.85 249.57 563.13
var 0.004 0.011 0.037 0.001 0.002 0.003 0.004 0.006 0.011 16.076 27.319 56.759
cv 0.134 0.156 0.232 0.029 0.041 0.051 0.066 0.051 0.064 18.726 10.946 10.079

4 n 61 61 61 61 61 61 61 61 61 61 61 61
Mean 2.93 6.82 16.49 2.43 4.60 5.58 6.17 12.37 17.09 83.30 256.31 584.20
var 0.003 0.009 0.025 0.001 0.001 0.002 0.005 0.005 0.007 22.345 30.218 70.194
cv 0.112 0.128 0.149 0.026 0.032 0.028 0.087 0.042 0.040 26.826 11.790 12.015

5 n 70 68 72 70 68 72 70 68 72 70 68 72
Mean 3.07 6.79 16.71 2.37 4.50 5.45 5.79 12.31 16.63 84.07 243.88 554.26
var 0.004 0.009 0.027 0.001 0.001 0.002 0.005 0.010 0.012 19.140 20.613 81.211
cv 0.145 0.134 0.163 0.025 0.030 0.028 0.086 0.080 0.073 22.766 8.452 14.652

6 n 70 67 69 70 67 69 70 67 69 70 67 69
Mean 2.99 6.77 16.80 2.37 4.42 5.37 5.76 11.89 16.34 86.53 247.40 562.59
var 0.004 0.014 0.031 0.001 0.002 0.001 0.007 0.016 0.016 16.166 32.547 42.774
cv 0.139 0.214 0.187 0.028 0.039 0.025 0.114 0.132 0.098 18.683 13.156 7.603

7 n 21 22 21 21 22 21 21 22 21 21 22 21
Mean 3.02 6.68 16.28 2.37 4.43 5.37 5.83 12.29 16.10 75.76 243.00 562.33
var 0.007 0.010 0.033 0.001 0.001 0.001 0.003 0.009 0.004 36.790 20.762 46.733
cv 0.229 0.143 0.202 0.010 0.022 0.022 0.054 0.077 0.028 48.561 8.544 8.311

Abbreviations: BN, batch (Lot) number; n, sample size; mean, average; var, variation; cv, variation coefficient.

on a Bayes estimator, the following equation can be

� � 1
1 n n=2
∑n ðxi μ0 Þ2
i¼1 e βθ (2) derived from Equation 1:
¼ pffiffiffiffiffi θ e 2θ �
2π ΓðαÞβα θαþ1 1
∑ni¼1 ðxi μ0 Þ2 þ β1
^θB ¼ EðΘjx1 ; x2 ; � � � ; xn Þ ¼ 2
Therefore, the marginal joint probability of X1 ; X2 ; � � � ; Xn α þ n2 1
n
α 1 1
can be deduced as: ¼ � þ 2
� Sn2 (4)
α 1 þ n2 βðα 1Þ α 1 þ n2
∑ni¼1 ðxi μ0 Þ2
where Sn2 ¼ is the variance of the nth detec­
� � �� � ðαþn=2Þ n
1 nΓ αþ2
n
1=2 ∑ni¼1 ðxi μ0 Þ2 þ β1 tion of a new QC reagent; and ^θB is the mean of prior
¼ pffiffiffiffiffi
2π ΓðαÞβα distributions and the weighted mean of the variance of the
(3) detection values of a QC reagent with a new batch number.

According to Equation 1 and Equation 2), it can be

deduced that when ðX1 ; X2 ; � � � ; Xn Þ ¼ ðx1 ; x2 ; � � � ; xn Þ, the Empirical Bayesian Method
posterior probability of Θ is The empirical Bayes method can be used to estimate
� � ! parameters through the marginal distribution of
n 1
IG α þ n2 ; 1=2 ∑ ðxi μ0 Þ2 þ 1β . Therefore, based a previous batch of QC data. Consider that a set of QC
i¼1

3042 submit your manuscript | www.dovepress.com Risk Management and Healthcare Policy 2020:13
DovePress
Dovepress Liao et al

Figure 1 Comparison chart of Level-1 PLT count detection value.

datax11 ; � � � ; x1n1 ; x21 ; � � � ; x2n2 ; � � � ; xm1 ; � � � ; xmnm in which Therefore, the empirical Bayes method can be used to
the quantity of old batch numbers is m,7–9 its likelihood adjust the variance of the detections of a new batch of QC
function will be where Sj2 is the sample variance of the jth reagents. The control chart of the X� , of which k samples
batch of QC data. are drawn each time, is where �xnew is the sample mean
during parallel testing.
8 qffiffiffiffiffi
� �
� � ∑m nj nj � >
> UCL ¼ �
x þ 3
^θEB
1 j¼1 � m
j¼1 Γ α þ 2
< X� new k
¼ pffiffiffiffiffi � CLX� ¼ �xnew q ffi ffi ffiffiffi (7)
2π Γ m
ð αÞβ mα (5) >
>
� � : LCL � ¼ �x 3
^θEB

m nj 2 1 ðαþnj =2Þ X new k

� �j¼1 S þ
2 j β

Equation 5 can be used to determine the maximum Results

likelihood estimator (MLE). Substituting α ^ into ^θB
^ and β In this study, the samples were compared with a computer
yields simulation. The sample of blood tests mentioned in sub­
n 2 1 section one is offered as an example of sample comparison
2 Sn þ ^β
n
^θEB ¼ α
^ 1 1
¼ n�^ þ 2
� Sn2 using computer simulations. This example illustrates the
^ þ n2 1
α α
^ ^ 1Þ α
1 þ 2 β ðα ^ 1 þ n2
steps of adjustment of the limits of the control chart, while
(6)
section 2 assesses the effectiveness of the control chart

Table 2 Levene’s Variant Homogenous Test Analysis of Variance Table

Source of Variation Sum of Squares Degree of Freedom Mean Square F P-value

Different lot 471.4 6 78.7 10.1 10

2:4 � 10
Error 2944.3 378 7.8

submit your manuscript | www.dovepress.com

Risk Management and Healthcare Policy 2020:13 3043
DovePress
Liao et al Dovepress

before and after the adjustment of the limit by an average Regarding the detection values of the other two QC
series of lengths. reagents (medium-concentration and high-concentration),
their control charts before and after adjustment are shown
in Figures 3 and 4, respectively. Based on these figures and
Adjustment of Hospital CBC Routine
Table 1, there is a greater change in variance of the low-
Test Examples and high-concentration detection values in reagents with
The research samples in Table 1 were tested during the
new and old batch numbers. As a result, there is a greater
morning and evening as part of the hospital routine. The
difference in the adjusted control limit. However, the
frequency of parallel testing was n = 10. In this section,
adjusted limits of the medium-concentration detection
the steps of adjusting the limits of the control chart are values are less significant as the difference between the
explained using low-concentration platelet detection variance values was marginal.
values as an example.
Step 1: According to Table 1, there are six (m ¼ 6)
batch of detection data pertaining to a QC reagent. The Comparison with Simulations
batch sizes is n1 ¼ 23; n2 ¼ 69; n3 ¼ 71; n4 ¼ 61; n5 ¼ In this section, it is assumed that significant differences
n6 ¼ 70 and the detected variance of each old batch is, exist in the variances of QC reagents with different batch
respectively, S12 ¼ 13:178; S22 ¼ 52:192; S32 ¼ 16.076 numbers. Computer simulations were performed to com­
pare the average run lengths of the control charts before
S42 ¼ 22:345; S52 ¼ 19:14; S62 ¼ 16:166. Substituting
and after adjustment. By using the detection and estimated
these values into the likelihood function (Equation 5))
values of the low-concentration platelet sample, the simu­
α ¼ 7:0892; β
will yield^ ^ ¼ 0:0071.
lation steps are as follows:
Step 2: The mean and variance of a low-
Step 1: Generate seven random variables
concentration platelet sample with a new batch number
from the inverse gamma distribution,
as calculated through parallel testing are, respectively,
namely θ1 ; θ2 ; � � � ; θ7 ,IGðα ¼ 7:0892; β ¼ 0:0071Þ
�xnew ¼ 74:4; S 2 ¼ 39:7; n ¼ 10. Substituting α ^ into
^; β
n Step 2: Generate the observed values of the six old
Equation 6) generates the empirical Bayes estimator of 0 0 0
batch numbers xi1 ; xi2 ; � � � ; xini ,N ð�xi ; θi Þ; i ¼ 1; 2; � � � ; 6,
the variance of a detection value with a new batch as well as the data of the ten (n ¼ 10) new batch numbers
number, that is, ^θEB ¼ 30:991. 0 0 0
used in parallel testing, ie x71 ; x72 ; � � � ; x7n ,N ð�x7 ; θ7 Þ.
Step 3: In that the morning and evening test data must Step 3: Substitute the simulation data into the three
be monitored, k ¼ 1, and the control chart can be adjusted steps in Section 4.1 and calculate the control limits before
according to Equation (7) to and after adjustment.
8 pffiffiffiffiffiffiffi Step 4: Simulate the shift (δ) in detection value and the
< UCLX ¼ �xnew þ 3 ^θEB ¼ 91:4 pffiffiffiffi
CLX ¼ �xnew SD ( θ7 ) of the data of the new batch numbers
: pffiffiffiffiffi¼
ffiffi 74:4 pffiffiffiffiffiffiffiffiffiffiffiffiffi �
^
LCLX ¼ �xnew 3 θEB ¼ 58:0 xδ7j ,N �x7 þ δ 30:991; θ7 ; j ¼ 1; 2; � � � , until the control

The control chart of the detection values of a sample limit appears. Record the run lengths of the control limits
with a new batch number can be adjusted through these before and after adjustment.
three steps. In addition, since the maximum variance of the Step 5: Repeat steps 1 to 4 for a total of 2000 run
previous five batches is S22 ¼ 52:192>^θEB , then the limits lengths and calculate the average run length (ARL).
of a control chart in a traditional hospital are wider, and its In the simulations of this study, the shift was set as δ ¼
0; 0:5; 1; and 2 times the SD. The simulation results of
control chart is:
8 pffiffiffiffiffi the detected platelet count at each concentration are pre­
< UCLX ¼ �xnew þ 3 S22 ¼ 96:1 sented in Table 3. The results showed that the ARL0 at
CLX ¼ �xnew pffiffiffiffiffi¼ 74:4 a controlled state and the ARL1 at uncontrolled states of
:
LCLX ¼ �xnew 3 S22 ¼ 52:7
the control charts before adjustment were all substantially
Figure 2 is the control chart of the platelet count large, which indicated that the control charts had lost their
detected in the low-concentration QC reagent with a new monitoring capacity. On the contrary, the ARL0 at
batch number. It is evident that there is a difference in the a controlled state of the control charts after adjustment
UCL and LCL before and after adjustment. were more adequate and are likely to be detected under

3044 submit your manuscript | www.dovepress.com Risk Management and Healthcare Policy 2020:13
DovePress
Dovepress Liao et al

Figure 2 Control chart of the platelet count detected in the low-concentration reagent with a new batch number before and after adjustment.

uncontrollable conditions. Based on the simulation results, blood tests in clinical patients, there must be a quality con­
when differences exist in the variances of QC reagents trol mechanism to eliminate deviations and abnormalities.14
with different batch numbers, Bayes method should be A full range of CBC blood routine test values (includ­
used to adjust the limits of a control chart, thus increasing ing WBC, RBC, Hb, Hct, PLT, MCV, MCH, MCHC, etc.),
its monitoring effectiveness. in the clinical setting can assist specialists to diagnose
whether there is anemia, inflammatory infection, blood
Discussion disease, and clotting disorders, blood cancer or leukemia.
Westgard (1992) and Petersen et al (1996) agree that inter­ For special patients, the critical test report values, such as
nal quality control (IQC) should be designed, the main deviation, will directly affect the doctor’s medication and
purpose of which is to ensure that individual hospitals follow-up treatment methods.
choose different laboratory instruments and methods. IQC To improve the stability and accuracy of the clinical
is an important basis for verifying the various methods.10,11 laboratory QC process, the hospital clinical laboratory
According to the latest version of the medical laboratory conducts parallel testing before replacing the new lot
international certification guidelines (ISO-15,189: 2012). number of the CBC QC reagent. When parallel testing in
The ISO 15,189 requires that ‘the laboratory shall design the range of the admission, the clinical laboratory will use
IQC systems that verify the attainment of the intended the SD of the old lot number to be the new SD of the new
quality of results’.12 Although some medical laboratory lot number. The SD of the old batch number is likely to be
scholars believe that the acceptability of the biological extended to the wider QC reagent standard deviation,
variability assessment method should be based on the rele­ resulting in the quality control being a more relaxed pro­
vant parameters, there is a consensus on the actual recom­ blem. Thus, when the laboratory instrument stability is
mended parameters.13 Therefore, for the results of CBC systematically shifted, system errors will occur. Control

submit your manuscript | www.dovepress.com

Risk Management and Healthcare Policy 2020:13 3045
DovePress
Liao et al Dovepress

Figure 3 Control chart of the platelet count detected in the medium-concentration reagent with a new batch number before and after adjustment.

charts made from existing control rules may not immedi­ new and old batches by weighing standard deviation when
ately detect abnormal alarms. the data of the new batch are smaller in volume and the
Through this study, we found that there are differences values are more dependent on the standard deviation of the
between the various lots of CBC QC reagents. Not correct­ old batch. However, when the new batch data increase in
ing the SD leads to the control chart monitoring declining volume, the values will be more dependent on the standard
or losing capacity. The control chart of the empirical deviation of the new batch. In other words, the proportion
Bayesian method can significantly reduce this problem. of new batch QC data used in the calculation process also
Previous lot QC data is combined with the new lot number increases, which can improve quality control performance.
data to adjust the parameters suitable for the new lot In practice, the weighted average of CV% in the past six
number QC reagent SD control chart. The chart will not months is generally used as the basis for the subsequent
rely on the old lot number of the QC reagent’s SD: the new setting of internal quality control limits in order to better
lot number test of the new quality control data will be reflect the long-term stability of the instrument.
more in line with the new lot of QC reagent’s SD. For Clinical laboratory routine blood tests are a large part
quantitative tests, parallel testing should be performed by of modern medicine in hospital including biochemical test
assaying the same samples or reference materials with and complete blood counting. Nearly all inpatients, about
both the old and new batch numbers to assess compar­ half of patients in the emergency department, and nearly
ability. Since the original CV% in parallel testing (if the one-third of outpatients had laboratory results during their
QC rules are accepted) will follow the old standard devia­ visit to the health-care center,15 comprising a quite portion
tion, the standard deviation of the new batch control chart (5–6%) of the total cost of healthcare.16,17 If we can
may lead to the problem of lax control. The rule applied improve the quality of CBC’s lab test reports, coupled
for the empirical Bayesian method is to use the data of the with future cloud consolidation to avoid duplicate testing

3046 submit your manuscript | www.dovepress.com Risk Management and Healthcare Policy 2020:13
DovePress
Dovepress Liao et al

Figure 4 Control chart of the platelet count detected in the high-concentration reagent with a new batch number before and after adjustment.

in the different hospitals even the same hospital, we exploration of the better daily quality control standards
believe that many health insurance medical expenses can for CBC routine blood testing.
be reduced, while improving the quality of care. At pre­ In addition to the importance that the United States and
sent, there is no golden standard protocol for clinical test some Asian countries attach to laboratory quality manage­
personnel to perform parallel testing of QC reagents with ment of the Clinical Laboratory Improvement Amendments
new and old batch numbers as part of their CBC routine. (CLIA) and the World Health Organization (WHO)
In general, the number of parallel tests performed is requirements,18,19 under the health-care insurance system,
approximately ten. There should be room for further several advanced countries in Asia have been constantly

Table 3 ARLs of Detected Platelet Counts at Different Concentrations and at Controlled and Uncontrolled States
Item Control Chart ARL0 ARL1 (UCS)
(CS)
Data Shifts 0.5 Data Shifts 1 Data Shifts 2
Times the SD Times the SD Times the SD

Low-concentration After EB adjustment 319.84 217.47 86.94 20.33

Level-1 Before adjustment 62,238.43 55,045.03 37,690.59 17,450.01

Medium-concentration After EB adjustment 307.01 185.12 76.68 13.41

Level-2 Before adjustment 1477.46 972.76 284.46 43.65

High-concentration After EB adjustment 306.26 196.47 78.95 13.93

Level-3 Before adjustment 3479.93 2699.23 550.83 60.84
Abbreviations: SD, standard deviation; CS, controlled state; UCS, uncontrolled state.

submit your manuscript | www.dovepress.com

Risk Management and Healthcare Policy 2020:13 3047
DovePress
Liao et al Dovepress

committed to the laboratory quality improvement,20,21 for supporting this research with ID TCGH 1060306001;
reduced medical resource waste and improved cost- MOST 105-2662-E-161-002-CC3; MOST 108-2221-E-
effectiveness of clinical laboratories.22–25 We believe that 161-003-MY2 and this research received no specific
with every laboratory continuous efforts, it will be able to grant from any funding agency in the public, commercial,
provide the more effective and accurate reporting for med­ or not-for-profit sectors. The authors are grateful for all
ical laboratories, become the better support for clinicians to medical technologists of the laboratory of Ten-Chen
diagnose diseases, and continuously improve the overall Medical Group, Ten Chan General Hospital (Chung Li)
quality of health-care system. for their support and help in this study. Chen-Mao Liao
and Chih-Ming Lin are co-first authors for this study.
Limitations
The major limitation of this study is that due to this being
Disclosure
a hospital-based analysis, variations across different insti­
The authors report no conflicts of interest for this work.
tutions must be considered. Is a pilot study, the lack of
laboratories in the hospital, and lack of comparison
between different labs is also the research limitation. References
1. Shewhart WA. Economic Control of Quality of Manufactured
Conclusions Product. London: Macmillan And Co Ltd; 1931.
2. Levey S, Jennings ER. The use of control charts in the clinical
Continuous quality improvement in health-care systems is laboratory. Am J Clin Pathol. 1950;20(11_ts):1059–1066.
very important and helpful in hospitals.26 This study offers doi:10.1093/ajcp/20.11_ts.1059
3. Westgard JO, Barry PL, Hunt MR, Groth T. A multi-rule Shewhart
an innovative application of the empirical Bayesian chart for quality control in clinical chemistry. Clin Chem. 1981;27
method to adjust the SD of CBC QC reagent daily control (3):493–501. doi:10.1093/clinchem/27.3.493
data. The research aims to effectively import the hospital 4. Plebani M, Piva E. Erythrocyte sedimentation rate: use of fresh blood
for quality control. Am J Clin Pathol. 2002;117(4):621–626.
clinical laboratory quality control process to improve the doi:10.1309/QB1G-6FRR-DNWX-BKQ9
quality control ability of the routine blood examination 5. Ezzelle J, Rodriguez-Chavez IR, Darden JM, et al. Guidelines on
good clinical laboratory practice: bridging operations between
CBC test reports. Further research may build upon the
research and clinical research laboratories. J Pharm Biomed Anal.
standard parallel comparison quality control procedures 2008;46(1):18–29. doi:10.1016/j.jpba.2007.10.010
in the different lots of QC reagents. Then, adjustments to 6. Berger JO. Statistical Decision Theory and Bayesian Analysis.
Springer Science & Business Media; 2013.
the control chart limits may be made that are more suitable 7. Carlin BP, Louis TA. Bayes and Empirical Bayes Methods for Data
for the new lot number of the QC reagents. To clarify the Analysis. Vol. 88. Boca Raton: Chapman & Hall/CRC; 2000.
variation of inter-/intra-laboratories, more evidence has to 8. Berger J, Berliner LM. Robust Bayes and empirical Bayes analysis
with $ _\epsilon $-contaminated priors. Ann Stat. 1986;14
be obtained from different institutions. (2):461–486. doi:10.1214/aos/1176349933
9. Casella G. An introduction to empirical Bayes data analysis. Am Stat.
1985;39(2):83–87.
Patient Consent for Publication 10. Hyltoft Petersen P, Ricos C, Stockl D, et al. Proposed guidelines for
Not required. the internal quality control of analytical results in the medical
laboratory. Eur J Clin Chem Clin Biochem. 1996;34(12):983–1000.
11. Westgard JO. Assuring analytical quality through process planning
Ethics Approval and Consent to and quality control. Arch Pathol Lab Med. 1992;116(7):765–769.
12. ISO. ISO, 15189: 2012, Medical Laboratories–Particular
Participate Requirements for Quality and Competence. Geneva: ISO; 2012.
Not required. There are no patient laboratory data in this 13. Cooper G, DeJonge N, Ehrmeyer S, et al. Collective opinion paper on
findings of the 2010 convocation of experts on laboratory quality. Clin
study. Chem Lab Med. 2011;49(5):793–802. doi:10.1515/CCLM.2011.149
14. Miller W, Erek A, Cunningham TD, Oladipo O, Scott MG,
Johnson RE. Commutability limitations influence quality control
Consent for Publication results with different reagent lots. Clin Chem. 2011;57(1):76–83.
This manuscript contains no person’s details, images, or doi:10.1373/clinchem.2010.148106
videos. 15. Ngo A, Gandhi P, Miller WG. Frequency that laboratory tests influ­
ence medical decisions. J Appl Lab Med. 2017;1:410–414.
doi:10.1373/jalm.2016.021634
Acknowledgments 16. Hu WY, Yeh CF, Shiao AS, et al. Effects of diagnosis-related group
payment on health-care provider behaviors: a consecutive
This research was funded by Ten Chan General Hospital three-period study. J Chin Med Assoc. 2015;78:678–685.
(TCGH) and Ministry of Science and Technology (MOST) doi:10.1016/j.jcma.2015.06.012

3048 submit your manuscript | www.dovepress.com Risk Management and Healthcare Policy 2020:13
DovePress
Dovepress Liao et al

17. Choi JW, Kim SJ, Park HK, et al. Effects of a mandatory DRG 22. Shimetani N. Medical reimbursement for diagnostic tests. Expert Rev
payment system in South Korea: analysis of multi-year nationwide Pharm Outcomes Res. 2004;4:127–130.
hospital claims data. BMC Health Serv Res. 2019;19:776. 23. Su BG, Chen SF, Yeh SH, et al. Cost evaluation of clinical laboratory
18. World Health Organization. WHO Manual for Organizing a Nati in Taiwan’s National Health System by using activity-based costing.
Onal External Quality Assessment Programme for Health Clin Chem Lab Med. 2016;54(11):1753–1758. doi:10.1515/cclm-
Laboratories and Other Testing Sites. Geneva, Switzerland: WHO 2016-0193
Press; 2016. 24. Jun JK, Sung NY, Song SH, et al. Budget impact of the accreditation
19. Code of federal regulations chapter IV SubchapterGPart 493 program for clinical laboratories on colorectal cancer screening via
SubpartASection 493.17-test categorization. Available from: https:// fecal immunochemical testing: results from the national cancer
w w w. e c f r. g o v / c g i - b i n / t e x t - i d x ? S I D = 1 2 4 8 e 3 1 8 9 d a 5 e 5 screening program in Korea. Ann Lab Med. 2018;38:249–254.
f936e55315402bc38b&node=pt42.5.493&rgn=div5. Accessed August doi:10.3343/alm.2018.38.3.249
17, 2020. 25. Inal TC, Goruroglu Ozturk O, Kibar F, et al. A. Lean six sigma
20. Kim S, Yun YM, Kim H, et al. The new diagnosis-related group methodologies improve clinical laboratory efficiency and reduce
reimbursement system and laboratory test quality in Korea: analysis turnaround times. J Clin Lab Anal. 2018;32(1):e22180. doi:10.1002/
of external quality assessment results. Healthcare. 2020;8 jcla.22180
(2):127–137. doi:10.3390/healthcare8020127 26. Zhang D, Liao M, Liu T. Implementation and promotion of quality
21. Kim H, Kim S, Yun YM, et al. Status of quality control for laboratory control circle: a starter for quality improvement in Chinese hospitals.
tests of medical institutions in Korea: analysis of 10 years of data on
Risk Manag Healthc Policy. 2020;Volume 13(13):1215–1224.
external quality assessment participation. Healthcare. 2020;8
doi:10.2147/RMHP.S261998
(2):75–82. doi:10.3390/healthcare8020075

Risk Management and Healthcare Policy Dovepress

Publish your work in this journal
Risk Management and Healthcare Policy is an international, peer- guidelines, expert opinion and commentary, case reports and
reviewed, open access journal focusing on all aspects of public extended reports. The manuscript management system is completely
health, policy, and preventative measures to promote good health online and includes a very quick and fair peer-review system, which
and improve morbidity and mortality in the population. The journal is all easy to use. Visit http://www.dovepress.com/testimonials.php
welcomes submitted papers covering original research, basic to read real quotes from published authors.
science, clinical & epidemiological studies, reviews and evaluations,
Submit your manuscript here: https://www.dovepress.com/risk-management-and-healthcare-policy-journal

submit your manuscript | www.dovepress.com

Risk Management and Healthcare Policy 2020:13 3049
DovePress