Koch - Control Charts
Koch - Control Charts
Michael Koch
Assuring the Quality of Test and
Calibration Results - ISO/IEC 17025 – 5.9
The laboratory shall have quality control
procedures for monitoring the validity of
tests and calibrations undertaken.
The resulting data shall be recorded in
such a way that trends are detectable
and, where practicable, statistical
techniques shall be applied to the
reviewing of the results.
Assuring the Quality of Test and
Calibration Results - ISO/IEC 17025 – 5.9
This monitoring shall be planned and reviewed and
may include, but not be limited to, the following:
regular use of certified reference materials and/or
internal quality control using secondary reference
materials;
participation in interlaboratory comparison or
proficiency-testing programmes;
replicate tests or calibrations using the same or
different methods;
retesting or recalibration of retained items;
correlation of results for different characteristics of an
item.
Control Charts
powerful, easy-to-use technique for the
control of routine analyses
ISO/IEC 17025 demands use wherever
practicable
History
introduced by Shewhart in 1931
originally for industrial manufacturing
processes
for suddenly occurring changes and for slow
but constant worsening of the quality
Immediate interventions reduce the risk of
production of rejects and complaints from the
clients
Principle
Take samples during the process
Measure a quality indicator
Mark the measurement in a chart with warning and
action limits
concentration
target value
sample-# 1 2 3 4 5 6 7 8 9 10 1112 13 14 15 16 17 18
Control Charts in Analytical
Chemistry
Target value
certified value of a RM
mean of often repeated measurements
Control Charts in Analytical
Chemistry
Warning / action limits
if data are normal distributed
95.5% of the data are in µ±2σ
99.7% are in µ±3σ
xtarget±2s is taken as warning limits
xtarget±3s is taken as action limit
Action Limits
There is probability of only 0.3 % that a
(correct) measurement is outside the
action limits (3 out of 1000
measurements)
Therefore the process should be
stopped immediately and searched for
errors
Warning Limits
4.5% of the (correct) values are outside
the warning limits.
This is not very unlikely
Therefore this is only for warning, no
immediate action required
Calculation of Standard
Deviation
measurements marked in the control
chart are between-batch
standard deviation should also be
between-batch
estimation from a pre-period of about 20
working days
repeatibility STD too narrow limits
interlaboratory STD too wide limits
Limits Fitness for Purpose
Action and warning limits have to be
compatible with the fitness-for-purpose
demands
no blind use
Out-of-control Situation 1
suddenly deviating value, outside the action limits
8
7
concentration
6
5
4
3
2
1
0
01
01
01
01
01
01
01
01
01
6.
6.
7.
7.
7.
6.
6.
6.
7.
.0
.0
.0
.0
.0
.0
.0
.0
.0
17
20
23
26
29
02
05
08
11
date
Out-of-control Situation 2
2 of 3 successive values outside the
warning limits
8
7
concentration
6
5
4
3
2
1
0
01
01
01
01
01
01
01
01
01
6.
6.
7.
7.
7.
6.
6.
6.
7.
.0
.0
.0
.0
.0
.0
.0
.0
.0
17
20
23
26
29
02
05
08
11
date
Out-of-control Situation 3
7 successive values on one side of the
central line
8
7
concentration
6
5
4
3
2
1
0
01
01
01
01
01
01
01
01
01
6.
6.
7.
7.
7.
6.
6.
6.
7.
.0
.0
.0
.0
.0
.0
.0
.0
.0
17
20
23
26
29
02
05
08
11
date
concentration
17
.0
6.
0
1
2
3
4
5
6
7
8
01
20
.0
6.
values
01
23
.0
6.
01
26
.0
6.
01
29
.0
6.
01
02
.0
date
7.
01
05
.0
7.
01
08
.0
7.
01
Out-of-control Situation 4
11
.0
7.
01
7 successive increasing or decreasing
Advantages of Graphical Display
much faster
more illustrative
clearer
Different Control Charts
X-chart
original Shewhart-chart
with single values from analysis
mainly to validate precision
trueness with reference materials
also possible for calibration parameters
(slope, intercept)
EXCEL-Example
for control charts
Different Control Charts
Blank Value Chart
analysis of a sample, which can be assumed
to not contain the analyte
special form of the Shewhart chart
information about
the reagents
the state of the analytical system
contamination from environment
enter direct measurements, not calculated
values
Different Control Charts
Recovery Rate Chart - I
reflects influence of the sample matrix
Principle:
analyse actual sample
spike this sample with a known amount of
analyte
analyse again
Recovery rate:
xspiked − x unspiked
RR = ⋅ 100%
∆xexpected
Different Control Charts
Recovery Rate Chart - II
detects only proportional systematic
errors
constant systematic errors remain
undetected
spiked analyte might be bound
differently to the sample matrix better
recovery rate for the spike
Target value: 100%
Different Control Charts
Range Chart
absolute difference between the highest
and lowest value of multiple analyses
precision check
control chart has only upper limits
concentration
target value
sample-# 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Different Control Charts
Difference Chart - I
uses difference with its sign
analyse actual sample at the beginning of a series
analyse same sample at the end of the series
calculate difference (2nd value
– 1st value)
mark in control chart with the sign
Different Control Charts
Difference Chart - II
target value: 0
otherwise: drift in the analyses during the
series
appropriate for precision and drift check
Different Control Charts
Cusum Chart - I
highly sophisticated control chart
cusum = cumulative sum = sum of all
errors from one target value
target value is subtracted from every
control analyses and difference added
to the sum of all previous differences
Different Control Charts - Cusum Chart - II
T = 80 s = 2.5 90
7 80 0 0 0 2 4 6 8 10 12 14 16
8 79 -1 -1 30
9 78 -2 -3
10 80 0 -3 20
11 76 -4 -7 10
12 77 -3 -10 0
13 76 -4 -14 0 2 4 6 8 10 12 14 16
14 76 -4 -18 -10
15 75 -5 -23 -20
-30
Different Control Charts - Cusum Chart - III
V-mask as indicator for out-of-control situation
30 30
20
in 20
out of
10
control 10
control
0 0
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
-10 -10
-20 -20
-30 -30
quickly detected
Different Control Charts
Cusum Chart - IV
Advantages
it indicates at what point the process went
out of control
the average run length is shorter
number of points that have to be plotted before
a change in the process mean is detected
the size of a change in the process mean
can be estimated from the average slope
Control Samples
no control chart without control samples
requirements:
must be suitable for monitoring over a longer time period
should be representative for matrix and analyte conc.
concentration should be in the region of analytically important
values (limits!)
amount must be sufficient for a longer time
must be stable for several months
no losses due to the container
no changes due to taking subsamples
Control Samples
Standard Solutions
to verify the calibration
control sample must be completely
independent from calibration solutions
influence of sample matrix cannot be
detected
limited control for precision
very limited control for trueness
Control Samples
Blank Samples
samples which probably do not contain the
analyte
to detect errors due to
changes in reagents
new batches of reagents
carryover errors
drift of apparatus parameters
blank value at the start and at the end allow
identification of some systematic trends
Control Samples
Real Samples
multiple analyses for range and
differences charts
if necessary separate charts for
different matrices
rapid precision control
no trueness check
Control Samples
Real Samples Spiked with
Analyte
for recovery rate control chart
detection of matrix influence
if necessary separate charts for
different matrices
substance for spiking must be
representative for the analyte in the
sample (binding form!)
limited check for trueness
Control Samples
Synthetic Samples
synthetically mixed samples
in very rare cases representative for
real samples
if this is possible precision and
trueness check
Control Samples
Reference Materials
CRM are ideal control samples, but
often too expensive or
not available
In-house reference materials are a good
alternative
can be checked regularly against a CRM
if the value is well known good possibility for
trueness check
sample material from interlaboratory tests
Which One?
There are a lot of possibilities
Which one is appropriate?
How many are necessary?