Quick start
1
2
4
3
Click the main menu <Configuration € Analyte € Create, edit, archive> to open the dialog Definition of analytes.
Two fields must be filled in to create a new analyte:
Type “Analyser X” in the entry field <Section>
Type “Glucose” in the entry field <Analyte>
Press the button <Create analyte>
That’s all: The analyte “Analyser X – Glucose” appears in the left tree view of the background main window. Three control charts are ready to plot QC points.
Press the button <Exit>>, the key <Escape> or click the Windows close-box to quit the dialog.
Click the main menu <Data entry € QC vector> (shortcut F2) to open the dialog Entering QC vector.
1
2
The date and time are initialised at the present date and time.
Type in the concentrations of the 3 control levels, navigating from field to field with the keys <>, <>,
<Tab> or <Shift+Tab>.
The association of the concentrations of the three control materials that were assayed in the same run of samples is named a QC vector.
Click the button <Rate> to open the validation dialog.
1
The first QC vector in a chart is always valid (green color of the background) because there is not yet any control interval.
1- Click the button <OK> to validate the first entered QC vector.
When a new analyte is created, the control parameters are set by default to learning mode. This means that a provisional reference pool is built with the previously entered QC vectors and updated whenever the operator validates a new control vector.
The level-1 QC chart of the example is updated with the first concentration (2 mmol/l) which becomes the new provisional target. The SD is still arbitrary because we need at least two concentrations to begin to estimate variability.
Fill in the charts with fake data
1
2
Click the main menu <Data entry € Create fake QC data> to quickly create a great number of demo QC vectors.
Fill in the entry fields of the dialog
Create fake data as shown opposite.
Click the button <Create data>
3
The red/yellow flashing pane <QC> at the bottom-left of the main window warns that QC vectors are waiting for validation in the pending queue of MultiQC.
1
Click the red/yellow flashing pane <QC> to open the dialog Pending data and display the 20 fake QC vectors that were randomly created.
2
Click the menu <Validate € Automatic> to automatically validate the QC vectors which are in- control. The control interval has not yet been estimated at this stage. Therefore all of the validity icons are in-control (green dots or arrows) and all the QC vectors are accepted in the
control charts.
In normal use, the queue of pending data is made of QC vectors for different analytes. It is the reason why you must reselect the working analyte <Analyser X / Glucose> when leaving the Pending data dialog.
There are now 21 QC vectors in the charts. This number is not high enough
3
3
3
for a reliable estimation of CVs. The true CV used to create the fake random data was 2.0 %. The CVs observed in the random sample below are 2.2 %, 1.6 % and 2.1 %.
Control charts
11
1
2
3
Quit MultiQC
4
5
6 7 8 9
13
12
10
Main menu.
Login of operators.
Choose between displaying current or archived QC data.
Two-level tree view to select the current analyte. Analytes (second level nodes) are grouped in sections (first level nodes).
A splitter between the tree view of analytes and the charts changes the width of the panels.
Flashing pane to remind that QC data are waiting for validation in the pending queue.
Flashing pane to remind that repeatability data have been received from a connected analyser.
Flashing icon to remind that the time interval between backups elapsed.
This eye shows that the current analyte is under monitoring for the time of its QC.
Two event-bar reminding changes of bottles of control materials or of control parameters.
Three Levey-Jennings control charts.
Three event- bars reminding of analytical events likely to impact the control charts (reagent events, calibration events, instrument events).
When the number of QC vectors is greater than the number of displayed points, a scroller is enabled to shift the time axis.
Two buttons can also scroll the dates.
You can also scroll the charts by left-clicking
outside a QC point. The mouse cursor becomes a hand that horizontally drags the charts as long as the left button is not released.
The visual aspect of the charts can be adjusted with buttons at the bottom of the window.
1
2
3
This dropdown pick list sets the number of displayed QC points. When <Public>is checked the value of the dropdown pick list applies to all of the analytes. If <Private> is checked, only the current analyte is changed.
This dropdown pick list expands the range of displayed concentrations along the vertical axis.
The button <Show/Hide> opens a local menu with three checkboxes that enable or disable special features :
The EWMA line (exponentially weighted moving average)
The T chart in multivariate QC can be replaced by a simple row of green/red indicators according to the global validity status of the
5
4
relevant QC vector.
Two additional menus are popped up by a right-click of the mouse :
In the left panel, the items of the popup menu expand or contract the nodes of the tree view of analytes
In the right panel, the popup menu shows the same items as the main menu <Current analyte>.
2
6
3 1
9
7
8
5
4
Actual target line (solid green line).
Target value assigned by the manufacturer of the control material (dashed green line).
Upper and lower control limits (solid red lines). They are calculated according to the ARL entered in the dialog Definition of analytes (section 6.1) as explained in the FAQ (section 25).
EWMA (solid red line) between its upper and lower control limits (dotted red lines).
EWMA (solid blue line) of the twinned analyte.
Hint window that pops up whenever the mouse cursor hovers a QC point or a date. This hovered QC point is highlighted yellow. The hint window shows the date, the concentration and the estimated bias (relative difference between the EWMA and the target value).
Hollow style (white-centred) indicating a comment associated to a QC point. This comment can be read in the hint of the point.
This outlier is displayed at the bottom of the graph in a triangular style. It was rejected. So it is drawn pink and not linked to other points. This point is stored but not included in the statistics.
Identification of the current control material with the relevant statistics. A more detailed table of control limits is available by means of the menu <Current analyte€Limits>. The background colour is a reminder of the parameter mode (see colour codes in section 7.2).
10
The background colours of the charts can be scaled as in the picture above or restricted to three colours. In the latter case, the limits between the green and the yellow areas are [target ± 2/3 control interval]. Thus, if the control interval (limit yellow/red) is ±3 SD, the limit green/yellow is ±2 SD.
Every operator can customise the style of its own charts (see section 13.3) or the style associated to the anonymous operator “Noname”.
There is a special set of background colours for acceptance charts (never scaled), independent from the common background colours of univariate and multivariate QC.
Item clicked
Left-click
Right-click
Any graphical item (QC point, EQA flag, event icon.
Opens the relevant table and selects the row of the clicked item
Pops up a menu to edit or delete the clicked item (if you are a supervisor)
Outside a graphical item
Hand cursor that scrolls the charts
Pops up the menu <Current analyte>
External Quality Assessment (EQA) is a service where participating clinical laboratories are sent samples on a regular basis which they test as if they had come from patients. Results are returned to EQA centres which provide a report that compares the participant's performance with that of all laboratories using the same test method (peer group).
EQA provides a retrospective evaluation of the trueness of an assay. The delay between the assay and the return of the report can range from a few days to several weeks. Before deciding a corrective action in today’s analytical method on the basis of a passed EQA report, it is necessary to refer to the internal QC on the day of the EQA assay. Therefore both of them are displayed on the same graphs within MultiQC allowing thereby a complete follow-up of precision and trueness of analytical methods on a unique screen.
Flags
MultiQC draws EQA flags on the control charts. These flags do not show the actual concentrations of EQA materials, which cannot be directly plotted among QC concentrations, but the percent error returned by EQA providers. EQA flags are drawn as small black-centred targets surrounded by a green, yellow or red ring according to the rating set out below.
Position
EQA flags are located at an ordinate that the EWMA red line should cross if the method were not biased against its peer group. Here is an example of the calculation :
The EQA flag is drawn 2.1% higher than the EWMA line
Given the following EQA report :
Laboratory EQA assay : 39.8 g/l
Mean of the peer group : 40.65 g/l
Lab bias : -2.1 %
The relative bias of the laboratory against its peer group is -2.1 % at the time when the EQA assay was performed. This is a retrospective diagnostic. It means that patients and internal QC results should have been 2.1% higher if the laboratory was aligned on its peer group.
An important property of the EWMA is that it is a forecast of where the process will be at the next time period. Therefore the EQA flag is drawn 2.1% higher than the latest calculated EWMA that estimates the set point of the analytical process when the EQA assay was performed.
Colour
Colour of EQA flags depends on the extent of the bias as compared to tolerance:
Green : Bias < ½ Tolerance
Yellow : Bias < Tolerance
Red : Bias > Tolerance
Interpretation
EQA flags on a QC chart are never aligned on a common horizontal line because every EQA assay is affected by the same random variability as patient or QC assays. EQA flags draw a secondary chart superimposed on the regular Shewhart’s chart.
Daily QC for the analyte “plasma sodium” with one EQA assay every fortnight.
The decision to correct an analytical bias must be based upon several agreeing EQA deviations. Compensation is particularly necessary when facing red or yellow EQA flags. The final aim is to get EQA flags equally distributed around the EWMA line which denotes that the analytical method is centred on the mean of the peer group.
Example
The picture below shows the daily QC for the Olympus hs-CRP method, control level 3. An acceptance chart is used because of the high capability of the analytical method. (Cp= 2). The medically allowed tolerance is 12%. Five EQA flags are visible. Three of them are green which denotes a bias against the peer group that is less than 6% (half tolerance). The two other ones are yellow because the bias is greater than 6 % without surpassing the medical tolerance of 12%.
There is a discrepancy between three target values :
The current target value (49.5 mg/l) of the QC chart which is based on the calibrator of the CRP method (Olympus).
The target value (48.6 mg/l) printed on the flyer of the manufacturer of the control material (Biorad).
The target value (46.0 mg/l) suggested by the peer group built by the EQA provider (Randox).
Target value (46 mg/l) suggested by EQA returns (Randox).
Target value (48.6 mg/l) printed on the flyer by the manufacturer of the control material (Biorad).
Target value (49.5 mg/l) based on the calibrator of the CRP reagent. (Olympus)
Daily QC for the Olympus hs-CRP method with one EQA assay every fortnight.
Which one is the right target value ? Nobody knows but the clinical chemist must however make a choice.
Practice shows that the target values provided by manufacturers of control materials are not reliable. So the Biorad target value (48.6 mg/l) is ignored.
The peer groups of EQA providers are often bringing together different analytical methods. It is presently the situation because Olympus CRP and Olympus hs-CRP are not separated. So the Randox target value (46.0 mg/l) is ignored.
Finally the Olympus-calibrator based target value was kept unchanged because there was no serious evidence that it be wrong.
Remarks
In multi level QC, each EQA flag is displayed in only one QC chart : the one which has the target value the nearest to the mean of the EQA peer group.
EQA flags are placed on each control chart in relation to the EWMA. So they cannot be drawn if this average is not yet available at the time of the EQA assay. This is the case when an EQA return is registered before the first QC vector. Even if nothing is displayed on the charts, data are however visible in the table of EQA results.
EQA returns cannot be validly interpreted if the analytical method was not stable (in-control) when the assay was performed.
Keyboard entry of data
1
2
MultiQC processes two kinds of data:
Internal quality control (IQC) vectors.
External quality assessment (EQA) returns.
Select the relevant sub-menu in the main menu <Data entry > or use the shortcuts F1 and F2.
For a big amount of data it is more appropriate to use a profile (section 3.4), on-line acquisition of data (section 24) or importation by means of a file (section 23).
2
4
5
6
7
8
1
9
3
Double click
Select the analyte by a left-click of the mouse on the tree view of analytes.
Click the main menu <Data Entry € QC vector> to open the entry dialog (shortcut F1).
Or more quickly, you can double-click the name of the analyte in the tree view.
The entry dialog is initialised with the current time. So the data and time fields usually do not need to be changed.
Fill in the concentrations fields. Use the keys <>, <>, <Tab> or <Shift+Tab> to navigate between the entry fields.
You can type in a comment or select a pre-defined one from the drop down pick list. Refer to section
13.2 to learn how to edit these pre-defined comments
Refer to section 11.5 for the use of the box <Initialise EWMA> and to section 11.10 for the use of the box <Previous reagent bottle>.
Press the button <Rate> or the key <Enter> to switch to the validation dialog.
To quit press the button <Cancel>>, the key <Escape> or click the Windows close box.
While entering data, the background charts remain active. The analyte can be changed in the tree view, the charts can be scrolled, and the hints can be opened hovering above the QC points or events with the mouse.
Entering and rating an EQA return
Before entering an EQA return, the tolerance limits of the relevant analyte must have been updated as shown in section 0.0, else the default tolerance interval will apply (± 10%).
Click the menu <Data entry€EQA result>.
1
2
4
5
6
7
While entering EQA data in the entry dialog, the background charts remain active. The analytes can be changed, the charts can be scrolled, and the hints can be opened hovering above the points with the mouse.
Fill in the fields: <EQA ID>, <Date>, <Time>, <Lab value>, <Peer group>. Use the keys <>, <>, <Tab> or
<Shift+Tab> to navigate between the entry fields.
Click the button <Rate> to display the EQA rating.
Frame EQA rating:
Error: Relative or absolute according to what is 3
applying.
Coloured bar: This coloured bar graphically
displays the bias of the assessed analytical method. This bar is calculated in relation to the allowed tolerance. An EQA result is OK (green spot) when the bias is lower than half the allowed error. The result is acceptable (yellow spot) when its bias is lower than the allowed error. Unacceptable EQA results are flagged red.
When relative tolerances and errors are displayed, the sign % is appended to the figures. When absolute tolerances and errors are displayed, the measurement unit is appended except when this unit is itself a percentage (e.g. for the hematocrit ratio).
Use the button <Edit> to correct a misentry.
Click the button <OK> to validate the entries. A new EQA flag is then displayed in the shewhart’s charts.
The button <Undo> is disabled when entering an EQA report. It is enabled only when editing.
To quit press the button <Cancel>, the key <Escape> or click the Windows close box..
Creating entry profiles for IQC and EQA data
In routine lab work, analysers typically print out QC data level by level and sorted like patient reports. Therefore, it is more convenient to manually enter QC data in MultiQC level by level as they are printed out. The programme can create entry profiles made of several analytes sorted in any order.
Likewise, some EQA providers return reports every week in the same format. An entry profile is also very useful in such a situation.
To create a profile of analytes sorted per your patient reports or your EQA returns, click the main menu
1
2
6
3
7
4
5
8
9
10
<Configure € Entry profiles>.
Creating a profile
In the Profiles dialog, select the tab <QC profiles> or <EQA profiles>.
Drag analytes (or a whole section) from the tree view on the left and drop them in the centre panel.
Sort the analytes now showing in the centre pane by dragging and dropping them or using the <Up>,
<Down>, <Top> and <Bottom> buttons.
Name the profile.
Validate with the button<Create profile>.
Editing a profile
To add an analyte to an existing profile, select the profile in the right pane and work like in (2).
To remove an analyte from an existing profile
select the profile in the right pane
select the analyte to be removed in the centre panel
delete it with the button <Delete line(s)> or press on the key <Del>.
After editing a profile, do not forget to click the button <Update profile> for the changes to be taken into account.
Any change can be reversed with the button <Undo> as long as the dialog has not been closed.
To quit press the button <Exit> , the key <Escape> or click the Windows close box..
Typing in QC data by means of a profile
Click the main menu <Data entry€QC Profile€XXX>. This menu is disabled as long as a QC profile has not been created. Selecting a profile produces a new entry window with a grid to enter the QC concentrations of several analytes, level by level.
Over-long lists are common causes of misentries so should be avoided. It is often better to create two middle-sized profiles than a single longer one.
1
2
3
4
Fill in the grid. Proceed column by column, going to the next cell with the keys <Tab>, <> or
<Enter>.
All of the analytes in a profile are not necessarily QCed with the same number of control materials. The unused cells of the grid are disabled.
When finished, the button <Validate> inserts the data into the pending queue, that you can validate immediately or later.
To quit press the button <Exit>, the key <Escape> or click the Windows Close box.
Typing in and rating EQA returns by means of a profile
1
2
3
4
5
Click the main menu <Data entry€EQA Profile€XXX>. This menu is disabled as long as an EQA profile has not been created. Selecting an EQA profile produces a new entry window with a grid to enter a whole EQA report.
Enter the EQA identification, the date and the time when the EQA tests were actually performed.
Type in the lab result, the peer group mean and a possible comment for every analyte. At this stage, the columns on the right hand side are grey.
At any time you can rate the entered values with the button <Rate> to display the rating information in the right hand columns.
The button <Save> stores the EQA data on the disk and closes the dialog.
To quit press the button <Exit>, the key <Escape> or click the Windows Close box.
When analytes are cloned (= duplicated) because a new lot of control material is about to be started, each EQA return is also added to the duplicated analyte (~Analyte) provided there is at least an IQC point older than the EQA return. It must be remembered that the EQA flags are positioned on the Shewhart’s charts in comparison to the EWMA curve and that there is no EWMA curve without QC point.
When scanning the different analytes of the profile to add the EQA returns, MultiQC can meet a “busy” analyte (i.e. an analyte currently processed by another workstation of the network). In such a case adding the relevant data is postponed and enqueued to be performed later (see section 0.0).
The menu <Data entry-->Create fake QC data> creates a set of random, multi-normal data. It is very useful for demos or training.
1
2
3
4
6
5
Type in :
The number of QC vectors to create.
The CV of the method (%).
The coefficient of correlation between control levels (0 to 0.999). There are N(N-1)/2 coefficients between N control levels. Practice of clinical chemistry shows that these coefficients are generally very similar. Therefore MultiQC provides a unique field to enter a common value.
The date and time of the first QC vector.
The target values of each control material.
After validating with OK, the fake data are enqueued in the pending list of data for future validation of QC vectors.
Validating QC vectors
1
4
3
2
5
6
7
8
9
The QC vector to be validated is displayed in the grid of the validation window and simultaneously the charts of the background window are updated with provisional points. These new points are displayed bigger and coloured green, yellow or red.
Icons in the column Ctrl denote the deviations around the target
: Inside ±1/3 of the control interval
: Inside ±2/3 of the control interval
: Inside the control interval
: Outside the control interval
You can revise the comment if needed.
To suppress one QC level, tick the relevant box in the column <Sup> of the validation grid. The button in the row of headers ticks/unticks the whole column.
If you tick a box in the column <Rej>, the QC result will be stored but will not be included in the statistics. On the charts, the point will be pink-coloured and not linked to other points to show that it has been “rejected”. The button in
the row of headers ticks/unticks the whole column.
Refer to section 11.5 for the use of the box
<Initialise EWMA>.
The button <Edit> rolls back to the entry dialog in order to modify the previously entered data.
The button <Pending> postponed validation. The QC vector is enqueued in a file for a future validation.
Validate with the button <OK>.
To quit without validation, press the button
<Cancel>, the key <Escape> or click the Windows Close box.
All the deviation icons may be crossed by a small red arrow to remind of an out-of-control EWMA (See section 4.5).The background colour of the grid denotes the global validity of the QC vector:
Grey: all of the levels are absent, rejected or suppressed (figure below, left).
Green: every not absent, not rejected or not suppressed level is in-control (figure of the previous page).
Red: at least one level is out of-control (figure below, right).
Hotelling's T is calculated and plotted in the bottom chart. Notice that the relative T is charted as the ratio T/T limit, according to the chosen ARL.
The global validity of the QC vector depends on the value of the relative T as compared to 1. An exception is raised when the relative T is greater than 1.
: Non significant Hotelling’s test (Relative <1)
: Significant Hotelling’s test (Relative>=1)
A discrepancy may sometimes occur between the global validity status of a QC vector according to the Hotelling’s test and the separate validity status of each level. See the last FAQ in section 25 for possible interpretations.
In the above picture, the QC vector is globally in-control according to the multivariate test (relative T <1). The third control material is however individually out-of-control. The arrow in the third row should be red. Priority is given to multivariate testing. The background colour is thus green denoting the global in- control situation.
: Orange arrows denote individually out-of-control QC points that should be red in univariate mode but that are superseded by a globally in-control multivariate validity test (icon ).When the Hotelling’s T cannot be calculated because one control level is absent or rejected, multivariate QC is validated in univariate mode.
Validation in acceptance charts
The global validity of a QC vector results from two simultaneous criteria
The global univariate validity as explained above
No capability alarm: The bottom row of the grid shows the lowest Cpk value observed among the control levels. The number of the relevant level is displayed between brackets (2 on the picture). An exception is raised if this lowest Cpk is less than 1. See section 6.1 to enhance the default threshold 1.
1
2
: Icons that show whether the capability of the analytical process is high enough to meet the medical tolerance. This icon is red if the lowest Cpk value among the QC levels is less than the threshold entered in the dialog Definition of analytes (default threshold = 1).
3
Control interval
Latest known EWMA
QC point under validation
Acceptance interval
The control interval of an acceptance chart is based on the latest known EWMA and three short-term SD’s.
The Cpk value taken into account for the acceptance test is the Cpk value that would be observed, assuming that the QC point under validation is accepted.
Validating a set of QC vectors from the pending queue
The pending queue of MultiQC is a FIFO file (First In, First Out) that stores QC vectors
The validation of which was postponed by the button <Pending> of the validation dialog.
Entered in an IQC profile but not subsequently validated.
Received from a serial port thanks to a specialised QC Receiver Interface program.
Retrieved from the clipboard by clicking the main menu <Data entry € Paste>
Retrieved from an appropriately formatted text file by clicking the main menu <Data entry €
File>.
Created by the menu <Data entry€ Fake>.
The panel QC of the status bar of the main window blinks yellow/red as soon as one QC vector has been enqueued. Click the blinking panel to show the table of pending data.
1
2
Double click
3
Coloured deviation icons are the same as for separate validation (refer to section 4.1, 4.2 and 4.3 for significance). The background colour of a row is red whenever the relevant QC vector is globally out-of- control.
1, 2- The main menu and the local menu that pops up with a right-click of the mouse inside the window give access to the following functions:
Manual validation successively presents all the vectors of the pending queue in the validation dialog (see section 4.1) to be individually accepted, rejected or deleted.
Automatic validation triggers an automatic validation. Pending QC vectors are checked one by one and inserted in the relevant analytes if they are in-control. Out-of-control vectors remain in the queue and need a manual validation.
Warning when working in learning mode: Manually validating and accepting one by one several QC vectors for the same analyte is not the same as automatically validating them. In the former case, the control interval is revised after each new accepted result whereas in the latter case the control interval remains unchanged during the whole validation process.
Deletion of all the QC vectors of the pending queue or of only the selected vectors.
Sorting the list of pending QC vectors either by date/time of by analyte name.
The main menu <Export> is aimed at re-processing pending QC data by means of another programme. The pending queue is cleared and its data are sent either to Microsoft Excel or to the clipboard. After re-processing, QC data can be retrieved in MultiQC by means of Copy/<Data entry -> Paste> .
3- A quicker way to validate a row is to double click it.
When closing the window (Menu <Exit>, key Escape or Windows Close box) without performing a complete validation, the remaining not yet validated QC vectors are inserted again in the pending queue and the panel QC of the status bar blinks again.
Drifts and trends are frequent in clinical chemistry. They may lead to an unacceptable number of non- conforming analytical results. The EWMA curve is a good tool for signalling such insidious errors that are not or lately flagged by conventional QC
charts.
Significant drift
EWMA
control interval
The control interval of the EWMA is limited by two red dashed lines on the QC plots (see opposite picture). A drift of the EWMA becomes significant as soon as the solid red line crosses the upper or the lower red dashed line.
When validating a QC vector, an out-of- control EWMA is pointed out by a thin red arrow superimposed on the regular validity icon.
These small red arrows remain hidden when validating the 3 first QC vectors that follow a re-initialisation of the EWMA. because an EWMA based on less that 4 points is not reliable enough.
With the default values of the smoothing factor and of the ARL, the EWMA becomes out-of-control when the bias exceeds about one standard deviation.
Most often a one-SD bias does not need an immediate corrective action. This action can be postponed particularly if the capability of the analytical process is high. However the little red arrows herald the likely next occurrence of out-of-control QC vectors. A precautionary re-adjustment of the set point of the method might be useful to prevent a possible break of the analytical flow.
Analytical events
It is necessary to be acquainted with some analytical changes to be able to understand and troubleshoot the hitches and snags revealed by QC charts. MultiQC logs these events to make them easier to consult from the QC charts.
Ten kinds of analytical events are logged by MultiQC. They are shown by specific icons.
Remark about reagents
New lot of reagent and new bottle of reagent 
Calibration and reagent blank
New bottle of control material
Verification of the reportable range 
Comparison of methods
Repeatability
Instrument maintenance
Change of control parametersEach kind of event has a dedicated event-window both to enter new events of the same kind and to display the cumulative history of these events. Event-windows are opened by clicking the relevant submenu of the main menu <Events> (shortcuts F4 to F11)..
2
1
Specific icons painted in five event-bars remind the occurrence of events:
Three bars are located at the bottom of the QC charts, under the row of dates:
the reagents bar plots the events:
the calibrations bar plots the events:
the analyser bar plots specific icons made of one character associated to a pre-defined maintenance action or the generic icon
Two event-bars are located at the top of the QC charts:
3
The QC parameters bar. The icons remind of the dates when the control parameters were changed.
The QC bottles bar plots the events:
When the mouse is moved over an icon of any event-bar, this icon is highlighted yellow and a hint shows the contents of the underlying event. When a highlighted icon is left-clicked, the relevant event-window is opened.
Events do not modify the scale of dates. Icons do not use a reserved column of the time-plot. They are placed below the next QC vector that might be affected by the event.
Several events may occur at the same time. Their icons should be superimposed in the event-bar. Practically they are replaced by a special icon .
When an icon is left-clicked, a local menu pops up. It details the multiple events that are hidden under the common icon .
4
Any icon in an event-bar can be right-clicked. This mouse action pops up a local menu with an option to erase the clicked event.
Warning: When the latest parameters-change event is erased, the control parameters are automatically rolled back to the previous ones
The reagent remark window logs the events Reagent-remark. A reagent-remark is a short text of 30 characters. It reports something that occurred in an analytical method and which might affect all of the following QC points. Remarks must not be confused with comments appended to QC vectors. The latter only concern one QC vector at a given time.
1
2
3
4
5
The reagent remark window is opened by clicking the menu <Events € Reagent remark> or on an icon
in the events bar.Entering a new remark
Select the tab <New reagent remark>.
The fields <Date> and <Time> are initialised to the present date and time.
Type in the text of the remark or select a pre-defined one from the drop down pick list.
Refer to section 13.2 to learn how to edit these pre-defined comments
Validate with the button <Save>.
Editing a remark
The left tab <History> of the remark window shows the cumulative history of remarks.
The cursor of the table of remarks 1
is synchronised with the cursor of the relevant event-bar in the background window. If you click a row in the table, the corresponding icon in the event-bar at the bottom of the main window is highlighted and vice versa.
You can edit a remark only if you are a supervisor or if there is no
declared supervisor among the users 3
of MultiQC. Press the button <Edit>
or double-click the line to edit in the 2
list of remarks.
To delete one or several remarks, select the relevant lines with the mouse (Shift+click and Ctrl+click for multiple
25
selections as usually in any Windows program) and press the button <Delete> or the key <Del>.
Other event-windows derived from the remark window
MultiQC can display six event-windows which are derived from the remark windows described above. These event-windows work like the previous one but have specialised additional entry fields.
1
2
Calibration window
Logging calibrations within the QC charts is very useful to be able to construe the frequent shifts of the QC plots resulting from the uncertainty of the calibration processes.
Kind of calibration : <full calibration> or simple <reagent blank>
Calibrator lot (optional): This field is enabled only when the box <Full calibration> is checked. By default the lot is the lot entered with the previous full calibration. It should rarely have to be changed, only when a new calibrator lot is started.
Automatic acquisition of calibrations by a serial connection is desirable whenever the analyser outputs calibration messages towards the LIS.
The new set of QC bottles window
3
QC materials are theoretically stable from the time when the bottle is opened or reconstituted to the time when the bottle is empty. In fact multi-analytes QC materials sometimes include unstable molecules whose concentration decreases from day to day. These materials should be discarded but sometimes they have no competitor and we must learn to get by with them. It is then necessary to log in an analytical event the day when the bottle of QC material was started to be able to understand the variations of its contents.
The button <Extend the event …> is provided to add the icon to all of the analytes that are controlled with the same potentially unstable control material.
1
2
The reagent lot / bottle window
Logging the changes of reagent bottles within the QC charts is very useful to construe the frequent shifts of the QC plots due to the replacement of an old bottle by a fresher one.
Select <Lot + bottle> or <Bottle only>
Typing in the <reagent lot> is compulsory but the field <bottle number> can remain blank.
Automatic acquisition of bottle numbers by means of a serial connection is desirable whenever the analyser outputs bottle messages towards the LIS.
The linearity window
This window gives access to a linearity software included in MultiQC. See section 18. This module is not intended to initial validation of new analytical methods but to quick and periodic verifications of established methods.
The method-comparison window
This window gives access to a comparison software included in MultiQC. See section 19. This module is not intended to initial validation of new analytical methods but to quick and periodic verifications of established methods.
The repeatability window
This window gives access to a repeatability software included in MultiQC. See section 17.
The instrument-window
Logging the maintenance actions of an analyser within the QC charts is very useful to construe the frequent shifts of the QC plots resulting from modifications of the instrument.
2
1
4
3
Type in the description of the maintenance- event (for instance “Lamp change”) in the field <Instrument event> and save it with the button <Save>. The event-bar will then show the non-specific icon
For a recurrent maintenance-event, it is more practical to register it as a pre-defined event with a specific icon. Thus it will be more easily entered by dropping down the pick list.
An instrument event entered for one analyte can be extended to all of the analytes of a whole section. This is often needed because the analytes belonging to a given section are generally equally impacted by the same instrument event. Just check one or several boxes(3) before pressing the button
<Save>.
When scanning several analytes to add an instrument event, MultiQC can meet a “busy” analyte (i.e. an analyte currently processed by another workstation of the network). In such a case, adding the event is postponed and enqueued to be performed later (see section 0.0).
5
6
7
8
It is recommended to create pre-defined events with specific one-character icons for recurrent instrument-events.
Click the button <Edit pre-defined instrument events> (page above) to open a new dialog.
Type in the caption of the recurrent event.
Select a one-character icon for instance L for “Lamp change”.
Click the button <Add> to add the new item
to the list of pre-defined events
Validate the list with the button <OK>
The new pre-defined event is now available in the drop down pick list 2.
This window displays the changes of control parameters. On the contrary of other event windows, there is no a tab to manually enter new changes of parameters. The only available tab is History. Parameter- change events are never manually entered or edited. A new row is automatically added to the history table whenever the control parameters are changed, which can take place by means of two actions:
Main menu <Configure -
1
2
3
> Analytes -> Control parameters> (see section 7.1)
Button <Set Targets = EWMA> of the bias-line window (see section 21.3)
The first row shows the initial parameters when the analyte was started. It cannot be deleted.
The following rows show the history of the control parameters.
A second special feature of the parameter-change window is
its button <Delete row>. It erases the selected row and the relevant event as in any other event windows but, when the selected row is the last one, the button rolls back the current control parameters to the previous ones.
This window shows in the same table all the events of an analytical method. Events are sorted by date/time of occurrence. To open this table:
Click the main menu <Events € All events>
Right-click inside any QC chart and select <Table of events> in the popup menu that opens.
Each row of the table can be right clicked or double- clicked to open the event window of the specific class.
Keeping the latest events when archiving and cloning analytes
When a new batch of control materials is started, the former QC data are archived and the control charts are cleared from their QC points. Events are also cleared except the following ones which are kept because the analytical method is always working with the former reagents and the former calibration:
The latest calibration.
The latest reagent lot / bottle.
The latest verification of the reportable range.
The latest comparison of methods.
The latest repeatability
These events are thus both stored within the archived chart and duplicated at the beginning of the newly restarted chart. Likewise, the same duplication of events is performed when an analyte is cloned because the newly created copy of the analyte is assumed to work with the same reagents and the same calibration.
Definition of analytes
Click the menu <Configure € Analytes € Create, edit, delete, archive> to open the dialog Definition of analytes.
1
2
3
4
5
6
7
8
9
12
11
10
Type in the name of the <Section> where the new analyte must be placed. If this is a new name, the section will be created in the tree view of analytes of the main window; else the analyte will be added to a pre-existing section.
Type in the name of the <Analyte>. Two analytes with the same name cannot be created in the same section.
Enter the number of <Decimal places>. This number will be used to format QC data before displaying or printing them. An additional decimal place will be used for the means. Note that computing and saving to files is made in full precision (real numbers). Rounding displayed concentrations does not damage the precision of calculations.
Choose a number of <QC levels> from 1 to 6.
Type in a <Unit name> or select it from the drop-down pick list.
The button <Edit list> opens a dialog to customise the pre-defined list of unit names.
MultiQC can monitor an analytical process by three control methods (see section 6.4):
Univariate quality control
Multivariate quality control.
Acceptance quality control.
This group-box of fields depends on the control method you have selected.
Univariate and multivariate QC charts: Thee ARL (Average Run Length) defines the risk of false rejection. An ARL equal to 370 means that an average of one false rejection occurs every 370 assays when the analytical method is in-control. This number 370 comes from the ARL of the historical Shewhart chart with a 3 SD wide control interval..
Acceptance chart: By default, capability alarms are triggered as soon as the capability index Cpk becomes lower than 1.00 but some authors advise a minimum operating value of Cpk equal to 1.40.
Click the button <Create analyte> to create a new analyte.
The button <Update analyte> is provided to change some features of a pre-existing analyte like the number of QC levels, the unit name, ….. but also
to rename an analyte : Change the analyte name and press <Update analyte>.
to move an analyte from a section to another one: simply rename the field <Section> with the new destination section and press <Update analyte>.
It is sometimes useful to have a look at the other analytes. They can be accessed by the button <List of analytes>.
QC data of each analyte are stored in a separate file with a random name and the extension <.qcf>. The path to the current file is shown in the bottom status bar.
To leave the dialog without taking into account the modified fields, press the button <Exit>, the key
<Escape> or click the Windows close box.
Editing / deleting an analyte or a section
Click the menu <Configure € Analytes € Create, edit, delete, archive> to open the dialog Definition of analytes .
While editing a section or an analyte, the background charts remain active. The analytes can be changed, the charts can be scrolled, and the hints can be opened hovering the plots with the mouse.
Editing / deleting a section
Select the section to be edited in the tree view of the main window. The four buttons of the frame
<Section> become enabled.
1
3
4
2
<Rename>: to change the name of the selected section.
<Delete>: to erase all of the analytes in the selected section.
<Clone/Unclone>: see section 14.2.
Editing / deleting an analyte
Select the analyte to be edited in the tree view of the main window and change the required fields of the analytes dialog.
After editing an analyte, do not forget to click the button <Update analyte>. The charts in the background window are immediately updated.
5
7
6
If you change the section name, the analyte is moved to the new section, which is possibly created if it is a new one.
Special functions are available in the frame <Analyte>
<Delete>: to erase one analyte.
<Clone/Unclone>: see section 14.2.
Section and analyte names are case-sensitive. The analyte Glucose is not the same as the analyte glucose.
Changing the section of an analyte
It is very easy to re-organise the analytes in the main tree view of MultiQC.
Open the dialog Definition of analytes.
Enter a new section name.
Validate with the button <Update analyte>.
If the new section name already exists in the tree view, the analyte will be moved to this section. If the new section is unknown, it will be created.
When tolerance is tight, it is essential to achieve stability of an analytical process to keep the method in- control. Univariate or multivariate QC are intended for this aim.
Univariate QC: Clinical laboratories usually monitor their analysers with control materials of typically low, medium and high concentrations. The common practice is to maintain an independent (= univariate) chart for each control level.
Multivariate QC: Multivariate quality control does not monitor control levels one by one but a global QC vector, the components of which are made of the concentrations of the three control materials. The theory of the Hotelling’s T is explained in any quality control textbook.
When tolerance is wider, a process can perform out-of-control and produce however acceptable results. Relaxing the level of surveillance provided by the standard Shewhart’s chart saves time and cuts costs. The acceptance chart accepts drifts of the mean as long as the number of nonconforming assays outside the tolerance limits remains insignificant. Acceptance charts are something new in clinical chemistry.
Therefore a special chapter of this manual (section 20) describes them with more details.
Control parameters
The five parameter modes of MultiQC
Any new analyte is created in learning mode. This is a provisional way to estimate the control parameters that allows an immediate acquisition and validation of QC data without having to type in anything else.
Sooner or later, the parameter mode will have to be switched to one of the four other modes available in MultiQC.
Reference pool : control limits are estimated from a set of QC points.
Specified statistics : control limits are forced from two pre-defined statistical constants: mean and SD (or CV).
Specified Interval : control limits are forced from two external constants: target value and allowed deviation.
Semi-learning : Special parameter mode used when starting a new batch of control materials. CV is kept from the previous batch whereas the chart self-learns its new mean.
Click the menu <Configure € Analytes € Control parameters> to open the dialog Control parameters .
Five radio buttons are provided to select a parameter mode.
1
2
Each parameter mode has its own colour code which can be changed pressing the button <Colours>.
The parameter mode Specified interval is not compatible with multivariate and acceptance control charts.
Shewhart’s control charts are built with two statistical parameters: mean and SD. Hotelling’s charts involve in addition the covariances between QC levels. When starting a new batch of control materials, these statistical parameters are most often unknown and cannot be estimated as long as a reliable reference pool has not been collected. MultiQC however immediately starts charting with provisional floating parameters.
The two first QC points are charted on an arbitrary ordinate scale. When entering the third QC point, an estimate of SD can be calculated from the two first ones and statistical QC can start. Of course, at this stage, alarms must not be taken too seriously but graphical display is immediately available. The beginning of the Hotelling’s chart comes later because the variance-covariance matrix needs at least N+1 independent QC vectors to be estimated (N is the number of control levels).
1
2
3
4
The learning mode is selected by default whenever a new analyte is created. It is also possible to manually set the learning mode by clicking the main menu <Configure € Analytes € Control parameters>.
Check the radio-button <Learning>. The analyte is immediately updated to learning mode without having to press the button <Apply> which is disabled.
The table cannot be edited : it displays means and CVs calculated by MultiQC on the basis of all the available valid QC points.
The button <Undo> reverts to the initial parameters and parameter mode.
The button <Limits> shows a more detailed view of the control parameters and intervals
To quit the dialog, press the key <Esc> or the button <Exit> or click the Windows close box.
It is risky to stay in learning mode with statistical floating parameters for too long. A slow drift of an analytical method may gradually shift the mean, widen the control range and stop the occurrence of out- of-control alarms. It is necessary to lock the computation of limits as soon as a stable reference period has been collected.
1
2
3
4
MultiQC can select a reference historical pool made of several unconnected series of QC vectors. This pool can be easily expanded when the number of QC vectors increases during the first months of use of a new batch of control materials. The final aim is to get a reference pool made of at least a hundred of QC vectors, during which the analytical process was stable.
The charts below show how MultiQC displays its reference pools : The QC vectors belonging to the pool are coloured light-blue. The relevant range of dates is also light-blue back grounded.
To select a reference pool in a series of existing QC points, click the main menu <Configure € Analytes € Control parameters>.
Check the radio-button <Reference pool> in the dialog Control parameters. The response of the background plots to a left mouse-click is thus changed.
Define the segments of the reference pool thanks to adequate click, shift+click and ctrl+click of the mouse on the QC points as shown on the picture below.
The grid cannot be edited. MultiQC updates in real time the columns mean, CV and N after each mouse-click on the charts .
The button <Undo> reverts to the initial parameters and parameter mode.
The button <Limits> shows a more detailed view of the control parameters and intervals. To quit the dialog, press the key <Esc> or the button <Exit> or click the Windows close box.
1- Mouse click
2- Shift + click
3- Ctrl + click
4- Shift + Click
Reference pool made of 36 QC vectors
To be taken into account the reference pool must be large enough to get valid SD’s for every control material. In multivariate mode, an estimate of the covariance matrix must also be calculated. With N control levels, this needs at least N+1 complete and independent QC vectors. If the calculation of control limits fails, the background of dates is made of light blue crosses.
The specified statistics parameter mode is needed to force external statistical parameters on the QC charts. Means and CVs are directly entered into MultiQC without any connection to a historical reference pool.
1
2
3
4
5
6
Click the main menu <Configure € Analytes € Control parameters>.
Check the radio-button <Specified statistics>.
The column Mean of the grid is initialised with the current targets. You can force any other value.
The last column of the grid is intended for acquisition of variability. You can enter either CV or SD. Tick the relevant box. If the mean of one level is nil, entering a CV value is forbidden because the CV for this level would be infinite.
Click the button <Apply> to take into account the entered data. The charts in the background are immediately updated.
The button <Undo> reverts to the initial parameters and parameter mode.
The button <Limits> shows a detailed view of the control parameters and intervals.
To quit the dialog, press the key <Esc> or the button <Exit> or click the Windows close box.
When the statistical distribution of QC points is not estimated but externally known, there is no uncertainty on means, SDs and coefficients of correlation. Control limits are therefore calculated from the Gauss’ law for the Shewhart’s chart and from the Khi2 law for the Hotelling’s chart. It is the reason why the same set of parameters lead to control ranges slightly tighter in specified statistics mode than in reference pool mode. The difference is all the less so as the size of the reference pool is big.
Every analyte should sooner or later be switched to and then stay in specified statistics parameter mode :
After a few months of QC, the EWMA generally suggests a QC target more appropriate than the target previously provided by the reference historical pool.
Analysers are sometimes subject to periods of undiagnosed increase of variability. Repairmen are then reduced to a try-and-see part exchange policy which can last several weeks. So laboratorians are compelled to go on working with increased CVs if the capabilities of analytical processes are great enough.
The specified interval parameter mode is needed to force an external control range on the QC charts. Target values and deviations allowed around these target values are arbitrarily entered without any connection to the actual analytical variability.
This is a simplistic way to QC an analytical method, generally relying upon data provided by the maker of control materials. The process is assumed to be acceptable whenever the assays of materials fall within the specified intervals [target ± allowed deviation].
Click the main menu <Configure € Analytes € Control parameters>.
1
2
3
4
5
Check the radio-button <Specified interval>.
Specified intervals are defined in the grid by a centre value (Target) and a maximum deviation around this centre value. Fields are initialised with the current data.
Click the button <Apply> to take into account the entered data. The charts in the background are immediately updated.
The button <Undo> reverts to the initial parameters and parameter mode.
The button <Limits> shows a more detailed view of the control parameters and intervals.
To quit the dialog, press the key <Esc> or the button <Exit> or click the Windows close box.
Non-statistical QC is always univariate: correlation between control levels are not taken into account.
The EWMV is disabled in non-statistical QC because there is no defined reference SD to test the moving variance. Likewise the control limits of the EWMA cannot be drawn.
This parameter mode must be used when analytical variability is a priori known whereas means have to be estimated. This situation occurs when a batch of control materials is exhausted and a new one coming from the same manufacturer must be started. Coefficients of variation estimated from the older lot have no reason to change and can be kept for the newer lot. Means are however prone to slight changes and need to be re-estimated.
Plots in semi-learning mode are thus self-centring on the new unknown targets, keeping the former spreads of the control ranges.
Click the main menu <Configure € Analytes € Control parameters>.
1
2
3
4
5
6
Check the button <Semi-learning>.
The column CV of the grid is initialised with the present CVs of the analyte. You can type in any other a priori known value of CV.
The column Mean of the grid cannot be manually edited. Means are calculated by MultiQC. They are updated every time when a new QC vector is validated or when the button <Apply> is pressed.
Click the button <Apply> to take into account the new semi-learning parameter mode. The charts in the background are then immediately updated.
The button <Undo> reverts to the initial parameters and parameter mode.
The whole table of estimated means and control limits can be displayed with the button <Limits>
The correlation matrix in multivariate QC
In multivariate QC with specified statistics, it is asked to type in the means and the CVs as in univariate QC. But the programme additionally needs the matrix of correlations between control levels to calculate the Hotelling’s T. There are N(N-1)/2 coefficients of correlation between N control levels. Practice of clinical chemistry shows that these coefficients are generally very similar. Therefore MultiQC provides a unique field to enter a common value.
Logging the changes of control parameters
When control parameters are changed, a parameter-change event is automatically generated and logged in the parameter-change window.
If the control parameters are changed several times without entering a new QC vector, only the latest change is logged.
Control materials are often delivered with data sheets giving a table of target values for different brands of analysers. Experimented laboratorians know that these manufacturer targets often do not match the locally determined targets. As long as we are not sure which targets must be trusted, it is useful to display the manufacturer targets on the QC charts.
Click the main menu <Configure € Analytes € Manufacturer targets>.
The column Working of the grid cannot be edited. It is a simple reminder of the current QC targets
Enter the manufacturer targets in the column Manufact of the grid .
As soon as the button <Apply> is pressed, a new dashed green line appears on the relevant QC plots.
Manufacturer target
= 470 µg/l
The dashed green line is erased when the manufacturer target is cleared.
While entering manufacturer targets, the background charts remain active. The analyte can be changed, the charts can be scrolled, and hints can be opened hovering above the QC points with the mouse.
Analytical capability
Making out a table of the acceptable analytical deviations that do not alter diagnosis, follow-up or treatment of patients is the cornerstone on which a clinical laboratory QC system must be built. For each assay, the permissible variations of the measured concentrations around the true values defines the frontier between conforming and non-conforming. It is a nonsense to claim performing QC without having first defined how much variation can be accepted.
Some authors speak of « allowed error ». This wording is disputable: An error which is allowed is no longer an error. It is the reason why MultiQC borrows from the engineering world the better word
« tolerance » which denotes the permissible deviation of an actual property of a product to the nominal value of this property.
Every analyte is created by MultiQC with a default tolerance equal to 10%. This means that an analytical assay is regarded as conforming if the deviation of any measured concentration to its true value does not exceed 10%. This default tolerance must always be replaced by analyte-specific values.
Analyte-specific tolerances are theoretically derived from clinical needs, biological variation or the state of the art. Unfortunately there are so many discrepancies and inconsistency between makers of regulations or recommendations that laboratories generally prefer the ready-made table of acceptable ranges published by their EQA provider (External Quality Assessment). The equation is :
Tolerance = ½ Width of the acceptable interval
Since medical needs are not clearly expressed, the analytical quality aim of a laboratory pragmatically becomes to pass all external quality assessments without unacceptable returns.
Click the main menu <Configure € Analytes € Tolerance> to open the dialog Tolerance of analytes.
You can define the tolerance within three ranges of concentration:
The whole analytical range (Row: General).
The lower and upper ends of the reportable range (Rows: Lower than and Higher than).
Tolerance can be expressed as a % of the concentration (Column Relative)
3
5
1
2
4
6
7
8
Tolerance can also be an absolute deviation (Column Absolute)
When both Relative and Absolute columns are filled in, MultiQC uses the greatest allowed error for the concentration under question. This means that relative tolerance applies to the higher end of the range and that the absolute tolerance applies to the lower end of the range. The switch concentration <Relative Absolute> is calculated in the column Switch;
When the tolerance of an analyte is changed, the previously entered EQA flags can be re-evaluated if the box is checked.
Do not forget to press the button <Apply> to validate any change in the dialog Tolerance of analytes.
You can have a look at the tolerance data of other analytes by means of the button <Show table of tolerances>.
Capability of analytical processes
Any laboratory assay is spoiled by an inherent uncertainty. Random deviations of results are Gauss distributed and thus theoretically boundless. The uncertainty interval is however conventionally limited to a 6 spread (m 3), letting apart extremely rare deviations, the frequency of which is less than 1/370.
As tolerance limits tell us how much variation can be accepted, a comparison between tolerance and uncertainty becomes unavoidable. It is the purpose of capability indexes.
Capability index Cp
6 SD
UTL LTL
p
C
The capability Cp of a method relates the tolerance interval to the inherent analytical variability. The capability is equal to the ratio of the width of the tolerance interval (Upper Tolerance Limit – Lower Tolerance Limit) to the spread of the natural variations of the analytical process (6 standard- deviations).
Target ± 8%
Briefly, Cp is the ratio of what must be done (the allowed tolerance) to what the process is able to do (the expanded uncertainty of the assay). The higher the capability, the lower is the risk of jumping the tolerance limits and therefore the higher the quality.
Target ± 15%
Histograms of 60-day QC data for amylase and bicarbonate assays.Estimated Gaussian distribution curve of QC data.
Expanded uncertainty of the assay (mean ± 3 SD).Medical tolerance interval. Out-of-tolerance assays
Capability =
LTL
UTL
LTL
UTL
The capability of the amylase method (above-left picture) is about 2. The assay is able to fit medical needs (± 15%) and should never produce out-of-tolerance results even if the QC mean slightly deviates from its target.
The capability of the bicarbonate method (above-right picture) is less than 1. The assay cannot fit medical needs (± 8%). A part of the analytical results does not meet the required tolerance. They are nonconforming. The analytical process must be improved.
One-sided index Cpk
ST
3 SD
UTL mean
pu
C
A major shortcoming of the index Cp is that it may yield erroneous information if the process is not on target, that is, if it is not centred. Cpk denotes the capabilities on one side of the distribution, the side for which the larger proportion nonconforming will result. According to the side (lower of upper) we can calculate Cpl or Cpu. Cpk is the less favourable index between Cpl and Cpu.
Target ± 7.5%
LTL
UTL
LTL mean UTL
< 1
Cpl =
Low-sided capability index
> 1
Cp =
Centred capability index
Looking at the left picture above, the phosphate test has a capability greater than one: the green double- arrow is longer than the light blue double-arrow. The phosphate test would be able to fit to medical needs, but this is only a potential capability.
The actual analytical process is biased. The mean is not centred on the target so that nonconforming results will be issued. Cpl is less than 1 (right picture). Some test results are out-of-tolerance.
The analytical capability window
1
2
3
Uncertainty interval
4
Tolerance interval
For each analyte you can open an analytical capability window either with the main menu <Current analyte€Capability> or with the menu <Performance> that pops up when right-clicking inside any QC chart.
You can browse across the analytes with the button.
The grid at the top of the window shows the distribution parameters for each control material.
The light blue double-arrow shows the uncertainty interval [mean ± 3 SD] in which 99.7% of the analytical production is assumed to fall, provided the method remains in-
control. The light green background shows the tolerance interval for the relevant level of concentration. If the light blue double-arrow does not encroach on the red borders, the number of non-conforming results is expected to be negligible.
A detailed histogram is be displayed for every control level.
The period of evaluation of the capability indexes can be easily set to any time interval with a series of buttons.
Every value of Cp or Cpk is flagged according to the scale below :
Flag
Cp or Cpk
Quality
Comment on the analytical method
More than 2.0
Top
Highly capable. Can work out-of-control up to a point. Acceptance chart is advised.
1.3 to 2.0
Good
Capable but needs a conventional QC to continuously stay in-control.
1.0 to 1.3
Insufficient
Theoretically “capable” but practically prone to non-conformities if a slight drift takes place.
Less than 1.0
Bad
To be improved.
Performance indexes must be considered
To select an appropriate QC method. A capability higher than 2 justifies a "light" QC. Conversely, the lower the capability, the lower are the acceptable errors and the more careful must be the QC
Whenever facing an out-of-control situation to decide if an immediate corrective action is needed or if this action can be postponed because the analytical process remains acceptable. A high capability method can work off-centre without any compensation. A low capability process must be driven much more cautiously and compensated as soon as a deviation occurs.
Tables within analytes
Three tables of data for each analyte
MultiQC can display the raw data in three tables
Table of QC data
Table of EQA results
Table of events. These tables can be accessed
By means of the main menu <Current analyte> or its shortcuts
By right-clicking a chart outside a QC point. This opens a popup menu with the same options as the menu <Current analyte>.
By left-clicking any QC point, any EQA flag or any icon of the events bar. This way has the advantage to directly position the cursor of the table on the row corresponding to the graphical item that was clicked.
There are 3 ways to open the table of QC vectors.
For any point on the charts, move the mouse cursor over the point until it is highlighted in yellow with a visible popup hint and then left-click it.
Left-click the main menu Current analyte €Table of QC (Shortcut Ctrl + F1).
Right-click anywhere in a chart outside a QC point to open a local popup menu and select the item
2
1
<Table of QC>.
The table of QC vectors and the background charts are synchronised. If you click a row in the table, the relevant column of QC points in the charts window behind is highlighted and vice versa.
Refer to section 4.1 for the significance of the coloured deviation icons. The background colour of a row is red whenever the relevant QC vector is globally out-of-control. Note that the colours depend on the status at the time of the validation. If the control interval is changed later, a QC point may be drawn between the new control limits but stay flagged out-of-control in the table of data.
From the table of QC values it is possible to export QC data to the clipboard, to Microsoft Excel (if installed on the computer) or to a file.
There are 3 ways to open the table of EQA reports.
For any EQA flag on the charts, move the mouse cursor over the point until it is highlighted in yellow with a visible popup hint and then left-click it.
Left-click the main menu <Current analyte€Table of EQA> (Shortcut Ctrl + F3).
Right-click anywhere in a chart outside a QC point to open a local popup menu and select the item
<Table of EQA>.
This table works like the table of QC vectors. It is synchronised with the EQA flags of the background charts.
The character % in the columns Bias or Specif. means that the figures are relative figures. When a unit name or nothing is displayed the figures are absolute figures.
The unit name % is never displayed (e.g. for the hematocrit ratio) to prevent any mix-up with relative data.
1 2
5
3
4
MultiQC can print its tables. A preview always comes before printing.
1- Start printing.
2- Printer set-up.
3- Previous/next and first/last pages. 4- Change the scale of the preview. 5- Customise printouts.
Tables across analytes
QC data are most often managed analyte by analyte. This is the reason why MultiQC stores each analyte in a separate file. It is however sometimes useful to scan all these files to get an across analytes view of data. The across analytes functions are accessed by means of two items of the main menu <Daily data> and <Across analytes>.
This list is useful for the QC supervisor who can easily consult a daily summary of all the QC assays performed on a given day in each section of the laboratory.
Click the main menu <Daily data -> QC> to open the Daily QC dialog.
The displayed section can be changed by means of a drop-down list.
The current date is selected when the window is opened. It can be changed to any other day.
Click the button <Apply> to activate any change of date which is thus taken into account in the list and in the status bar at the bottom of the window.
Jump to the day after of to the day before with the arrow-buttons.
Two radio-buttons are provided to sort the table by analyte or by time. This sorting goes for both displayed and printed lists.
2
3
4
5
1
Refer to section 4.1 for the significance of coloured deviation icons. The background colour of a row is red whenever the relevant QC vector is globally out-of-control.
The main window with its QC plots remains active in the background of the daily QC window. Clicking a row of the daily QC list, or selecting it with the arrow keys ( ), immediately updates the background plots to the relevant analyte and highlights the QC vector of the chosen day.
You can print a separate report for each section of the laboratory by clicking the menu <Print € Displayed sections>. It is also possible to print a cumulative report for all the sections by clicking the menu <Print € All sections>.
Click the main menu <Daily data -> Calibrations> to open the Daily calibrations dialog. It displays and print a summary of all the calibrations and reagent blanks of the day. Reagent blanks that are followed by a calibration within less than 30 minutes are not taken into account.
The Daily calibrations dialog works as the Daily QC dialog above.
A global diagnostic of analytical quality can be easily put forward for each section of the laboratory. Click the main menu <Across analytes € Quality diagnostic> to show a list of all the analytes with the relevant capability indexes estimated for each QC material.
8
1
3
4
6
7
5
2
Unfold the drop-down list box to access another section.
Change the sorting order of the list. When Cp or Cpk is selected, the first analyte is the analyte with the lowest value of Cp or Cpk in anyone of the control materials. Thus the worst analytical methods are flagged down at the top of the list.
The date interval for the quality diagnostic can be defined by two extreme dates. Do not forget to press the button <Apply> to take into account the entered dates.
Fixed date intervals are often more practical. The buttons <Days back> (30 or 60) create an interval of dates back from the current date.
The current section and the date interval are displayed in the status bar.
Refer to section 8.4 for the significance of the performance icons.
Checking a box before the name of an analyte opens the Capability window for this analyte with the same time interval as the Quality diagnostic window (refer to section 10.3).
The menu <Print> can print the quality diagnostic of the current section or of all the sections.
Cumulative EQA data are most useful if you wish to access them date by date.
1
3
4
2
3
1
Data can be sorted by analyte or by date by means of the main menu <Group> or the local menu which pops up when the table is right-clicked.
Each node can be expanded/collapsed by clicking its sign +/-.
All the nodes can be simultaneously expanded/contracted by means of the main menu <Nodes> or the popup menu.
By default, the interval of dates includes the two last full months and the current month, but this can be changed to any range of months or dates
6
2
4
5
1
3
Statistics for a given period
1- Four parameters (m, SD, CV and N) are displayed. Anyone can be hidden by unchecking the relevant box.
The range of dates can be set to the latest 30 or 60 days (2), any month (3), any range of days (5) or the whole set of QC data (4).
6- Displayed statistics can be printed or exported to another program.
Additional submenus of the main menu <Across analytes> show and print summary reports for all of the analytes in a section:
Control limits
Parameters of analytes
Tolerances
Control materials
Reagent lots
Special management of data
Editing or deleting a QC vector
It is possible to edit or erase a misentered QC vector after its validation. Changing a validated QC vector requires the rights of a supervisor.
Important: Validation is always connected with the present reference pool of QC vectors. When re- validating a passed QC vector, the Hotelling’s T2 and the relevant control status may change if the reference pool or the control parameters were changed .
Two ways are provided to edit of a QC vector:
Editing from the charts
1
2
3
4
Right-click a QC point in a Shewhart’s chart or a T chart to pop up its local menu.
<Edit comment> to edit only the comment without having to re-validate the QC vector.
<Edit QC vector> removes the clicked vector from the sequence of QC points and displays it in the QC entry dialog where you can edit and re-validate it.
<Erase QC vector> erases all QC levels in the clicked column.
<Enqueue QC vector> removes the clicked vector from the sequence of QC points and sends it to the pending queue of MultiQC. The left panel “QC” in the status bar begins to blink red/yellow to inform of the QC vector waiting to be re- validated.
Editing from the table of QC vectors
Open the window Table of QC vectors as shown in section 9.1.
To erase one or several QC vectors, select them in the table and
Click the menu
<Edit€Delete>
Or right click inside the table to access a local popup menu with the same item
Or press the key <Del> of the keyboard.
To edit a QC vector, use the main menu, the popup menu or double-
click the line containing an erroneous datum. Re-validation works as indicated above. The QC vector to be re-validated is removed from the charts and inserted in the entry dialog as if it were a new entry.
One or several QC vectors can be resent to the pending queue of MultiQC by the menu item <Enqueue selection>
Editing or deleting EQA results or events
It is easy to edit or delete a misentered EQA result. This requires the access rights of a supervisor. It works like editing QC points.
1
2
3
It is sometimes necessary to simultaneously erase all of the data entered on a given time to allow an easy correction of misentries which concern several analytes. MultiQC can scan the analyte files to search for the items to delete. Without this function, it would be very tedious to have to open every analyte one by one to erase the wrong items. Erasing across analytes is only accessible to supervisors.
Open the window Delete in all analytes by clicking the main menu <Across analytes € Erase across analytes>.
Left-click the graphical item to erase across (IQC point, EQA flag or Instrument event icon).
The foreground dialog is immediately updated with the kind of data to erase and the date/time of the wrong items.
Press the button <Delete> to delete all the items of the same kind and date/time as the clicked, highlighted item.
Erasing across may require up to 2 steps
A confirmation is first requested before irreversible erasure.
When scanning the different analytes, MultiQC can meet a busy analyte (i.e. an analyte currently processed by another workstation of the network). In such a case the erasure action is postponed and enqueued to be performed later.
Initialising the EWMA and the EWMV
Mean of the first 5 points
ting value
EWMA star
The EWMA associated to the first QC point in a Shewhart’s chart is somewhat arbitrary. The iterative formula fails for this first point because there is obviously no point before. A possible starting value might be the process target but this target is unknown at this time. MultiQC uses an algorithm that revises the EWMA starting value as QC progresses.
The first EWMA value is
The first QC value when only one point has been entered
The mean of the 2 first QC values when 2 points have been entered
…….
The mean of the 5 first QC values when 5 points or more have been entered.
The EWMV is initialised likewise. The EWMV bars are displayed only after 5 different QC points have been entered to cut non-significant signals that might be triggered when a chart is started,
The power of EWMA comes from its ability to accumulate small lasting deviations to produce an early out-of-control signal when the analytical method begins to drift. This essential property relies upon the inertia of the EWMA curve that smoothes random variations to only show significant tendencies.
This inertia of EWMA is however a drawback when a wilful and sudden shift is forced on the analytical process. Such shifts often occur in QC charts after routine changes or adjustments of analytical conditions (calibration, new lot, maintenance …). Due to its inertia, the EWMA will follow the abrupt analytical step with a too long delay.
It is therefore necessary to manually re-initialise the EWMA to create a gap in the plotted line at the date of the analytical change thus showing the intentional discontinuity in the analytical process.
1
2
Calibration
The EWMA line follows the shift caused by a calibration with a time lag because of its inherent inertia.
A false increase of imprecision is flagged by the EWMV bars.
The EWMA must be re-initialised when a known external event has suddenly shifted the output of the analytical process. This can be done :
When entering and validating the first QC point after the shifting event
Retrospectively: Right-click the first QC point on the charts after the shift to open a local popup menu and left-click the submenu <Initialise EWMA>. Left-clicking again this submenu rolls back the re-initialisation.
3
EWMA initialisation
The re-initialisation of the EWMA appears on the charts as a “broken” style of the EWMA red line.
The EWMV is not re-initialised but calculated related to a new shifted EWMA. There are no longer alarms of imprecision in the EWMV bars.
A frequent way to back up data is to use a specialised network hard drive but a simple USB stick can be more practical. Each analyte needs a few tens of KB. The size of each analyte file can reach 100 KB after one or two years of daily QC without restarting a new batch of control material. Hence, each backup for the whole set of analytes can need a disk space up to a few megabytes.
1
2
3
4
Click the main menu <Data security € Backup> to open the dialog Backup data. It is necessary to carry out a short configuration before the first backup :
Click the tab <Configure backup>
Browse to the wished backup folder. You can only select an existing folder of the local disk, of a remote disk, or of an external storing device. It is sometimes useful to create a sub-folder. Its name must be entered with the keyboard in the entry field.
A flashing reminder is displayed in the status bar of the charts window when a given number of days elapsed since the previous backup. The default value is 7 days.
To prevent the backup folder from always increasing its size, it is recommended to activate an automatic purge which erases the oldest
5
6
backed up files, keeping only the three latest ones. This purge occurs every time a new backup has been successfully completed.
Now select now the tab <Perform backup>.
And press the button <Backup>.
Most often, restoring is requested because one analyte has been erased or corrupted by mistake. During the learning period of MultiQC, a good practice is to backup before any maintenance action to be able to reverse to the initial state if it is necessary.
If you want to selectively restore one or several current analytes, click the main menu
<Data security€Restore> to open the dialog Restore an analyte or a section.
1
2
3
4
5
6
7
The latest backup folder and the date of this backup are displayed.
You can browse to any other backup folder to reach a previous backup. It is also possible to browse the network files to the installation folder of MultiQC on another computer to copy any analyte from the distant computer to the local computer.
Two tabs are provided to restore either QC data or configuration data (8).
The backed up analytes are shown in a tree view. Select either a single analyte or a complete section.
Three boxes are provided to restore either the QC charts only, or the MAOP only or both simultaneously.
Press the button <Restore the selected item> or double-click any item (section or analyte) in the tree view to start a partial restoration.
8
The button <Restore all> starts a complete restoration of all analytes, ignoring the item selected in the tree view.
Transferring MultiQC to another computer
Launch the setup package on the destination computer to install MultiQC.
Locate the data folders of the source and of the destination computers. They depend on the version of Windows :
Windows XP data folder : C:\Documents and Settings\All Users\Application Data\MultiQC7
Windows 7 data folder : C:\ProgramData\MultiQC7
By default, Windows hides the folders C:\Documents and Settings\All Users\Application Data\ and C:\ProgramData. You must check the box <Tool menu -> Folder options -> View tab -> Show hidden files and folders> in the Windows Explorer.
Copy by any means the 3 following sub-directories from the source folder to the destination folder :
QCdata: current data and configuration data
QCarchi: archived data
QCusers: users’ settings.
Every backup creates a folder QC-yymmdd-hhmmss where yymmdd is the date of the backup and hhmmss is the time of the backup. You can find in these folders the three directories QCdata, QCarchi and QCusers that may be used as above.
Data restored from CDs are usually “read only”. Do not forget to uncheck the box “Read only” in the property dialog of the Windows Explorer because MultiQC needs to access its QC files in read/write mode.
Creating archives
It is also possible to manually archive any analyte or section.
Select the analyte or the section to archive.
Click the main menu <Configure€Analytes€Create, edit, archive> to open the dialog Definition of analytes.
Press one of the two buttons <Archive> in the frames Section or Analytes.
1
2
Consulting archives
To consult archived QC data check the box <Archived QC>.
Archived files are sorted in a 3-level tree view by
Section name
Analyte name
Date of the first QC vector. The number between brackets is the number of archived QC vectors).
3
Configuring archives
Many items of the main menu are hidden when the box Archived QC is checked. This is because archives theoretically must not be edited. However practice is different from theory and MultiQC provides a special menu for that purpose.
4
The menu <Unarchive> resends any archived file from Archived QC to Current QC. To prevent a mixture of archived data with current data, the section name becomes @Section with a leading @.
In the opposite picture, a new section @2700-Serum was created for the unarchived Albumin which can be edited, reprocessed … as any current analyte. Then the unarchived analyte can be re-archived.
Deleting and purging archived files
After several years of QC, archived files are accumulating in the folder QCarchi of the installation directory of the program. Archived analytes can be erased one by one with the menu <Configure € Delete archive>.
The menu <Configure € Purge archives> triggers an automatic purge of the oldest files.
Choose the age of the archives that need to be purged.
4
3
1
5
2
Press the button <Search> to scan the archived files and display the oldest ones.
The archives are listed whenever the last recorded QC vector is older than the entered age.
Lines can be individually selected / unselected with checkboxes.
The columns From and To show the dates of the first and of the last QC vector for a given file.
Selected files are sent to the Windows recycle-bin when they are erased. Thus they can be restored in case of error.
When the available volume of a reagent is not great enough for the anticipated analytical workload of the day, some analysers can manage a second bottle of the same reagent waiting on the tray. Bottles are automatically switched when the first one is exhausted to prevent from a break in the analytical throughput.
When the QC computer is connected to the analyser and directly receives bottle information, MultiQC displays the icon in its reagent event bar as soon as the second bottle is calibrated or QCed. However the older bottle may be not yet empty and may be QCed again after the new one has been started.
Assay with the new bottle
Assay with the old bottle
Old bottle of reagent
New bottle of reagent
MultiQC uses a special drawing style to show that a QC point has been assayed with the older reagent bottle: regular blue point surrounded by a thin red line.
Operators and logins
It is not necessary to log in for a simple consultation of QC data. Anybody can browse anonymously (under Noname) across the analytes and across the QC plots without wasting time entering a user-name and a password.
Anonymous entry and validation of QC data can be forbidden by checking the box <No anonymous entry> of the configuration dialog (refer to section 13.1). This box is unchecked after the installation of MultiQC and can be left unchecked during the trial period of the program. Later in routine laboratory use, anonymous entries should be forbidden to comply with traceability regulations.
To log in you must drop down the list box in the top-left corner of the main window and click your user name. Your password will be requested except if you log in under “Noname”.
There are two kinds of operators
Basic operators who can enter data but who are not allowed to edit data or to configure the program
Supervisors, who can enter data, edit data and configure MultiQC.As long as no supervisor has been declared, all of the functions of MultiQC are accessible to anyone. As soon as there is one supervisor, the configuration and edition tasks of multiQC are accessible to supervisor logins only.
Click the menu <Configure €Operators> to open the dialog List of operators. This menu is accessible only if you are a supervisor or if no supervisor has yet been declared in the list of operators.
Logins are created by supervisors with a provisional user-name and without password. Later the users will have to update their login themselves to keep their password hidden to supervisors.
Enter a new operator's name.
Check one of the boxes <Basic operator> or <Supervisor>.
Validate the new operator with the button <Create>. The new login is then inserted in the left list box and sorted in alphabetic order.
1
2
3
4
The new user-name is not yet password protected. On the first logon, a reminder will ask each operator to protect his user-name by a password.
To edit an operator, click his/her name in the list, change the name, the rights and do not forget to validate using the <Update> button, otherwise the modifications are not saved.
If all supervisors have forgotten their password, erase all the files *.usr in the data folder QCusers. This will reset the list of operators to the unique default operator Noname..
The place of the the folder QCusers depends on your version of Windows :
Windows XP : C:\Documents and Settings\All Users\Application Data\MultiQC7\QCusers
Windows 7 : C:\ProgramData\MultiQC7\QCUsers
Each user is compelled to enter a password when he logs in for the first time.
When logging in again later, the button <Change login> enables each user to change his user-name and his password without referring to a supervisor.
Miscellaneous configuration
Miscellaneous configurations dialog
Click the main menu <Configure € Miscellaneous> to open the dialog Miscellaneous configuration.
1
2
3
4
5
6
Tab : General
<Header on printed forms> :
Header that is printed on all the printed forms.
<No anonymous entry> : Box to check to prevent from data entry without logging on.
<Delay to self idle> : When MultiQC works in a network of computers, two PCs cannot simultaneously access the same analyte. Analyte files are not shared. To prevent a user from locking an analyte for too long, multiQC automatically returns to an idle status if no mouse movement or key press occurs
during a user-defined time period. By default this period is 5 minutes. Moreover, if a technician leaves his or her login open, it is automatically reset to Noname after the same time interval.
When MultiQC is connected to an analyser, QC results are sequentially received. For each analyte it is necessary to lump together the different QC levels in a unique QC vector. This is made on a time interval basis. QC levels for an analyte are associated in the same QC vector if the time interval between the assays is less that the limit entered in the field <Max time interval …>.
Two entry fields are provided to set the percent of tolerance that is allocated to non-linearity error and non-equivalence error. Refer to section 18.7 and 19.4.
Changes are taken into account when the dialog is closed. Before you can revert to the initial data with the button <Undo>
Tabs : Timetable Monitoring and Holidays
Refer to section 15.4 (Monitoring the time of QC assays).
Editing lists of pre-defined captions
Several lists of pre-defined captions are maintained by MultiQC.
Names of unit
QC vectors comments
EQA returns comments
Reagent remarks
Calibration comments
All these lists are edited in a common dialog which is shown above.
Press the button <Insert> or the key <Ins> to show an entry field at the end of the existing list. Leave the entry mode by clicking anywhere on the dialog of pressing the key <Escape>
Do not forget to press the button <OK> to validate the updated list before quitting the dialog.
Every operator can customise his charts. When he logs on, MultiQC shows the analyte that was displayed on the previous logging off with his own style of control charts.
3
1
2
The dialog Style of control charts is opened by the main menu <Maintenance€Charts style> or the same command in the local popup menu that opens when right-clicking anywhere on a chart.
Enter 0 in the field <One vertical every> if you want to hide the vertical lines.
<Analyte tree view> : The tree view to select an analyte can be displayed on the right or on the left hand of the charts. When charts are completely filled with QC points, the new points are added on the right of the screen and it is more practical to show the tree view beside the most recent QC points.
<Colour scale> : Uncheck the box to restrict the colours of the background to 3 discrete colours.
MultiQC is designed for a peer-to-peer architecture in which each workstation has equivalent capabilities and responsibilities. QC data are different from patients data. The latter data are best managed with a clients/server architecture, in which one computer is dedicated to serving the others, because patients data are of general concern for the whole laboratory.
Local installation
QC and EQA data are closely related to the analytical instrument they originated from. There must be a real-time interaction between analytical processes and the software devoted to control these processes. Therefore MultiQC must be installed on the local disk of a computer near the laboratory bench. This installation guarantees a permanent and instantaneous availability of the QC software which is as essential to an analytical process as its reagents.
Distant access
The laboratory supervisor is a minor off-line user of MultiQC. He may wish to access and review the data distributed on the QC computers of his laboratory without leaving his office. This is possible thanks to a series of shortcuts on his desktop. Each one is connected to a distant program file MultiQC7.exe.
1
Overwriting analyte files is impossible because these files are opened in Share-exclusive mode. When a workstation tries to display an analyte which is being processed by another one, the former one receives a warning message Busy. Thus, for instance, the QC supervisor cannot update the control parameters of an analyte from his office while a technician is validating QC results for this same analyte. The supervisor must wait till the analyte is freed to be allowed to access it. The self-idle function of MultiQC (section 13.1) prevents anybody from
2
indefinitely locking an analyte.
3
Some actions of MultiQC need an update of all the analytes (actions across analytes). If one analyte is busy at this time and thereby cannot be accessed, the action is postponed to be automatically tried again every 1 minute till it can be performed.
Some configuration actions cannot be done if several instances of MultiQC are running. A warning message is displayed.
4
Conversely, when a workstation is configuring MultiQC, other workstations cannot start the program as long as the configuration task is not completed
Setting up a distant access to MultiQC
The program and data folders of MultiQC must be declared as “shared” and network users must have the permission to modify the files in the data folder. The Windows documentation explains how to share sub- directories and apply permissions. The dialogs are not the same in Win XP, Win Vista and Win 7.
However the course is the same:
Laboratory local computer where MultiQC is installed (It is located in the laboratory, connected to one or several analysers)
Open the Windows explorer and browse to the program folder C:\Program files\MultiQC7.
Right-click this item.
Choose Properties in the local menu that opens;
In the following dialog depending on the version of Windows you must check the box <Share>, type in the share name <MultiQC7 program> and set the share-permissions to <Read>.
Browse to the data folder which depends on your version of Windows
Win XP = C:\Documents and Settings\All Users\Application Data\MultiQC7
Vista or Win 7 = C:\ProgramData\MultiQC7
By default, Windows hides the folders C:\Documents and Settings\All Users\Application Data\ and C:\ProgramData. You must check the box <Tool menu -> Folder options -> View tab -> Show hidden files and folders> in the Windows Explorer.
Right-click this item and proceed as above. Check the box <Share>, type in the compulsory share name <MultiQC7 data> and set the permissions to <Change> (read+write+delete).
Secondary computer (which is for instance in your office).
Open the Windows explorer and browse through the network files to find the laboratory local computer where MultiQC is installed. Then go on browsing to find the folder with the share name <MultiQC7 program> and then the distant executable file <MultiQC7.exe>.
Monitor of the secondary computer
Share name of the distant program folder
Distant QC computer
Create a short-cut on the desktop to this distant executable file.
Windows Explorer of the laboratory local computer
The registration key that is emailed after purchasing MultiQC is requested whenever a distant computer tries to launch the programme for the first time.
Control materials
Each control material is usually employed to control several analytes. This is the reason why the names and lots of control serums are entered first in a table and then assigned to analytes. Creation and assignment of control material are done from the main menu <Configure € Control materials>.
1
2
3
4
5
While control materials are created, the background charts remain active. The analyte can be changed, the charts can be scrolled, and the hints can be opened hovering points or events with the mouse.
Create the control materials filling in the fields of the left hand panel. When the button <Create> is pressed, the new control material is inserted in the list of the centre panel.
Assignments are made with the button <Assign> or by dragging and dropping from the centre panel to the desired QC level in the right panel. You can assign a control material either to a unique analyte or simultaneously to all of the analytes of a section. For instance in the picture above, the target analyte is the direct bilirubin of the section 640-Serum, but it should be possible to select the section “640-Serum” to simultaneously update all of the analytes it contains.
To edit the lot or the name of a control material you must
select it in the centre panel
Change the name or the lot in the left panel
press the button <Update>. This action updates the list of the centre panel but also all the analytes which are utilising the edited control material.
Do not mix up editing a control material with changing to a new batch. This latter action is described in the next section .
The button <Material utilised by> shows which analytes are controlled by the material selected in the list above the button.
The button <List of materials> displays the list of all the analytes with the relevant control materials. Printing this table is also possible.
Starting a new batch of control material
A good laboratory practice is to start a new batch of control materials before the current one is exhausted. So, during a few days or weeks, it is necessary to separately collect QC data for the new batch while keeping on controlling analytes with the older batch. Two sets of QC charts must temporarily coexist for the same analyte.
An e-learning how-to tutorial for starting a new batch of control materials is available at www.multiqc.com.
Cloned analytes
MultiQC has a built-in easy way to manage this duplicated QC by duplicating the analytes. It is designed to apply to the normal case when all the levels of control materials are simultaneously changed. The QC status of a method on a given time is testified by the association of the concentrations of all the materials gathered into a non-dissociable QC vector. It would be therefore a great error to separately archive one QC level alone because this would cut the traceability of passed QC.
The clone of an analyte is a copy of this analyte (father) which is used to temporarily store the QC data relevant to the new batch of control materials during the time necessary to learn the new control parameters. When time occurs to switch to the new batch, the older charts are archived and the clone takes the place of the regular analyte.
A cloned analyte :
Keeps the features of its father-analyte.
Has the name of its father-analyte with a leading ~ (~Analyte).
Is freed from all QC points and EQA flags
Is set to semi-learning mode (see section 7.7)
can be accessed for entering or editing QC vectors in the same way as any other regular analyte.
Cloned analytes should not be used when the control mode is Specified interval : there is no need to estimate the new control parameters because they are provided by the maker of the control materials. You have only to archive the older data and update the specified interval.
Redirection of QC data to ~Analytes
A mechanism is available in MultiQC to redirect entry profiles and connexions from analysers to
2
1
~Analytes. This redirection must be used every time when you are assaying the new batch of control materials. It tells MultiQC that the next QC data to be keyed in by means of a profile or to be received from the connected analyser must be sent to clones instead of father- analytes.
Click the main menu <Data entry € Entry to ~Analytes>. As soon as the button <OK> of the confirmation dialog is pressed, all of the QC results are re-directed to the cloned analytes.
The redirection will be automatically ended one or two hours later. Be careful if you check the box <Never>. Practice in the author’s lab has shown that technicians often forget to manually stop redirection to ~Analytes. This omission may lead to a mixture of batches the day later when the current batch of QC material is assayed again.
3
When the redirection to cloned analytes is active, a new checkbox is displayed in the top-left corner of the main window. At any time, you can manually uncheck it to end the redirection.
Redirection of data entries concerns
Manual keyboarding of IQC profiles
QC data loaded from a text file.
QC data received on-line by means of a serial QC receiver interface (refer to section 24). The redirection must be activated before the transmission of concentrations by the analyser.
IQC data sent to MultiQC working as an automation server
Automatic copying of analytical events and EQA returns to clones
When two batches of control materials are coexisting thanks to cloned analytes, EQA flags and analytical events apply to both Analytes and ~Analytes. Therefore new entries and changes in Analyte are automatically copied by MultiQC to ~Analyte.
To prevent any duplication, it is impossible to edit EQA returns and events in cloned analytes. You can only edit the father-analyte.
Exception : For obvious reasons, the events Parameter-change are independently managed in Analytes and ~Analytes.
Hint : EQA returns can be inserted in the provisional ~Analyte if at least one IQC vector of ~Analyte is dated earlier than the EQA result. Remember that EQA flags are drawn in comparison with the moving average and thereby cannot be drawn if this average is not available at the time of the EQA result.
Creating cloned analytes
In normal use of MultiQC, analytes never need to be manually cloned. You have only to activate the redirection to ~Analytes before filling in an entry profile or receiving data from an analyser. Cloned analytes are automatically created if they do not exist.
1
2
3
However it my be sometimes useful to manually create the clone of a given analyte.
To clone an analyte, simply select this analyte in the tree view and click the menu <Configure €
Analytes € Create, edit, delete, archive> to open the dialog Definition of analytes.
Press the button <Clone> in the analyte frame.
The name of the new ~Analyte is shown in the same section as the father-analyte.
1
Updating the lots of the new control materials
Click the main menu <Configure € Control materials> to open the dialog Administration of control materials (See section 14.1).
Create the new control materials in the left panel. Be careful not to edit the older batch. It is yet in use and must not be changed as long as it is not exhausted.
A special box is displayed in the right frame Assign to when any ~Analyte is selected. Checking it will save you from having to individually update all of the clones. Assigning the new batch of control materials to one is extended to all of them.
Switching to the new batch of control materials
When the older batch is exhausted, it is time to unclone the cloned analytes. This operation
Archives the data of the father-analytes
Erases the father analytes
Rename the ~Analytes as Analytes
Click the menu <Configure € Analytes € Create, edit, delete, archive> to open the dialog Definition of analytes.
Two buttons <Unclone> are available: The left one in the frame Section is the most useful. It simultaneously unclones all of the ~Analytes in the selected section.
Monitoring the time of QC assays
MultiQC can monitor whether QC assays have been performed in compliance with a timetable separately defined for each analyte. Every hour, the programme scans its files looking for QC assays that were missed. If a too long delay is found, the programme opens a warning window which displays a list of the missing tests and sends out a customisable music.
Activating the monitoring of QC time
1
2
4
3
Open the configuration window with the menu <Configure € Miscellaneous>
Select the tab <Timetable monitoring> .
Tick the box <Activate QC time monitoring>.
Press the <Play> button to listen to the default music.
If you dislike the default music, you can browse to any other
<*.wav> file of the disk.
When the QC monitoring is activated, a reminder icon is displayed in the status bar at the bottom of the main window
5
The current analyte is under QC time monitoring 
No timetable was entered for the current analyte
Setting up the QC timetable of each analyte
1
2
Select an analyte in the left tree view of the main window.
Click the main menu <Configure € Analytes € Timetable> to open the dialog Timetable of QC assays.
Check the week-day boxes on the left to select the days when the program must verify whether the timetable has been adhered to.
The 24-hours clock on the right is provided to define the deadlines for QC assays. Click one or several hour-buttons to enlighten them and thereby create the QC timetable.
It is sometimes necessary to enter a different timetable for Saturdays and Sundays. Ticking the box <Different
…> adds three tabs to the timetable window for separately programming week days, Saturdays and Sundays.
Entering the same timetable for all of the tests performed by the same analyser would be very tedious. It is possible to select a section in the left tree view of the main window to simultaneously set up the timetables of all its analytes.
The width is rounded to the lower quarter of an hour The width is limited to a maximum of 3 hours.
4
Width of the allowed time intervals shortest time between successive QC deadlines
QC assays must be performed not too early before the deadlines. Assays must fall within allowed intervals ended by the deadlines. These time intervals are displayed as green pieces of pie on the QC timetables. Their widths depends on the frequency of QC assays.
For instance, the timetable of the above picture requires two QC assays per day, the first one between 8 AM and 10 AM and the second one between 5 PM and 7 PM.
The menu <Configure € Analytes € Timetable> is allowed only to supervisors. Basic operators can however consult timetables by means of the menu <Current analyte € Timetable> with no editing right of click the status bar icon
.
1
3
2
On the hour, 24 times per day, MultiQC automatically scans all of the QC data to search for missed assays. Whenever the program finds something late, it opens the Missed QC dialog.
The column <QC delay> displays the time from the missed deadline.
If you close the dialog with the button
<Exit> the same warning will appear one hour later unless the relevant QC assays are performed.
Check the boxes in the column
<Cancel>, to cancel the forgotten deadlines and stop any further warning.
The Missed QC window can be also manually open at any time with the main menu <Across analytes €
Missed QC>.
It is possible to halt QC monitoring during holidays.
Click the main menu <Configure€ Miscellaneous€ Holidays> to open the Holidays dialog.
This dialog is self-explanatory.
QC monitoring is halted on holidays. The next QC deadline is then shifted to the next non-holiday day.
Twinned analysers
Analysers in medical laboratories are often doubled either to increase the analytical throughput or to guarantee the customer a constant turnaround time in spite of unavoidable breakdowns or halts for maintenance. Using two different instruments for the same test means that successive samples from a given patient may be randomly assayed on anyone of the two instruments. For example, this configuration of twinned analysers is implemented in the author’s laboratory for
Blood gas
Emergency immunochemistry (chiefly cardiac markers)
Basic chemistry
Issues raised by twinned analysers
The obvious drawback of twinned analysers is the bias that is likely to occur between them. The calibration process of the presently marketed clinical chemistry analysers is very far from a genuine metrological procedure. Calibration assays are generally repeated only twice which leads to a significant uncertainty of set points.
In routine work, twinned analysers are independently calibrated. Each one is therefore affected by a random and independent error of its set point. So we cannot rely upon calibration to equalize them. A variable bias between the two instruments is unforgiving. The discrepancy may range within an interval of [ 3 SD ] (where SD is the standard deviation of each instrument). This potential error is unacceptable for low capability analytical methods
Managing twinned analysers requires not only a classical separate QC but also a continuous monitoring of the bias between instruments. Any significant deviation should be corrected as soon as possible by adjusting one set point or both. Three inappropriate practices must be discarded :
Setting up as twinned analysers two instruments of different brands, that are not using the same reagents and the same calibrators. Some clinical chemists are wrongly thinking that they can rely upon a slope and an intercept resulting from a method comparison to cancel out the bias between their analysers. This computation surely cancels out the average bias between instruments but not the individual difference of biases for every patient.
Planning weekly or monthly reviews of the means of control material to decide whether the twinned analysers were equalized or not. This retrospective control is much too late.
Comparing the Levey-Jennings charts of the twinned tests: This comparison is not very informative even when both QC plots are painted on the same screen. QC points are too much scattered on the plots to provide an efficient visual indicator of the difference between set points.
MultiQC7 proposes an alternative way to monitor the bias between twinned analysers : The paired exponentially weighted moving averages (PEWMA) of control materials : At first, twinned analytes must be controlled with the same materials. Then the two moving averages are simultaneously plotted on the same chart. They are paired on the basis of the same date.
The exponentially weighted moving average (EWMA) is a cumulative score that weights the earlier observations successively less than subsequent observations in such a way as to automatically phase out distant observations almost entirely. The EWMA is both
a statistical process-monitoring tool: It detects the presence of assignable causes that result in a process shift (bias).
a forecast of where the process will be at the next time period. An estimate of the bias of a method is given by the difference between the EWMA and the target. This estimate can be used as the basis for a dynamic process-control algorithm to determine how much adjustment is necessary.
For every analyte MultiQC draws the EWMA line (red in the picture below) superimposed over the regular Levey-Jennings chart. When an analyte is twinned with another one, a second EWMA line (light blue below) is added to the main chart to provide the clinical chemist with a simultaneous and real time display of both individual biases ( the distance of each EWMA line to the target line) and of the bias between tests (distance between the two EWMA lines).
K+ assay on twinned analysers
Main analyser Twin analyser
QC points EWMA
EQA target flagsEWMA
Control limits of the EWMA
Remarks
The QC assays for an analyte are not necessarily done at the same time as those of its twin. Furthermore an analyte may be controlled twice or thrice whilst its twin is controlled once or vice versa. Time scales do not coincide. To be able to pair the two EWMA lines, MultiQC must therefore locally expand/shrink the time scale of the twin before inserting it in the time scale of the main analyte
The plot of paired moving averages only shows the QC points of the main analyte. The QC points of the twin analyte are hidden not to cram the chart with useless marks.
The smoothing factor of the main analyte is applied to both EWMA lines, even if the twin analyte has a different factor. Thus the red and the blue lines are comparable because equally smoothed
Target value
The paired moving averages are aimed at monitoring the bias of twinned analytes and at suggesting calibration adjustments as soon as the gap becomes too high. In the picture below, the bias reaches about 2 SD (grey double arrow). Such a discrepancy is too big for a low capability method. It should trigger a corrective action. The twin analyser (blue curve) is pretty much stable whilst the main one (red curve) has drifted of about +1.5 SD from its target value.
Bias of the main analyser
Bias of the twin analyser
Bias between analysersThe EWMA of the main analyser (red line) has drifted out of the control interval whereas the twin analyser (blue line) is more stable. Equalizing the twinned analysers may be necessary if the capability of the analytical method is low.
Because of the uncertainty of the calibration process mentioned above, it would be fruitless to try to reset the main analyser to its genuine set point thanks to another calibration. This might equally either improve or worsen the situation. Presently marketed analysers are unable to guarantee calibration biases less than
2.0 SD (in good conditions).
The only way to equalize the main analyser with its twin is to manually tune the calibration factor in an engineering process control fashion. The drift of the EWMA provides us with the precise value of the required adjustment. Unfortunately, direct access of calibration factors is rarely available on today’s medical laboratories instruments. Manufacturers are more and more forbidding what they call “fudge corrections” hoping to guarantee a floor level of (poor) analytical quality with unskilled operators. If you are a skilled and perfectionist clinical chemist you have to purchase another instrument that allows feedback adjustments of calibration factors.
Software of clinical chemistry analysers often have entry fields named Slope and Intercept. These coefficients cannot help us because they are intended for a definitive change of analytical methods. They are used for adjustment factors that create permanent shifts. The best example is the compensation of creatinine Jaffe assay to match the ID- MS method.
What we need is a direct access to the calibration factors to be able to slightly adjust them when the EWMA shows that a feedback action is necessary. This adjustment is provisional. It must disappear with the next calibration.
Twinning / Untwinning analytes
Click the menu <Configure -> Analytes -> Twins> to open the dialog Define twin analytes. You can browse to any analyte or section :
By clicking the top-left spin button of the dialog (1)
By clicking the tree view of analytes in the main window that is always active in the background.
1
2
3
Twinning two analytes
Browse to anyone of the two analytes to twin (1)
Then select the section (2) and the name (3) of the twin analyte
Validate with the button Apply. You can go back to the previous state with the button Undo or cancel the twinning with the button Untwin.
1
4
Twinning two sections
It is possible to quickly twin all of the analytes of a section to the analytes with the same name in another section
Browse to the section to twin (1)
Select the name of the twin section (4)
Validate with the button Apply
A list of the successfully twinned analytes is displayed at the end of the operation.
5
When an analyte is twinned, the name of its twin is displayed at the bottom of the treeview in the main window of MultiQC.
The button Switch (5) is then enabled to allow an easy switching to the twin of any analyte and vice versa (shortcut Ctrl + w).
This is very useful to compare either the QC charts or the MAOP curves.
Verifying the repeatability of a method
Verifying repeatability is particularly quick and easy when the QC computer is connected to the analyser.
Refer to the user manual of the QC receiver interface to learn how to define the reserved identifier thanks to which MultiQC recognizes the repeatability assays among the flow of patient assays.
MultiQC regards repeatability assays as completed when a 3 minutes lag without new assay has elapsed.
A reminder Rep flashes red/yellow at the bottom-left of the main window to warn that repeatability data are ready.
6
Tolerance
5
Uncertainty interval
1
4
2
3
Click this reminder to open the Repeatability dialog.
The histogram is drawn in the same way as in the capability window (section 8.4).
A table displays the statistics of the distribution.
The grid of data can be edited if you press the button <Edit>.
An outlier can be withdrawn from the computation of statistics. Switch to the Edit mode and check the box besides the concentration to disable. The disabled concentrations are greyed in the histogram.
You can browse to the previous and next verifications of repeatability with the buttons and .
Repeatability data can be typed in manually by means of the tab <New repeatability>.
The capability index Cp which is computed in the repeatability window must not be taken as the actual operational capability of the analytical method. Cp is computed in repeatability conditions i.e. within the same run. Long term Cp is often 1.5 to 2.0 times smaller.
Every newly saved repeatability verification creates an event Repeatability
which is logged by MultiQC like all the analytical events.
Repeatability events can be recalled by :
Clicking the icons
which are drawn in theReagent event bar.
Clicking the main menu <Events -> Repeatability> (shortcut F6).
Verifying the reportable range
Establishing and verifying the reportable range is an integral part of quality control. Therefore, MultiQC includes a linearity module for recurrently monitoring the non-linearity error of all the analytes which it controls. MultiQC does not use the protocol EP6-A by the CLSI / NCCLS.
An e-learning how-to tutorial for the linearity module of MultiQC is available at www.multiqc.com.
What is not appropriate in the EP6-A protocol ?
The EP6-A CLSI protocol is artificial because it is not based on the very definition of the reportable range. The error of its editors was to think primarily in terms of linearity instead of terms of allowed error. Straight line or not straight line ? That is not the question. The true question is to decide whether the departure of the actual response curve of an analytical method from the ideal straight line is acceptable or not. Stepwise polynomial regression is recommended by the EP6-A protocol. It is an efficient statistical tool to verify linearity but verifying linearity is unimportant in a clinical laboratory. What we do need is to establish a reportable range for a given shape of response curve and for the allowed tolerance of the analyte.
Response curve and tolerance area
The response curve of an analytical method is a plot of the measured concentration as a function of the true analyte concentration. Ideally, the measured and the true concentrations should be equal whichever these concentrations might be. Thus the ideal response curve is the bisecting line in a plot of measured- versus-true concentrations.
Absolute tolerance
Relative tolerance
True concentrations
Reportable range
Non conforming
Ideal response line
Actual response curve
Non-linearity error
Tolerance area
Point beyond which non- linearity error exceeds tolerance
Measured concentrations
A significant departure from this perfect agreement is nevertheless acceptable because of the error tolerated for the analyte. The vertical distance between the ideal and the actual response curves is named non-linearity error. It must not exceed the tolerance of the method. This tolerance can be expressed as an absolute or as a relative acceptable error. Most often in clinical chemistry, both are associated so that the absolute error applies to the lower concentrations and the relative error applies to the higher concentrations. Graphically, the tolerance intervals for each concentration are merged into a polygonal area framing the ideal bisecting line. The edges of this area are parallel for an absolute tolerance. The edges are diverging for a relative tolerance.
A curved response line can partially meet the tolerance of an analytical method provided that the interval of measured concentrations is adequately limited. The reportable range is the range of concentrations within which the non-linearity error is smaller than the tolerance. Analytical results inside this range are conforming and can be reported. The other ones must be discarded and the sample reprocessed.
The reportable range of an analytical method is established searching for the segment(s) of the actual response curve which is/are interior to the tolerance area and then projecting it/them onto the ordinate axis. Why not project onto the X axis? Because of the practical use of the reportable range. Technicians will compare each concentration measured by their instruments to the reportable range to decide whether the assayed values are conforming or not. It is the reason why the reportable range must be also expressed in terms of measured concentrations (ordinates) and not in terms of true concentrations (abscissas).
Practically we have to search for the intersections of the response curve of the method with the top and bottom edges of the tolerance area. This area may be a rather complex polygon when setting different tolerance values for low, mid and high concentrations of the analyte as it is possible in MultiQC. For a given tolerance polygon, the number of intersection points depends both upon the shape of the response curve and upon its position relative to the axes of the plot. The former is linked to the principle of the analytical method. The latter is linked to the calibration of the method. Our first step will focus only on the shape.
Determining the shape of the actual response curve
Obtain 5 or 9 levels of concentration over a range that is a bit wider than the anticipated reportable range and with equally spaced concentrations. See below how to make intermediate dilutions of two pools by means of sequential mixing. The exact concentration of each linearity material can be ignored but the dilution ratios between successive samples must be very accurate. Run one, two or three replicate samples according to the degree of precision that you need for the response curve that is going to be estimated.
Open the linearity window by clicking the main menu <Events € Linearity> (Shortcut F7).
Select the tab <New range verif>.
Set the number of <dilutions> to the desired value.
Enter the assayed concentrations in the grid.
Validate with the button <Apply>.
MultiQC searches its library of mathematical functions for the one which best fits the experimental points. This function is adopted as the estimated response curve of the analytical method and the relevant curve is drawn. At this step, we have found a shape but the response curve remains uncompleted because the X axis is yet graduated with the number of the dilutions instead of the true concentrations.
1
2
3
4
A faulty calibration might shift the response curve out of the tolerance area and reduce the reportable range. But this is another problem. We must pull apart non-linearity error and inaccuracy supposing that the method was precisely calibrated before assaying the set of samples with equally spaced concentration.
2-point calibration
Let us assume that the instrument is in-control and that it was calibrated in 2 points: 0 g/l and 2.34 g/l. This means that, by definition and ignoring the uncertainty of calibration, the actual response curve is true for these two concentrations 0 and 2.34 g/l. Hence we know two points of the ideal straight response line and consequently can draw it on the plot.
Move the mouse cursor over the plot and drop two calibration marks (blue crosses) at the ordinates 0 and
As soon as the two calibration points are set, MultiQC updates the plot:
It draws the ideal straight response line that crosses the two calibration marks.
It draws the tolerance area framing the ideal response line with tolerance intervals.
It graduates the X axis with a scale of assigned values replacing the scale of dilutions.
It searches for the segment of the actual response curve interior to the tolerance area and project it onto the Y-axis to graphically show the reportable range as a green background behind the axe.
It displays the numerical value of the reportable range in the bottom status bar of the Linearity window.
Reportable range
Calibration points
Moving the calibration marks
It is very easy to move the calibration marks to test if a better choice of the calibration concentrations would improve the reportable range.
Hover the calibration mark to move with the mouse cursor to enlighten it in yellow.
Left-click the enlightened yellow cross. It turns red.
Keeping the left button of the mouse pressed, drag the calibration mark along the response curve and drop it where you wish by releasing the mouse button.
The tolerance strip and the frame Calibration interval are simultaneously updated.
Multi-point calibration
Let us now assume that the instrument was calibrated in 6 points. This would mean, as above, that the actual response curve is true for these six concentrations. Practically because of the calibration uncertainty these six points are never perfectly aligned. In this case you must check the box <Multi point calibration> and drop two calibration marks on the curve at the two ends of the calibration interval. Then MultiQC calculates the ideal response line as the regression line of all the points of the response curve between the two calibration marks.
The response curve is drawn blue between the two calibration marks to remind that the position of the ideal response line is based on the whole interval and not only on its two ends.
Reportable range
Ends of the calibration interval
Linear response curve
If the best response curve estimated by MultiQC is a straight line, there is no need to enter calibrations marks. The reportable range equals the tested range.
Click the tab <History> and select the item to edit in the drop down pick list.
Click the button <Edit> which is enabled only if you logged on as supervisor or if there are no supervisor.
Now you can change anything in the entry fields or move the calibration marks with the mouse. The button <Apply> recalculates the new response curve and the new tolerance strip.
Do not forget to press the button <Save> when everything is OK in the edited response curve.
Tolerance
The item <Tolerance> of the main menu opens the tolerance dialog of MultiQC (see section 8.2).
When a new range verification is being typed in, the displayed tolerance is the current tolerance.
When going back in the history of reportable range verifications, the dialog shows the tolerance at the time when the assays were performed. This tolerance may be different form the current tolerance.
1 2
3
Print
You can get a printed report of the reportable range verification by means of the menu <Print>. A preview is displayed before printing.
Start printing
Printer setup
Customize printing format.
Each user of MultiQC has his own printing style.
Style
With this dialog you can customize the on-screen style of the linearity plots.
Each user has his screen display style
Tolerance for non-linearity error
Tolerance for non-linearity error is based on the overall tolerance of each analytical method which is recorded in MultiQC to create acceptance charts and to calculate the capability indexes. Non-linearity is a component of total error, but not the only component. So tolerance for non-linearity error must be smaller than the overall tolerance to take into account the other causes of error (imprecision, bias, interferences…). MultiQC makes use of a reduction factor named Non-linearity error budget. Its default value is 50%. This means that if the overall tolerance for serum glucose is 4%, the reportable range will be the range where non-linearity error does not exceed 2%. The actual value of this Non-linearity error budget is shown in the bottom status bar of the Linearity window. To change it, click the menu
<Configure € Miscellaneous € Tab Error budgets>.
MultiQC can associate a relative and an absolute tolerance in three different intervals. This may lead to complex tolerance areas and discontinuous reportable ranges.
Preparation of samples with equally spaced concentrations
The most precise way to mix low and high concentrations pools to produce samples with equally spaced intermediate concentrations is sequential mixing. A middle pool is obtained by mixing equal volumes of the low and high pools. Then the middle pool is mixed with the high and the low pool, in equal volumes, to produce a mid-high and a mid-low pool. Thus a set of five equally spaced concentrations is made up. A set of nine equally spaced concentrations can be easily prepared mixing again the adjoining samples of the set of five concentrations.
The volumes in the above dilution scheme should not be reduced to keep a good precision. Conversely, it is highly recommended to work with higher volumes if enough pool is available. Increasing the volumes is also required for the preparation of a set of 9 equally spaced concentrations which requires a third dilution step.
Low pool 500 µl
High pool 500 µl
0
2
4
.
200 µl
200 µl
100 µl
100 µl
100 µl
100 µl
200 µl of five equally spaced dilutions
0
1
2
3
4
Sequential mixing is very fast, accurate and precise even with volumes as small as 100 µl. Errors might come from the nature of pools materials which generally have a high viscosity and a tendency to foam easily. These issues are overcome by using the reverse pipetting technique to dispense materials and by a careful mixing of tubes. A bias in the calibration of pipettes is not harmful because of the principle of mixing equal volumes from the same pipette.
Saline or even water can be taken as low pool with a nil concentration in many cases. The objection of matrix effect seems to be largely overemphasized. A high pool should be easily found among the highest daily samples of the laboratory that need to be re-processed after having been diluted. Considering the accuracy of in-house sequential mixing by trained operators, purchasing commercial linearity verifiers often appears as a waste of money which is not balanced by more reliable samples.
Every new saved verification of the reportable range creates an event Reportable range which is logged by MultiQC like all the analytical events.
Reportable range events can be recalled by :
Icons
which are drawn in the Reagent event bar.The main menu <Events -> Linearity> (shortcut F7).
Method comparison
Method comparison is performed in clinical chemistry laboratories to evaluate the agreement between two analytical methods. Data are obtained collecting samples uniformly distributed in the reportable range. These samples are split in half and one piece is assayed by each method. MultiQC includes a module to calculate, plot and archive comparison data.
An e-learning how-to tutorial for the method validation module of MultiQC is available at www.multiqc.com.
Method comparison is performed whenever a new method is considered for replacing a current one. Interchanging methods may be
Permanent if the new method has better operational qualities than the former one.
Temporary if the new method is only an alternative method, which can replace the current one in case of failure of the routine analyser.
Cyclic when two analysers are performing the same assays at different hours of the day.
Method comparison is usually recommended when a new analytical method is started. But it is also useful to resort to method comparison in routine work either to troubleshoot an analytical issue or to demonstrate the permanent agreement between two analysers within a laboratory or between two laboratories.
Changing an analytical method to another one is only possible if they agree sufficiently closely. In this respect, two analytical methods can be equivalent, commutable or incompatible.
Methods are equivalent if they give equal results within their inherent uncertainty. They can be interchanged without loss of analytical accuracy.
Methods are commutable if they give equal results within the medical tolerance interval. They may be not rigorously equivalent but they can be however interchanged without loss of diagnostic power for patients.
Incompatible methods lead to results with differences greater than the tolerance interval.
Taking into account that our laboratories are intended to a clinical use, a departure of a new method from the current one can be accepted if it does not impair the medical diagnostic or follow up. So, the best cost- effective criterion that allows interchanging two analytical methods is commutability and not equivalence. Error made by replacing an analytical method by another one which is not rigorously equivalent is named the non-equivalence error. It is a component of total analytical error.
There are two ways to compare split samples assayed by two analytical methods.
The scatter-plot : the values of the new method are plotted against the corresponding values of the current one. The mathematical relation between methods is estimated by a regression line. The disagreement between methods is measured by the departure of the regression line from the bisecting line of the plot (identity line).
The difference-plot : the differences between concentrations in every split sample are plotted against the means of each pair. The disagreement between methods is measured by the deviation of the points from the horizontal nil-bias line.
Field - Ref (mmol/L)
2,0
1,0
0,0
-1,0
-2,0
110
120
130
140
150
160
Mean (mmol/L)
160
Field method (mmol/L)
150
140
130
120
110
110
120
130
140
150
160
Ref method (mmol/L)
Comparison of two serum sodium methods : difference-plot on the left and scatter-plot on the right (same data)
Both methods have pros and cons. They provide complementary information. CLSI (former NCCLS) has published guidelines for method comparisons where both scatter-plot and difference-plot are advised.
Medically acceptable error is a basic figure whose knowledge is essential to a cost-effective management of quality in a laboratory. It can be an absolute or a relative allowed error. Most often in clinical chemistry, both are associated so that absolute error applies to lower concentrations and relative error applies to higher concentrations. MultiQC maintains a table of medical tolerance intervals for every analyte that it controls. The software allows sophisticated tolerance schemes because it is possible to separately define relative and/or absolute errors for low, mid and high concentrations.
Difference-plot
For every concentration on the X axis, the error allowed for the difference within each split sample is represented by a vertical segment centred on the horizontal line of nil bias. On the whole, separate tolerance segments are merged into a polygonal area framing the nil-bias line. The top and bottom edges of this polygon are parallel for an absolute tolerance. The edges are diverging rightwards for a relative tolerance.
Interpretation of a difference-plot is easy : any point inside the tolerance polygon satisfies the tolerance conditions. Any point outside of the tolerance polygon denotes the non-commutability of the two analytical methods for the relevant sample. The final verdict is based on the percentage of non- commutable points found on the whole plot.
Concentrations
Allowed total error
< 130 mmol/l
5 mmol/l
130 to 150 mmol/l
3 mmol/l
> 150 mmol/l
5 mmol/l
Difference-plot with its tolerance polygon : Comparison of two serum sodium methods
Tolerance polygon
Regression of differences against means
Out-of-tolerance pair 
In-tolerance pairs
± 3 mmol/l tolerance 
± 5 mmol/l tolerance
6,0
5,0
4,0
3,0
2,0
1,0
0,0
-1,0
-2,0
-3,0
-4,0
-5,0
-6,0
Mean (mmol/L)
Field - Ref (mmol/L)
110
120
130
140
150
Scatter-plot
Building a tolerance polygon on a scatter-plot is less straightforward. The aim is to evaluate the departure of the regression line from the identity line. In routine laboratory work, slope and intercept of regression lines are generally calculated from a set of, say, 30 to 100 samples. Random error on regression lines is thus minimised to put non-equivalence error in prominent position. So the error allowed for a regression line is smaller than the total error allowed for separate
concentrations.
MultiQC makes use of a reduction factor named non- equivalence error budget. Its default value is 33%. This means that if the overall tolerance for serum cholesterol is 6%, the criterion allowing the interchange between two methods will be : non-equivalence error less than 2%. The tolerance polygon is built on a scatter-plot with this reduced tolerance. It frames the bisecting line.
Interpretation of a scatter-plot does not depend on separate points but only on the regression line. Any point of this line which is inside the tolerance polygon satisfies the commutability criterion. So the commutability range is established searching for the segment(s) of the regression line which is/are interior to the tolerance area.
160
Field method (mmol/L)
150
140
130
120
110
Scatter-plot : Comparison of two serum sodium methods.
(Background enlarged below)
110
120
130
140
150
160
Ref method (mmol/L)
MultiQC searches for the intersections of the regression line with the top and bottom edges of the tolerance area to find out the ends of all the in-tolerance segments. Then the software projects these segments onto the abscissa axis. The commutability range is outlined by a green background on the X axis. The methods under comparison are commutable if the commutability range is wider than the reportable range of the reference method.
Maximum non- equivalence
error = ± 1.7 mmol/l
Maximum non- equivalence
error = ± 1.0 mmol/l
Ref method (mmol/L)
Commutability range
Equality line Regression line Tolerance polygon
Point where the regression line crosses the edge of the tolerance polygon.
Non-equivalence error budget = 33%
Field method (mmol/L)
160
150
140
130
Concentrations
Allowed regression error
< 130 mmol/l
1.7 mmol/l
130 to 150 mmol/l
1.0 mmol/l
> 150 mmol/l
1.7 mmol/l
120
110
120
130
140
150
160
110
Enlarged background of the above scatter-plot : Comparison of two serum sodium methods
Is there a best regression method for a scatter-plot ?
Practice shows that quality of data matters much more than statistical models. Anyone can make his own opinion with MultiQC which can instantaneously switch between ordinary linear regression, Deming regression, weighted Deming regression and Passing-Bablok non parametric regression.
Discrepancies between scatter-plot and difference-plot
The figures below show a comparison between two serum calcium reagents : cresol phtaleine versus arsenazo. Plots are based on a medical tolerance of 4% and a non-equivalence error budget of 33%.
The scatter-plot (left picture) shows a commutable range of [74 to 120 mg/l]. So, the two methods should be directly commutable, at least for normal and elevated calcemias. The commutable range might be widened to lower concentrations by setting calibration correction factors, easy to calculate from the regression parameters.
There are however 5 out-of-tolerance pairs in the difference-plot (blue in the right picture). This bad agreement between methods cannot be explained by the imprecision of the one or the other. Capability indexes calculated by MultiQC are about 2 for each method. Both are therefore highly capable to individually meet the medical tolerance. Why then do the differences between methods do not meet the medical tolerance ? The answer is generally referred to as “aberrant-sample bias” whose origins may be differences in specificity, matrix effects or many other unknown causes.
Cresol phtaleine cpx (mg/L)
120
110
100
90
80
70
70
80
90
100
110
120
Arsenazo (mg/L)
8
Field - Ref (mg/L)
6
4
2
0
-2
-4
70
80
90
100
110
120
-6
Mean (mg/L)
The conclusion is that the scatter-plot and the difference-plot are complementary. The former evaluates the average agreement between all of the pairs whereas the latter individually focuses on each pair.
The scatter-plot is a tool to search for differences of calibration. The difference-plot is a tool to search for aberrant pairs.
A prerequisite to the difference-plot
The difference-plot implemented in MultiQC compares deviations between analytical methods to medical tolerance taken as criterion of judgement. This comparison is senseless if each method does not individually meet the criterion. So a prerequisite before any method comparison by means of a scatter- plot is to check that the capability index of each method under evaluation is greater than 1.
The Bland-Altman plot
Bland and Altman published a paper in The Lancet (1986) which popularised the difference-plot among clinical researchers. MultiQC does not implement the difference-plot exactly as it was described by Bland and Altman because it is not well adapted to clinical chemistry data :
The reportable range of analytical methods is much wider than the range of clinical measurements. Range ratios in routine clinical chemistry generally exceed 10. So the hypothesis of homoscedasticity is far from being fulfilled. A log transformation would be always necessary, making difficult reading of the plots.
The bias is rarely constant over the whole reportable range. Most often, differences of calibration create a proportional bias that depends on concentration. So the average bias has no practical meaning.
Allowed error schemes are more complex in clinical chemistry than the simple agreement limits of Bland and Altman.
Correlation coefficient
Correlation coefficient use is inappropriate for comparing analytical methods:
The correlation coefficient measures the strength of the relation between two variables, not the agreement between them. Two analytical methods may be highly correlated whereas a huge bias makes them completely incompatible.
The magnitude of the correlation coefficient is affected by the range of concentrations studied. The correlation coefficient can be made smaller by measuring samples that are similar to each other and larger by measuring samples that are very different from each other.
Correlation coefficient has no place in method comparisons because it does not answer the actual question of agreement and has an arbitrary value depending on the choice of analytical samples. Correlation coefficient is therefore not displayed by MultiQC.
Data entry for method comparisons
7
1
4
6
2
3
5
Click The main menu <Events -> Comparison> (shortcut F8) to open the dialog Method comparison. The left part of this dialog is devoted to data entry, the right part to plot and statistics.
The tab <History> gives access to the previous method comparisons. The tab <New comparison> enables all the entry fields.
Type in the grid the comparison pairs
Validate with the button <Apply>. This action triggers the computation.
Two tabs are available to display either the <Plot> or the relevant <Statistics>.
Five buttons instantaneously change the kind of computation for the same set of data.
An outlier can be withdrawn from the computation, without being withdrawn from the plot and from the grid.
Either click the relevant box of the column <Out> in the grid
8
11
9
10
13
12
Or click the outlying red point in the plot to turn it grey.
MultiQC has a generator of random pairs to easily investigate the different statistical methods proposed to compare analytical methods.
Click the main menu <Fake data> to show the panel opposite.
Type in the number of pairs of concentrations to be created.
Enter the lower and the higher concentrations of the random sample.
Select a type of distribution : <Gauss> creates data the more spaced out as we move away from the centre of the range. <Uniform> creates data equally spaced out within the whole range.
Type in the parameters of the theoretical regression between Ref and Field analytical methods.
Type in a CV for every analytical method.
The button <OK> draw lots for fake pairs of concentrations and load them to the data grid in the comparison window.
Every new saved method comparison creates an event Method comparison. An icon
is drawn in theReagent event bar.
Every new saved method comparison is logged by MultiQC as all the analytical events.
Method comparison events can be recalled by :
Icons
which are drawn in the Reagent event bar.The main menu <Events -> Comparison> (shortcut F8)
Acceptance charts
The analytical quality of an assay in clinical chemistry is satisfactory when the deviation of measured concentrations to the true values are not greater than the medically allowed tolerance. Keeping the whole analytical output within-tolerance should be the chief quality aim of any laboratory. It is very important not to confuse within-tolerance and in-control. They are not the same.
An in-control process is not necessarily within-tolerance : when the capability index is low, a process may produce out-of-tolerance results even if it is perfectly stable and in-control. Conversely when the capability index is high, a process may perform out-of-control and produce however within-tolerance acceptable results.
Control charts are often too strict
EWMA line
Control interval of the QC points
Target value
Control interval of the EWMA
Stable
All the points on the levey-Jennings QC chart below are « in-control ». They are randomly distributed around a mean equal to 70 U/l. They fluctuate within the control interval [68 - 72]. The EWMA line also remains within a narrower control interval [69.4 – 70.6]. The process is performing in a stable fashion.
One QC point is out-of-control
The EWMA is out-of-control
Drift
Stable
During the following days, the analytical process drifted upwards for unknown reasons. A first alarm was triggered by the EWMA line when it went across its upper control limit. Three days later, a second alarm was triggered by an out-of-control QC point.
According to the principles of quality control, the analytical process would need an adjustment. However, do not forget that we are in a medical laboratory. The analyte under control is ALT for which a medical
tolerance of 8 % is generally required. The observed drift of the QC material ranges to much less than 8
%. It has no clinical meaning. The process is obviously acceptable though out-of-control. A corrective action would waste time and money without any benefit for the patient. The level of surveillance provided by the standard Shewhart’s chart should be relaxed.
Acceptance charts, a pragmatic concept
A control interval which self-adjusts to slow drifts of the mean
EWMA line
Moving control interval
We need a chart that allows analytical drifts to a certain extent. Instead of building the control chart around a fixed target, let us build it around the EWMA with control limits calculated from the short-term standard deviation excluding the long-term variations of the mean. The control interval forms a green band moving around the mean.. In spite of a slow drift of the mean, the QC points stay in-control because the control interval automatically follows the fluctuations of the moving average.
A control interval manually re-initialised on intentional shifts
The EWMA was re-initialised
Sudden shifts often occur after routine changes or adjustments of analytical conditions (calibration, new lot, maintenance …). Due to its inertia, the EWMA would follow the analytical step with a too long delay. It is then necessary to manually re-initialise the EWMA to reset the centre of the control interval to the new shifted average. The gap in the green band below reminds of the intentional discontinuity in the analytical process.
Calibration
A fixed tolerance interval
Tolerance interval
To make the acceptance chart efficient we must set limits to the wandering of the control interval : these are the upper and the lower limits of the tolerance interval.
The green Gauss curve below represents the random distribution of analytical results due to common causes of variation. The span of this distribution is theoretically infinite but conventionally limited to 6 SD and named « analytical uncertainty ».
Out-of-tolerance fraction of the analytical output
Too great drift
Greatest permissible drift
Time
Uncertainty
Tolerance
A long-term cause of variation slowly moves the green Gauss curve rightwards and leftwards. A conforming analytical output is achieved as long as the Gauss curve does not encroach on the left or on the right out-of-tolerance band. It is the reason why the acceptance chart gives up controlling the slow drifts of the mean as long as the uncertainty interval remains nested within the tolerance interval. This is all the easier since the uncertainty is small and the tolerance is large.
The capability index of a method is precisely the ratio of the two intervals we are trying to compare :
Cp
Tolerance
Uncertainty
Acceptance charts are only feasible with high capability analytical methods. Several years of practice in a clinical laboratory have shown that Cp > 2 is a good threshold. Thus about 50% of the analytes in routine clinical chemistry can be controlled by an acceptance chart.
Working with acceptance charts
Three different situations can be encountered :
Within-tolerance process : the analytical method is within-tolerance as long as the control interval (the moving green band) remains nested within the tolerance interval (fixed yellow band). The analytical output is thus zero-defect (actually less than 1/740 non-conforming).
Out-of-tolerance analytical output
Capability exception
Capability exception : The moving green band encroaches on the lower or on the upper out-of- tolerance red band. A part of the analytical output is out-of-tolerance. This means that the drift of the EWMA is too great in comparison to the capability of the process to ensure a zero-defect production.
Control exception
Control exception : One QC point falls out of the mobile green band. In the absence of any intentional shift, this is an out-of-control condition. The process might go on at a pinch if the control interval remains nested within the tolerance interval. However, a special cause of variation occurred and a detailed investigation of the process should be started sooner or later. For the time being, the process can go on.
Parameters of an acceptance chart
An acceptance chart is built with three parameters
The tolerance limits, possibly tightened when the lower allowed Cpk is greater than 1.
The short-term SDST. This parameter can be floating (learning mode), estimated from a reference pool or specified. When SDST is not specified, its estimate from a reference pool is not very consistent if the number of QC points is too low. Therefore, when an acceptance chart is started, MultiQC does not build the chart with the true estimated SDST as long as the number of QC points is less than 10. The provisional control interval (green band) is
Current EWMA 3 Allowed error
4
The target value which can be also floating (learning mode), estimated from a reference pool or specified.
2
1
X X 2 n 1
SDLT
Variations of the analytical process charted in the picture above can be measured in terms of
1- Long-term (LT) variations: The double arrow shows the total spread of
QC values when all of the points are taken together. Long-term variations take into account the shift that occurred when the lot of reagent was changed. Long-term variability is measured by the average of the squared deviations from the target value (the centre green line on the picture below).
2
d
R
ST
SD
2- Short-term (ST) variations: The double arrow (2) shows the spread of the QC values about the mobile average of the process (the red line). Short-term variability is measured by the mean range of span 2, excluding the points where the EWMA has been restarted (e.g. the yellow point on the picture above). The significance of the coefficient d2 is explained in all QC textbooks.
When the analytical method is stable SDLT = SDST. Here, the method is not stable. SDLT > SDST because of the drifting and shifting average of the process.
Calibration charts
Calibration charts are a feature of MultiQC which must not be taken as an established method, but as a first step toward an alternative approach to the analysis of QC data. In the fifties, Levey and Jennings based their interpretation of QC data on the very concentrations resulting from the assay of each control material. Their QC protocol would have been exactly the same if the reportable range of analytical methods was restricted to two or three discrete concentrations. Nevertheless, laboratory practice has shown that analytical quality can be efficiently controlled by means of the classical Shewhart’s charts of the concentrations of three control materials.
Reducing an analytical process to a set of discrete and independent quality characteristics is however conceptually unsatisfactory. It would be desirable to take into account the continuous nature of analytical processes. So QC in medical laboratories should take benefit from the recent research activity in the area of quality profiles which was reviewed by Woodall et al. (Using Control Charts to Monitor Process and Product Quality Profiles. Journal of Quality Technology 2004, 36: 309-20).
Profiles are useful when a quality variable is functionally dependent on one or more explanatory or independent variables. In our medical laboratory field, profiles are the response curves of analytical methods. They relate measured concentrations to true concentrations. The theoretical relationship is equality, a very basic linear profile.
Instead of dealing with each QC vector as a set of a low, a medium and a high concentration that are directly controlled, we use them to calculate the regression line that relates the three assayed values to the 3 target values. This relationship between the “output” (actual assayed concentration) and the “input” (true target concentration) provides us with an estimate of the instant calibration line which becomes the new QC object. The resulting chart is therefore named calibration chart. It controls the day-to-day variations of the instant calibration line by means of a single chart.
Several pitfalls must be overcome before drawing calibration charts:
Analytical methods are heteroscedastic. So we are compelled to calculate the estimates of instant calibration lines by a linear regression weighted by the reciprocal of variances.
Instant calibration lines are defined by three attributes: slope, intercept and residuals. MultiQC does not take into account residuals yet because of theoretical issues that remain to be solved.
The X-values are coded with an offset equal to the barycentre of the target values of the control materials. Thus the estimators of the Y-intercept and slope are statistically independent, which makes interpretation easier.
1
The aim of calibration charts is not to bring to light sudden gross errors in analytical processes that are better detected by conventional QC but to follow the slow drifts of calibration in order to be able to decide a preventive corrective action before being out-of-control. Therefore slopes and intercepts of instant calibration lines are plotted after smoothing by an EWMA computation (exponentially weighted moving average).
Elements of a calibration chart
Calibration charts are drawn in MultiQC below the regular QC charts. The box <Calibration chart> of the main window displays/hides the calibration chart of each analyte.
Calibration charts are disabled when control is performed in specified interval mode.
The EWMA-smoothed slope and Y-intercept of instant calibration lines are simultaneously plotted on the same chart.
The red curve shows the EWMA of the slopes
The blue curve shows the EWMA of the Y-intercepts
The scale of slopes is drawn in red on the left side of the chart with a centre value equal to one.
The scale of Y-intercepts is drawn blue on the right side of the chart with a centre value equal to 0. The number displayed in the light-blue box (1.23 below) is the offset used to code the X-values (barycentre of the target values of the control materials).
8
9
Common control interval
10
7
2
3
5
6
The instant calibration line is out-of-control as soon as either the slope curve (red) or the intercept curve (blue) encroaches the upper or the lower yellow strips of the calibration chart.
The control interval is limited by two red dotted lines. It is computed for an ARL equal to 370. The extents of the right and left scales are chosen so that the same control interval applies to both slopes and intercepts curves.
4
The red and blue lines of the calibration chart are useful:
To evaluate the long term stability of an analytical method
To find out when instant calibration lines are out-of-control.
However the deviation of instant calibration lines to the theoretical response curve (slope = 1, intercept=
0) is difficult to imagine on the basis of the sole red and blue lines. The bias window is an explanatory popup window designed for that purpose. It plots the difference (bias) between the smoothed instant calibration line and the theoretical response curve as a function of the measured concentrations.
The calibration chart displays a vertical dotted line whenever the mouse hovers a column in correspondence to a QC vector. Clicking this dotted line turns it to a solid orange line and shows the bias window.
The equation of the smoothed instant calibration line is displayed at the top of the window. The green bias line does not directly represent this equation but its difference to the ideal calibration line Y = X. The bias line is green if both slope and Y-intercept are in-control, else it is red.
The vertical blue dotted line is a reminder of the offset of X-values.
Practice has shown that a calibration chart is the best tool to locate the periods of perfect stability of analytical methods. Therefore a button <Set targets = EWMA> is available to be able to automatically re- target the QC plots on the actual control concentrations.
Any change of the control targets is logged as a parameter-change in the top event bar of the main window. If the targets are changed several times without entering a new QC vector, only the latest change is logged.
Assay of total proteins with a 3-level QC: The biuret reagent is very alkaline and fixes the atmospheric CO2 when it is left open on the reagent tray of the analyser.
Results are high whenever a new reagent bottle is opened and go down little by little when the reagent gets older.
The calibration chart shows that the slope (red line) remains stable. The drift with time comes from a simple change of the intercept. The suggested corrective action is therefore to perform every day a reagent blank (and not a whole calibration).
The moving average of patients (MAOP)
Using patients' results for quality control in medical laboratories dates from 1955 (Hoffman et al). The basic assumption underlying this method of QC is that the population served by a clinical laboratory is medically invariant. This means that patients may change but the distribution of wellness and disease is statistically the same days after days.
The former implementation of quality control by patients' samples consisted in computing at the end of every working day the average of the patients (AOP) of the day and then in plotting these averages on Levey-Jennings charts. This retrospective control is too late and not suited to today's laboratories which are working 24/24 - 7/7 with a turn around time of less than 1 hour.
MultiQC implements a real-time moving average of patients (MAOP). The average of patients is no longer computed only once a day but is updated whenever a new patient is assayed. The programme continuously monitors the value of the MAOP and produces a real time visual and sound warning to notify the operator of any out-of-control value.
The MAOP is not a universal method of QC because of its conditions of use:
Its efficiency depends on the ratio of the variability in the population served by the laboratory to the analytical variability. The lower this ratio, the greater is the ability of the MAOP to detect drifts of analytical methods. Thus some analytes are better candidates for QC by MAOP than other ones.
The computer running MultiQC must be directly connected to the analyser to receive the patients’ results as soon as tests are completed.
The computation of MAOP smoothes the short term variations between patients to isolate longer term variations. The main feature of MAOP is thereby its inertia. It responds to analytical drifts with a delay that can be adjusted by the smoothing modulus. The higher the smoothing modulus, the longer the delay. The analytical throughput must therefore be high enough to allow the detection of drifts within a reasonable time. MAOP is not a QC tool for small laboratories.
The invariance of the population served by the laboratory is the most demanding condition of validity. The question raised by any out-of-control alarm from the MAOP is always the same : Have I just received a series of unusually low or high samples or has my analyser actually drifted?
This short story illustrates both the power and the limitation of QC by MAOP: During several months the MAOP for plasma potassium in the author's laboratory had been very stable and had never raised a false alarm. An unexpected out-of-control warning occurred on a cold afternoon of December. The MAOP for plasma potassium was increased by about 0.1 mmol/l. Several repeated assays of the mid QC material proved that the K+ electrode was perfectly calibrated. It was easy to find the source of the shift: a heavy snowfall had crippled traffic and we had received samples from distant hospitals several hours later than usually. A delayed spinning of blood was the obvious cause of the increased average plasma K+.
Patients’ results are stored in circular buffers as and when they arrive from the QC receiver interface. Each analyte is associated to a fixed-size buffer that is filled with the incoming analytical results. When a buffer gets full, the oldest concentration is deleted to make place for every new one.
1
2
4
3
4
The size of each buffer can be adjusted for the number of patients it can hold at a time. The possible sizes are 500, 1000, 2000, 5000 or
10 000 patients. The optimum size for each buffer depends on the analytical throughput of the laboratory. A satisfactory size should be able to keep one to three months of results.
Acquisition of patients can be disabled for the analytes which have a high between-patients variability that would impair the efficiency of QC by MAOP.
MAOP control can be disabled on Saturdays, Sundays and Holidays. Practice shows that the population served by a hospital laboratory working 24/24 and 7/7 changes with the day of the week. The ratio of disease on wellness increases on Saturdays, Sundays and Holidays because there are no outpatients. Therefore the MAOP may trigger false alarms on these non working days.
The buttons <Extend to whole section> facilitate the extension of options to all of the analytes in the same section as the current analyte.
The raw analytical output is enqueued in patients’ buffers without any filtering and the contents of each buffer is displayed in a real-time bar chart. The range of concentrations can thereby be very wide because it includes very pathological patients and analytical errors. MultiQC processes patients' data after two successive truncations.
Exclusion of extreme concentrations from the bar charts of patients
Extreme concentrations are excluded from the chart to prevent the interesting central part of the distribution from being too tight.
The picture below shows the distribution of 10 000 K+ tests from routine plasma samples in the author’s laboratory. Two interruption marks are drawn on the X axis to show the limits of its graduated portion. The excluded concentrations are merged into an extreme-low and an extreme-high bars. The exclusion interval results from a compromise between a complete but nevertheless detailed view of the distribution.
10 000 routine plasma K+ from the routine of an hospital laboratory.
Exclusion interval: extreme concentrations are taken out of the graduated portion of the X axis and merged into 2 bars:
Bar of very low K+ < 2.5 mmol/l Bar of very high K+ > 6.0 mmol/lLimits of the graduated part of the X axis
Truncation interval used to compute the MAOPReference range
Truncation of unusual data to compute the MAOP
1
2
Concentrations that are not excluded from the bar chart must be truncated again to compute the MAOP. The aim is to withdraw unusual concentrations from the moving average. A good starting truncation interval is the centred interval that holds 95% of the population served by the laboratory.
Do not confuse this truncation interval based on the very distribution of the results that you observe among your actual customers with the reference interval which is the centred interval that holds 95% of a theoretically normal population. For an hospital laboratory, the former is generally much wider than the latter.
The truncation interval can be manually typed in.
The % of the excluded population can be directly selected in the drop down pick list <Do symmetric truncation>.
Remark : The actual percentage displayed in the field <Truncated patients> is generally not exactly equal to the selected value because of the rounding of patients concentrations.
Reference range
"Normal" patients are differentiated on the bar chart from "pathological ones by two colours (default dark green and dark red). When a reference range is changed, the twin analyte is automatically updated with the same range.
Some authors recommended a truncation based on the reference range (the so-called AON = average of normals). This is a mistake. The aim of QC by patients’ results is to monitor the stability of the distribution of your own population of patients after having cleared it from the most unusual ones to improve the power of error detection. The exclusion limits must be derived from the locally observed bar chart on the basis of its actual shape.
Plotting the moving average of patients
The curve
The below curve shows the MAOP for plasma sodium.
5
2
3
4
1
Control interval of the MAOP
6
Natremias are stored in a circular buffer of 10 000 patients. The bottom X axis is graduated in number of patients. It shows 2000 patients. The graduation is linear.
The top X axis is graduated in time. A vertical grey line is drawn at the beginning of each day. The day of the week is displayed in orange when there is room enough between two successive vertical lines.
Some of these grey lines may be hidden if they are too near. The graduation is not linear because the laboratory does not process the same number of samples every day.
The left Y axis is graduated in concentrations.
The right Y axis is graduated in deviation around the target value of the MAOP. An alarm is triggered whenever the MAOP goes out of the interval 137.76 0.56 mmol/l. The meaning of the icon drawn at the right of the target value is explained in the next page:
Floating target value
Fixed target valueWhen the mouse moves on the plot, a vertical dotted cursor-line follows the horizontal movements and shows a hint with the date and time.
The MAOP is exponentially weighted (EWMA). The smoothing factor used for computation is not very informative. It is therefore displayed on the plot by the derived smoothing modulus (= ln 2/) that is more concrete because it is expressed in number of patients.
2
1
3
Changing the parameters of the curve
Several buttons are available at the bottom of the MAOP plot:
Move the scroller to glance through all of the patients stored in the buffer.
Two buttons can also scroll the patients
You can also scroll the charts by left-clicking the chart. The mouse cursor becomes a hand that horizontally drags the charts as long as the left button is not released.
A drop down pick list changes the number of patients displayed on the X axis.
The button <Control parameters> opens a dialog to change the parameters of the MAOP (see next section).
Setting up the parameters of the MAOP
Parameters of the chart
Smoothing modulus: The moving average of patients is used to smooth out short-term fluctuations between patients and highlight longer-term trends or cycles. The threshold between short-term and long-term is defined by the smoothing modulus. The longer the modulus, the greater is the smoothing of the curve and the inertia of the MAOP.
The MAOP is exponentially weighted. The smoothing modulus is directly related to the smoothing factor of the EWMA through the relation :
Smoothing modulus = ln 2 /
Coverage factor: This factor specifies the range of the control interval of the MAOP. The default value is 3 It means that an out-of-control alarm is triggered whenever the gap between the MAOP and its target value exceed 3 SD. The SD of the MAOP is computes, after truncation, from all of the patients stored in the buffer. The SD is updated with every new patient even if the target is fixed.
Target of the MAOP
Floating target: the target value of the moving average of patients is the mean of all the concentrations stored in the patients' buffer after truncation. This target is updated whenever a new patient is received by MultiQC.
Fixed target: The floating target is locked to a constant value as soon as the checkbox fixed target is ticked. This value is displayed on the right Y axis of the MAOP plot. To change the fixed target:
Move the mouse on the plot to place the green solid horizontal line on the wished new value
Press the left or the right button of the mouse to replace the green dotted line by the solid one.
As soon as a MAOP goes out of its control interval:
An alarm window is shown in the foreground of the screen and a sound warning is produced by the computer.
1
2
The name of the involved test is highlighted in the tree view of analytes.
Steps to take to troubleshoot a MAOP alarm
Init MAOP
Mute the sound by pressing the button <Mute> of the alarm window.
Select the highlighted analyte name to display the involved MAOP curve and un-highlight the name by pressing the button
<Test name OFF>.
To prevent the alarm from being raised again by the transmission of the next result, press the button <Init MAOP> This action restarts the curve at its target value within the in- control interval. However the MAOP curve will drift again and trigger later a new alarm if the analytical method has actually shifted. A corrective action is necessary.
Out-of-control MAOP warnings are very rare when the regular QC with control materials is correctly planned and performed. The role of the MAOP is secondary:
It detects analytical errors that might occur between regular QC points.
It creates a netting able to catch malpractices in QC procedures.
Mid term variations between patients
The distinction between short term and long term variations between patients may become confused when an additional mid term source of variation appears. This difficulty is exemplified in the picture below by the curve of the MAOP for plasma proteins in the author's laboratory. We observe a strong circadian rhythm which results from the habit of wards with acute patients (and low plasma proteins) to always bring their blood samples to the laboratory very early in the morning. Practically we are compelled to increase the smoothing modulus and thereby reduce the power of drift detection of the MAOP.
MAOP for plasma proteins in a hospital laboratory:
The first samples received every day mainly come from intensive care units whose patients usually have low plasma proteins. Therefore the MAOP falls down during the early morning. Then the MAOP goes up again when the laboratory begins to receive samples from outpatients or chronic patients.
Saturdays and Sundays must be excluded from the MAOP QC plot because on these days there is a great change in the population served by the laboratory: It is reduced to a majority of critically ill patients.
What is the power of a MAOP plot to detect errors ?
The distribution of the population served by a laboratory cannot be summarized in a theoretical model. The variety of breakdowns of our analysers is countless. The frequency of failures is unpredictable: An analyser may be subjected to several issues during the same week and then nothing during several years. So any computer simulation is worthless to validate a QC laboratory method or to evaluate its power of error detection. It is the reason why MutiQC has a built-in tool based on your actual patients.
The picture below shows a MAOP plot for plasma calcium.
Right click the plot to show its local menu and then select the submenu <Fake shift>.
This opens a green bubble window that can simulate the changes of the MAOP that you would have observed if you had undergone at this time a shift of your analytical method.
Select between absolute and relative shift and type in the wished value.
1
The button <Apply> re-computes the MAOP plot after having added the entered shift to all of the concentrations after the green vertical line. Note that if you enter a too high or too low value for the shift, many shifted concentrations will be truncated.
MAOP for plasma calcium
Top: the actual MAOP
Bottom: Calcium concentrations after the green vertical line were increased by 2%
The 2% shift would have produced an alarm after about 150 patients This value represents 2 SD on the control materials chart. It is far from sure that the drift would have been detected in the regular QC the day after.
2
3
Consulting MAOP curves from a distant computer
Several instances of MultiQC can be launched from distant computers (see section 13.4). Each instance has its own memory buffers storing the results of patients. Buffers are loaded from a disk file when the instance is started. The buffers of the main instance launched from the laboratory local computer are updated in real time by the flow of analytical data and are backed up to the disk every hour.
The buffers of the other instances hold the latest back up and are not updated in real time. Their changes are not taken into account by the main instance.
Authors have recommended the average of “normals” (AON) for laboratory QC. This is a moving average of patients (MAOP) that is not moving and whose truncation interval is the reference interval of the controlled analytical method.
Setting up an arbitrary truncation interval that is not related to the actual distribution of the concentrations found in the population served by the laboratory is an error: The bar chart on the right shows the distribution of 2000 partial pressures of oxygen in samples from patients of a general hospital. The green reference interval [70 – 100 mm Hg] is obviously totally inappropriate for the computation and the follow up of the average of patients.
Partial pressures of oxygen in 2000 samples received in the laboratory of a general hospital.
The reference interval [70-100 mm Hg] includes hardly 1 patient out of 4. It is much too narrow and off-centre to be an efficient truncation interval.
Importing, exporting data
Formatting the columns of the spreadsheet
You can easily import data from Excel by copy/paste. The QC vectors to be imported must be arranged in the columns of the spreadsheet according to the following pattern (one QC vector per row):
A- Section name: only the first 12 characters are taken into account.
B- Analyte name: only the first 12 characters are taken into account.
C- Date: the format is the "short date format" set in the locale parameters of Windows.
D- Time: if the time column is blank or incorrect the default time 12:00:00 is used.
E to J- Concentrations for the 6 control materials QC1 to QC6. Let blank if unused.
K- Comment: it is truncated to the first 40 characters, if it is longer than this. The first three columns are mandatory. Incorrect rows are ignored.
Importing
In Excel: Select the rows to copy (or the whole spreadsheet with the key Ctrl+A) and click the menu
<Edit€Copy> (blank columns or rows are ignored).
In MultiQC click the main menu <Data entry € Paste>. Data will be retrieved from the clipboard. You can validate immediately or enqueue data in the list of pending data for a later validation.
Importing from a pre-formatted “text” file
MultiQC can load QC data from “text” files which are formatted with the same pattern as the Excel spreadsheet in the previous section:
Section <#9> Analyte <#9> Date <#9> Time <#9> QC1 <#9> <#9> QC6 <#9> Comment <#13>
<#9> is the “Tab” character.
<#13> is the EOL character
Click the main menu <Data entry€File> of MultiQC and browse to the file to be loaded in the dialog that opens.
Open the table of QC vectors as indicated in section 9.1. Four options are available in the <Export> menu :
Copy to the clipboard.
Copy to Excel (if Excel is installed on the computer).
Save to an XLS (Excel) file.
Save to a text file.
On-line acquisition of QC data
QC receiver interfaces are independent programs to be installed on the same computer as MultiQC. They are intended to
Receive analytical data from a clinical chemistry analyser by means of a serial connection.
Retrieve control and patients results from the flow of analytical data.
Retrieve the dates and times of calibrations, the reagents lots and the bottle numbers.
Re-send all these data to MultiQC.
MultiQC works as a controller: It receives data from and sends commands to the QC receiver interfaces. A single controller can be connected to several analysers in the laboratory to manage QC data from different sources by the same software application. Refer to www.multiqc.com for further details.
QC file-link interfaces are designed to work combined with data processing software or a LIS which is in charge of outputting QC data to temporary shared files stored in an exchange subdirectory. File-link interfaces are programmes which regularly scan this exchange subdirectory. When QC files are found, they are read. Data is transmitted to MultiQC and the temporary files are erased. Refer to www.multiqc.com for further details