1- Statistical methods

1.1- Description of the software
1.2- Probability control limits
1.3- Mobile limits mode
1.4- Locked limits mode
1.5- Selecting risks
1.6- Fixed limits mode
1.7- Common CV method

1.1- Description of the software

MedLabQC is a programme designed for internal quality control in clinical laboratories. Two versions are available. The stand-alone version with keyboard entry of data is mainly aimed at manual analytical techniques and low throughput analysers. The automation server version, which complies with the Microsoft COM standard, can be interfaced with any laboratory information system.

The programme records the control charts, archives the data and calculates the validation ranges. Simple access to the main functions means that MedLabQC is a programme which can be immediately utilised without training by all laboratory personnel who are familiar with Levey and Jennings control charts. The number of analytes is unlimited and each is provided with up to three control levels. The three levels are validated simultaneously after display within coloured graphics (green, orange, red) which represents the control range of acceptable, warning and rejection. The Westgard's multirules 13S/22S/R4S/41S/10m are optional. Rejected control points remain displayed, however, but they are excluded from the statistical calculations. Validation of the control points can be postponed by storing data in a queue. At the end of each day, all of that day's control results can be displayed and printed. Thus daily viewing or the control performance is available in each section of the laboratory where the software is in use.

Whenever the lot number of a control material changes, the data from the previous batch are archived and the control charts of all of the relevant analytes are simultaneously restarted. All deletion and correction facilities are password-protected so they may be restricted to senior laboratory staff if desired. The data file is backed-up from time to time on a diskette at a pre-definable interval. A reminder is provided on the main screen if the backup interval has been exceeded. Exchanges between applications are possible through the clipboard or delimited text files. A random data generator creates simulated Gaussian-distributed data and can turn the programme into an educational tool.

MedLabQC breaks new ground in data management, as explained in the following paragraphs.

1.2- Probability control limits

Traditional QC requires two nested intervals limited by warning and rejection thresholds. These limits are calculated with the formulae m±2s and m±3s where m and s are the mean and analytical standard deviation of the control material. The choice of the multiples 2 and 3 of the standard deviation dates from before the computer age when it was necessary to simplify the design of control charts as much as possible.

Unfortunately, the multiples 2 and 3 of the standard deviation do not correspond to simple probabilities. According to the Gauss distribution, an in-control analyser goes out of the interval m±2s once every 21 times and out of m±3s once every 357 times. These fractions measure the average frequency of false alarms in traditional QC. They are often unknown in spite of their major conceptual significance. A more logical approach is to define simple probability figures and abandon "rounded" limits. Moreover, it is the common rule in all biological experiments. The researcher chooses an empirical accepted risk for type I error, then deduces the decision limits.

By default the programme uses the probabilities 1/20 (5%) and 1/500 (0.2%) for false warning and false rejection. The decision limits are therefore m±1.96s and m±3.09s. The difference compared with the traditional intervals is small and normal practice remains virtually unchanged. But the probability definition of control ranges has the advantage of allowing the calculation of alarm limits with a constant risk, irrespective of the number of available control points.

1.3- Mobile limits mode

By default the programme works in a statistical mode called "mobile limits". It is also adopted when a new batch of control material is started. It allows the start of pre-control during the preliminary period. Whenever the operator validates a control point, he accepts it as consistent with the previous ones. It is then justified to include the new point in the available sample and to re-calculate a new mean and standard deviation with an additional degree of freedom. Estimations become increasingly precise. At the same time, the control limits are updated with Student's t for the two risks 1/20 and 1/500 and the new number of degrees of freedom.

Practice demonstrates that the pre-control usually becomes reliable when about 10 points have been entered. When the number reaches about 30, Student's t approximates to the standard normal variable and the limits become independent of the size of the preliminary sample (m±1.96s and m±3.09s). They are not theoretically constant, however, because m and s are always recalculated whenever a new QC point is acquired.

1.4- Locked limits mode

It may be risky to keep mobile limits too long. It is easy to imagine how a slow drift in the analyser could gradually shift the mean, widen the control ranges and prevent alarms from occurring. It is necessary to lock the computation of m and s on a reference period (the so-called preliminary period). This statistical mode is called "locked limits". With MedLabQC it is possible to select as the reference period any series of consecutive points. Each control level has its own reference period. Thus one level can be individually restarted with the analytical method staying under the control of the others.

1.5- Selecting risks

The probabilities of the controls giving a rejection signal when an analytical error is present (the probability of false rejection) and giving a rejection signal in the presence of error (the probability of error detection) are antagonistic. When narrowing the control range, the chance of detecting an out-of-control state increases, but the number of false rejections increases at the same time. Five values are available for the probability limits of MedLabQC : 5%, 2%, 1%, 0.5% and 0.2%. By default, the risk of false alert is set at 5% and the risk of false rejection at 0.2%.

Westgard's multirules are very popular in the USA. Their efficiency is not constant. In completely random analytical series, the multirules detect more errors than the simple 13s rule but they increase at the same time the frequency of false rejections. They are mathematically equivalent to narrowing the m±3s rejection limits on a traditional QC chart. In return, with real analytical series, multirules are claimed to provide better detection of out-of-control states without increasing false rejections. This paradox is explained by the incompletely random dysfunctions of manual or automatic analytical methods. The series of control points are structured according to failure models (shift, drift ...) that are detected by Westgard's multirules with a particular efficiency. The value of multirules depends on the analytical system. They are optional in the programme.

1.6- Fixed limits mode

Statistical quality control is self-calibrated. Only the mean and analytical standard deviation define warning and rejection limits without any reference to diagnostic or therapeutic need. MedLabQC can work in a fixed limits mode where the control limits are specified empirically and independently of the analytical CV. This mode is appropriate for several laboratory situations.

Analytes with a high within-subject variability may often tolerate a higher level of imprecision without any consequence for medical decisions. Guidelines recommend careful planning of QC procedures to minimise costs. The wasteful practice of repeating analytical runs is unnecessary for drift without clinical significance. A frequently-quoted rule is that the tolerable analytical variability should be less than half the intrinsic biological variation. CLIA 88 document minimum performance characteristics which can be the basis for fixed control limits that maximise the cost-effectiveness in routine service.

Fixed limits are also mandatory in centralised QC of several identical analysers (e.g. in point of care testing). With statistical QC, a poorly maintained instrument has a larger acceptance range than one working well. It is necessary to fix absolute limits, common for all of the instruments.

Statistical QC fails with modern analysers having infrequent calibrations. To cut costs, each calibration is only duplicated or triplicated. Under those conditions, the calibration CV is not negligible compared with the analytical CV. When the method is re-calibrated, once a week, once a month or sometimes more rarely when a new batch of reagent is started, the analytical results are slightly shifted and the control materials trigger false alarms. With a fixed limit mode it is easy to define control ranges wider than m±1.96s and m±3.09s.

1.7- The common CV method

The theoretical probability of obtaining a rejection signal in a process within-control is easy to derive from the control limits. For instance, the rule 13s gives an average false rejection rate of 0.3%. But in practice, the control ranges are calculated individually for each level from the data of the preliminary period. The mean m and standard deviation s are subject to sampling errors and therefore the two limits m ± 3s. Depending on chance during the preliminary period, the real false rejection probability may fluctuate in a given chart and widely deviate from the theoretical value of 0.3%. This is particularly puzzling in a multi-level control because it can result in large differences between the rejection rates within an analyte.

When possible, it is better to avoid constructing independently the different control charts of the same analyte. The law of the common CV, whatever the concentration may be, is rather general in clinical chemistry. If verified, it is justified to estimate a common CV across the control levels and to derive the control limits from this common CV. The method equalises the false rejection probabilities for all of the control materials assigned to the same analyte. The probabilities do not vary independently any more, but do so simultaneously on the basis of the global sampling error of all of the acquired preliminary periods. Thus the frequent ambiguous situations with error signals concentrated on a single control level are ruled out. The common CV method is adopted by default in MedLabQC for the two statistical modes : mobile and locked limits.

In the case of a two-level control, if the mean and standard deviations of the two preliminary periods are (m1, s1) and (m2, s2), the common CV is calculated with the usual common variances formula :

The formula is easy to generalise for any number of control levels. In the common CV method, the control charts are constructed with the parameters (m1, m1.CV) and (m2, m2.CV) instead of (m1, s1) et (m2, s2).

2- Software configuration

2.1- Analytes
2.2- Control materials
2.3- Assignment of control materials to analytes

2.1- Analytes

Open the menu <Parameters | Analytes> to set the analytes (also accessible with a double-click on the list of analytes in the main window). Each one is defined by four entry-fields :

Four types of operations are possible :

Exit from the dialogue. The above changes can be reversed with the <Undo> button as long as they have not been accepted with <OK> :

2.2- Control materials

MedLabQC manages a list of all of the control materials used in the laboratory. Each one is registered once and then assigned to one or several analytes. In this way, whenever a lot number is updated due to a batch change, all of the relevant charts are simultaneously re-started.

Open the menu <Parameters | Control materials> to access the dialogue "Definition of control materials". Each control analyte is defined by two entry-fields :

Four types of operation are possible :

The button <Material used by> displays the list of the analytes controlled by one material.

Exit from the dialogue. The above changes can be reversed with the <Undo> button as long as they have not been accepted with <OK> :

2.3- Assignment of control materials to analytes

Open the dialogue "Assignment of control materials" with the button <Assign lots>. Two ways for assigning a control material to the current analyte are proposed :

The modified boxes are indicated by a yellow background.

Exit from the dialogue. The above changes can be reversed with the <Undo> button as long as they have not been validated :

3- Routine work

3.1- Main window
3.2- Control charts
3.3- Data entry
3.4- Daily summary
3.5- Table of numerical data
3.6- Printing the control charts

3.1- Main window

The whole display of the main window relates to the current analyte, the name of which can be read at the top of the right panel "Entry of QC points" (large yellow letters on a dark blue background). To change to another analyte, click on its name in the lower right list.

The nine buttons at the bottom are associated with the usual routine functions :

The more specialised functions can be accessed through the menus of the main window. Password-protection is provided whenever a correction or deletion of data is required. When the programme is first opened, the password is disabled. See paragraph 8.2 to learn how to set it again.

3.2- Control charts

The control charts are displayed in the left panel of the main window.

Mouse actions :

3.3- Data entry

The panel "Entry of QC points" comprises the entry fields needed for acquiring QC data :

To navigate in the fields use the mouse or the keyboard. The <Tab> key gives the entry focus to the next field and <shift + Tab> to the previous one. Two buttons are available :

3.4- Daily summary

The window "QC daily summary" provides an overall view of the complete series of control analytes performed on a particular day in each section of the laboratory where the software is in use. The default starting date of the calendar is the current date. It can be changed at will. In each case, a coloured spot indicates the range within which a data point was located on the control chart when it was validated :

Buttons :

3.5- Table of numerical data

The window "QC data" opened with the button <Data table> displays the QC numerical data for the current analyte and allows edition of incorrect entries. In each case, the coloured spots indicate the range within which a data point is located on the control chart.

Buttons :

3.6- Printing the control charts

The dialogue opened with the button <Print charts> provides 4 options :

4- Validation of entered data

4.1- Validity indicators
4.2- Westgard's multirules
4.3- Delayed validation

The window "QC validation" is opened after pressing the <OK> button of the panel "Entry of QC points".

4.1- Validity indicators

Each new control point is indicated by a small black vertical line on a three-coloured horizontal bar representing the control ranges :

Each data point can be individually rejected or retrieved. Rejected points remain displayed, however, but they are excluded from the statistical calculations of m and s.

4.2- Westgard's multirules

MedLabQC implements the standard protocol 13S/22S/R4S/41S/10m. Its display needs three conditions to be simultaneously fulfilled :

The within-material rules are represented by a row of small grey ellipses under each validity indicator. The colour turns red if a rule is violated except for the rule 1:2s which is yellow coloured because it is only a warning and not a true rejection signal. According to Westgard, the rules 1:4s and 10m are only significant if associated with a 1:2s violation. They are coloured yellow in the programme if individually violated.

Westgard's nomenclature 2s and 3s has been maintained despite the probability control limits used in the software. The figures 2 and 3 should be replaced by the Student's t for the number of degrees of freedom of s and the risks selected in the configuration of the programme. It would be better here to use the terms warning deviation and rejection deviation. The meanings of the rules are :

When more than one control material is available the rules are applied across the materials. The meanings are :

4.3- Delayed validation

Two exits from the validation dialogue are possible :

5- Control limits

5.1- Locking the control limits
5.2- Fixed control limits
5.3- Displaying the control limits

5.1- Locking the control limits

The default mode of the programme is the mobile limits mode described in paragraph 1.3. When the preliminary observation period is over, the QC mode can be switched to locked limits. In that case, new data do not change either m or s.

To define the QC reference period, activate the menu <Special functions | Lock limits>. A new panel "Locking limits" appears on the right of the main window (password-protected). To select the date limits of the reference period, click with the left button of the mouse on the dates scale displayed under the charts. Two possibilities are provided :

The selected period appears in blue inverted contrast. As soon as a date interval is defined, the grid in the right panel shows the partial statistics for the chosen period. To lock a control level on this reference period, press the relevant <Lock> button. The control charts are immediately updated. The operation can be reversed with the button <Mobile>. On the charts, the reference interval is indicated by a light blue background.

Note : It is possible to lock the three control materials on the same period, to define a particular period for each one or even to mix the different modes mobile, locked and fixed in the same analyte.

Final validation :

5.2- Fixed control limits

The importance of fixed limits is set out in paragraph 1.5. The menu <Special functions | Fixed limits> makes a new panel "Fixing control limits" appear on the right of the main window (password-protected). Three entry fields are provided for each control level :

The <Fixed> and <Mobile> buttons switch each control material from one mode to the other. The control charts are immediately updated. The final validation uses the same buttons as in the previous paragraph.

5.3- Displaying the control limits

The <Control limits> button opens a window that gives a summary of all of the statistical parameters of the control :

The checkbox <Common CV> activates the common CV method described in paragraph 1.7. When changed, the underlying charts are immediately updated.

6- Changing control materials and archiving

6.1- Simultaneously re-start all of the analytes
6.2- Re-start analyte by analyte
6.3- Initialisation of one control chart
6.4-Overlapping of new and old batches of control materials

Three changes are needed whenever a new control material is introduced :

Three ways of archiving are provided by MedLabQC.

6.1- Simultaneously re-start all of the analytes

This is the most frequently used procedure. A batch of control material is exhausted and replaced by a new one. All of the analytes controlled by the former batch have to be simultaneously initialised.

Open the window "Definition of control materials" with the menu <Parameters | Control materials>. Change the lot number and the name if necessary. Validate with the button <Change and erase former data> (password-protected). The new lot appears in the right list with a red cross as a reminder for data deletion. It is possible to re-start simultaneously several control materials. All of the changes can be reversed as long as neither the <OK> button or the <Enter> key have been pressed. A final security confirmation is required before each deletion of data. Erased data are saved in archive files (see paragraph 7.4).

6.2- Re-start analyte by analyte

Sometimes it is useful to change a control material for only one analyte, keeping the former material for the remaining analytes. First, register the name and the lot number of the new material in the window "Definition of control materials" (menu <Parameters | Control materials>). Next select the analyte to be updated in the main window and press the button <Assign lots> to open the dialogue "Assignment of control materials".

Make the changes as indicated in paragraph 2.3 and validate with the button <Change and erase former data> (password-protected). A security confirmation is required before the deletion of data. Erased data are saved in archive files (see paragraph 7.4).

6.3- Initialisation of one control chart

It is possible to erase all of the data in one chart without changing the control material. The menu <Special functions | Erase/archive levels> opens a panel "Erasing QC levels" on the right of the main window. Clear the control charts with the buttons <Erase/archive data>. The points appear grey. They can be retrieved with the button <Un-erase data>.

Final validation :

6.4- Overlapping of new and old batches of control materials

It is good practice to switch to a new batch of control material only after the data from a preliminary period have been aquired. During a transitional period, both the new and old batches are analysed. This situation is easy to manage with MedLabQC, irrespective of the number of control levels :

7- Files and exchanges between applications

7.1- Retrieving the data file of QualCont
7.2- Exchanges between applications through text files
7.3- Exchanges between applications through the clipboard
7.4- Archive files

MedLabQC stores data in a file "MedLabQC.qcf" (qcf for Quality Control File). Backward compatibility with the previous programme QualCont is ensured. Exchanges with other applications are made possible through the clipboard or delimited text files.

7.1- Retrieving the data file of QualCont

MedLabQC can read the data file of QualCont 2. Data can be retrieved either from the backup diskette or from the disk. Click on the menu <File | Import> to open the file selection dialogue. Drop down the picklist "type". Three types of file are available :

Select (*.dat ) or (*.sav) and find the QualCont file in the tree of the sub-directory/reader where it is stored. The analytes of the former programme are loaded with all of the control points. Only the names of the control materials are lost. It is necessary to enter them again manually (menu <Parameters | Control materials>) and assign them to the relevant analytes (button <Assign lots>).

7.2- Exchanges between applications through text files

MedLabQC can export the data of each analyte to a text file with fields delimited by tabulations. The data are organised in 5 to 7 columns : date, time, control levels (1 to 3), operators' initials and comments. This kind of file can be read by any word processor or spreadsheet programme. Click on the menu <File | Export> to type in the name of the export file.

MedLabQC can import a text file delimited by tabulations. The first column must contain valid dates. The second column is reserved for times. If it is empty or if the format is not valid, the default time 12:00:00 is used. The next three columns are reserved for the results of the three control levels. Any additional columns are ignored. Imported data are inserted in the current analyte. To prevent any mixing of data, it is advised that the current analyte is cleared or a new one created before importing.

7.3- Exchanges between applications through the clipboard

A simple exchange method is copy/paste. Use the menus <Clipboard | Copy> and <Clipboard | Paste>. The rules are the same as in paragraph 7.2.

7.4- Archive files

Control points erased in a chart following the change of a control material are stored in archive files named QCXXXXXX.qca (qca for Quality Control Archive) where "XXXXXX" is a number incremented from 000001 to 999999. They are also text files with fields delimited by tabulations, which are easy to read with any word processor or spreadsheet programme.

8- Miscellaneous functions

8.1- Backup
8.2- Password-protection
8.3- Random data generator

8.1- Backup

It is advised that a data backup is performed regularly on a diskette. MedLabQC provides a reminder for the operation. Open the dialogue "Backup/restore options"with the menu <Configuration | Backup/restore options> . Enter a backup interval in days (default 7 days) and select the drive letter (usually A or B) where the following files will be copied : MedLabQC.sav (last backup) and MedLabQC.old (previous backup).

When the backup interval has elapsed, the green spot in the button <Backup> of the main window blinks red. To back up, simply click on this button after placing a diskette in the drive. To restore click on the menu <File | Diskette restore>.

8.2- Password-protection

Correction/deletion operations are not normally available to some grades of laboratory staff, but are restricted to senior staff. These operations are password-protected. When the programme is initially opened, the password-protection is inactivated. To re-activate it, open the dialogue "Change password" with the menu <Configuration | Password>. Fill in the fields "New password" and "Confirm password" and validate with <OK>. If the two entry fields are left empty, the password-protection disappears again.

8.3- Random data generator

The random data generator is started with the menu <Special functions | Random generator>. It creates simulated Gaussian-distributed data to provide the user with an easy means of learning how to use the software and to test it. The new data can be either stored in the queue or directly inserted in the charts. To prevent any mixing of data, it is advised that the current analyte is cleared or a new one created before generating data. The entry fields are :

 

_________________________