Council for europe portal
Language : en Search
Choose language

CombiStats: What's new in version 6.1?

CombiStats 6.0 was released on 1 January 2020 following the change made in Section 6.2.3 of Chapter 5.3 of the European Pharmacopoeia 10th Edition concerning the semi-weighted combination of assay results.

Since then, some users have reported an issue when opening CombiStats files with a double-click. This issue has been resolved in version 6.1 (no other modification has been made to this new version of the software).

CombiStats 6.1 can be downloaded using the following link.

Your credentials for the 2020 licence are still valid so you don’t need to request a new password.

This is a quick guide to some CombiStats 6.1 features:

 

Semi-weighted combination of assay results

CombiStats allows you to combine the results of several assays according to the 3 calculation approaches described in Chapter 5.3 of the European Pharmacopoeia. The datasheet initially displays the individual potency estimates along with their 95% confidence limits. The p-value of the test for homogeneity of potency estimates is then calculated and tables of results are displayed for the unweighted, weighted and semi-weighted combination approaches. Version 6.1 implements the change introduced into the 10th edition of the Ph. Eur. to calculate the between-assay variability involved in the semi-weighted combination approach. Here is an example.

 

Password protection of templates and datasheets

This protection is provided via the button with a padlock icon https://www.edqm.eu/sites/default/files/cadenas.png. There are 4 levels of protection and each level can have its own password. Passwords are optional. If you do not enter a password when increasing the level of protection, a user can remove the protection without a password. More information on the 4 levels of protection and the use of passwords is provided in the manual.

 

Support for equivalence testing

Equivalence testing can be activated in the Options Wizard’s ANOVA tab to show the equivalence between samples and a reference standard.

For a parallel line model, the CombiStats datasheet includes a table with the estimated slopes of the individual samples and their 90% confidence intervals. This table also shows the differences in slopes and the slope ratios with respect to the reference standard with their 90% confidence limits. The aim is to check whether the confidence limits are contained within the pre-defined equivalence limits. If this is the case, the slopes are considered to be equivalent.

For slope ratio models, the test uses the y-intercepts and the table shows the differences in the y-intercepts with respect to the reference standard but not the ratios of y-intercepts since these are not meaningful in this case. Here is an example.

 

Robust regression to reduce the effect of outliers

Although the European Pharmacopoeia discourages users from applying outlier tests to exclude outliers, many laboratories still apply such a test systematically without looking into an experimental explanation for the occurrence of these values. To encourage these laboratories to abandon this practice in favour of better alternatives, CombiStats offers robust regression by the use of Huber’s weights. This can be activated in the advanced options by specifying a weight of w=h. Here is an example.

 

Automatic invocation of the Spearman/Kärber method

Some assays that are designed to be analysed with probit models (or similar parametric curves) may have insufficient data in the central part to allow estimation of the slope. Often, an estimate can still be obtained using the Spearman/Kärber method. CombiStats detects when parametric models cannot be used and automatically invokes the Spearman/Kärber method. A message that this is the case is printed on the output. Here is an example.

 

5-parameter sigmoid curve regression

It allows a 5th parameter to model asymmetry of the curves. It can be activated in the Options Wizard’s “transformation” tab and is available only for quantal models and quantitative sigmoid models. Here is an example.

twitter facebook googleplus pinterest linkedin email