2008 Courses and Workshops

IBC’s 16th International Intensive Symposium Biological Assay Development & Validation – San Francisco, CA

David Lansky will begin the workshop on Practical Approaches
to Establishing Parallelism for Your Potency Assay with a

Parallelism Primer
 

Carrie Wager will present a case study on

Generalized equivalence testing for parallelism

Dose-response curves for potency bioassay are typically modeled parametrically, using either linear subsets or four-parameter logistic functions. The aim of equivalence testing is to ensure that the difference between two curve shapes is small; for parametric models this involves simultaneous testing of equivalence of the model parameters. For some assays, such as those with partial curves or hooks, parametric models may be too limiting. We generalize the concept of equivalence testing to encompass an arbitrary curve shape.

A keystone of any quantitative method which utilizes reference material is that the test sample must behave similarly in the method as the reference material. In the field of biological assays – this characteristic, similarity, is a formal system suitability requirement for the assay. Similarity of the biological response for reference vs. test material in a biological assay is assessed by comparing the dose-response curves of the two sample types. This comparison must be a formal assessment, with pre-established acceptance criteria, and must include the critical parts of the dose response curves.

Historically, for animal assays, this involved comparing the slopes of two supposedly parallel lines, thus similarity is often referred to as parallelism. However, with the popularity of 4-parameter model fits, similarity has expanded to include similarity of asymptotes as well as slopes.

In the past five years there have been many heated discussions, not only about what parts of the dose response curve should be compared, but also about how this comparison should be carried out. Currently the US Pharmacoepia <111> and the European Pharmacoepia Section 5.3 recommend the F-test, however the USP is currently drafting a chapter recommending an equivalence testing paradigm. Meanwhile in the face of all this turmoil, assays are developed and curve similarity (parallelism) is ascertained.

Come to this workshop, learn the terminology, the math, the history and most importantly what bioassay professionals are doing in their labs to establish this critical assay system suitability.

IBC’s 3rd Annual BioProcess International Analytical and Quality Summit – Cambridge, MA

David Lansky present a paper on

One Statistician’s View of Effective Ways to Satisfy the Forthcoming New USP Guidelines on Bioassay Validationn

The revision of USP <111> .has been in process for several years. The USP committee quickly came to strong early consensus on a handful of issues that have been reported and published. On other issues there is no clear direction from the committee. This talk will contain one statistician’s view of the background for as well as the open questions in bioassay design and analysis. The open issues include: linear vs. nonlinear designs and analyses for bioassay, criteria for and appropriate methods for combining assays, weighted fits vs. mixed models, and big vs. small assays. This material will be presented to stimulate discussion.

JSM 2008 – Denver, CO

David Lansky will present a paper entitled

Challenges and opportunities for nonlinear mixed models in biological assays

as part of the Nested and Crossed Random Effects in Nonlinear Models Session.

Modern biological assays contain many sources of variation due to complex design structures. Randomized incomplete-block strip-plots protect assays from many sources of bias, but are not practical without robots. Related manual non-randomized designs protect against up to quadratic location bias. Analysis models contain more random effects than can be estimated; model selection is important. Equivalence testing for fixed effects selection is appropriate and leads to questions whether AIC & BIC (used in selecting random effects) are analogous to difference rather than equivalence testing. Assays designed for and fit with mixed models handle non-additive effects particularly well, yielding large improvements in assay performance. Statistical issues include: how to analyze a non-randomized design and whether equivalence-based methods are more appropriate for selection of random effects.