Henry Stewart Briefing on
ICH Statistical “Points to Consider” when
Analysing and Presenting Clinical Trial Data
in European Licensing Applications
20 & Friday, 21 February 2003
Radisson SAS Portman Hotel, London W1
Background to this
The ICH Guidelines concerned with statistics do not, and cannot,
address all the statistical topics encountered in licensing applications in
sufficient depth. Difficulties continue to arise in practice and a need for
additional guidance has been identified. To address this need, the EMEA’s
Committee for Proprietary Medicinal Products (CPMP) has released Points to
Consider (PtC) documents on methodological topics. They are targeted at
statistical topics found to be a regular source of problems during the
assessment of licensing applications, chiefly those concerned with the
design and analysis of clinical trials.
Data : Problems highlighted with the “Points to Consider” document on
The CPMP “Points to Consider” on missing data was adopted in
November 2001, but it may be revised. Missing values lead to bias, but judging
and correcting for this bias is fraught with problems. The interpretation for
an equivalence/non-inferiority study is different from that in a study
designed to show a difference. This PtC document is especially useful as it is
the first substantial regulatory guidance in this area and it is designed to
be readable by both statisticians and non-statisticians. This presentation
will highlight the important points of the document and raise some serious
questions concerning its overall applicability.
observation carried forward’ be the standard approach or have its
limitations been overlooked?
Is ‘worst case
imputation’ too conservative to be useful?
Will newer statistical
methodology, such as multiple imputation, provide better results and be
acceptable to regulatory authorities?
What about ‘missing
at random’ ?
Is there any real value
in recording data from patients who withdraw?
for Baseline Covariates
The CPMP “Points to Consider” on adjustment for baseline covariates was
issued as a draft in December 2001 but has not yet been adopted. The document
is very thorough and it expands on ICH E9 recommendations by clarifying which
baseline covariate should be included in the protocol and how results should
be reported. This presentation will highlight the important points in the
document and also challenge some of the contents.
How strong should an
expected correlation be to justify including it in the primary analysis
Is there a need for
unadjusted analyses when use of a covariate is standard for that
Can a covariate be used
if it was not specified in the protocol?
Is there now a standard
approach for adjusting for differences between centres?
Do the assumptions of
linearity and parallelism need to be checked?
Is it sensible to
categorize if the covariate is non-linear?