Estimands at ICH
Presented by Rob Hemmings
Problem statement
• Lack of clarity in trial objectives:
• Potential to confuse developers and decision makers.
• Misleading information to patients and prescribers
• So many conversations about HOW to estimate treatments
effects, and so few about WHAT to estimate.
• The HOW has determined the WHAT.
• WHAT do we want to estimate?
2
Context
• Patients differ in response to treatment.
• Some patients will tolerate a medicine and adhere to its
administration schedule, others will not;
• Some patients will require additional medication, others
will not.
• Some patients die.
• More than one treatment effect can be described and
estimated.
• Are these ‘intercurrent events’ problems for estimation of a
‘true’ effect or are they to be embraced?
3
Current practice under ICH E9
• E9 introduces the ITT principle as the best way to assess a
‘treatment-policy’.
• ‘Treatment discontinuation’ has been conflated with ‘Trial
discontinuation’. Data after treatment discontinuation or
perhaps use of rescue medication is not collected / not used.
Multiple problems have been labelled as ‘missing data’.
• Today’s practice doesn’t adhere to the ITT principle. It
hasn’t been clear WHAT treatment effect is then being
estimated, and arguably estimators (HOW) do not fully
respect randomisation.
• Can we answer questions other than treatment-policy whist
maintaining the benefits of randomisation?
4
ICH E9 (R1)
• Estimands; we have these NOW, in every trial, we just don’t
talk about them.
• ICH E9 addendum is currently being developed to provide a
structured framework for talking about WHAT to estimate =
the estimand.
• Statistical approaches then need be aligned with the
treatment effect of interest.
• How to write a document that fits all therapeutic and
experimental conditions?
5
ICH E9 (R1) … work continues …
6
Framework I
Choice of estimand may impact study design and conduct and needs
to be discussed first.
Trial
Objective
Target of
estimation =
WHAT
Estimand
Main Estimator
Sensitivity
Estimator 1
…
Sensitivity
Estimator k
Main Estimate
Sensitivity
Estimate 1
…
Sensitivity
Estimate k
Method of
estimation =
HOW
7
Sensitivity
analysis
Supplementary
analysis
Estimand attributes
8
Strategies
• At least 5 strategies for addressing intercurrent events e.g. the
intervention effect “regardless of switching to rescue medication”.
• Determine which intercurrent events need to be considered
explicitly in the construction of the estimand in order to give a clear
understanding of the treatment effect to be estimated.
• Multiple intercurrent events per trial means multiple strategies per
estimand
• E.g. Difference between mean change from baseline to week 24
in HbA1c in the target population, regardless of adherence, if
rescue medication is not available
9
Construction of an estimand
• Should be:
• consequent to the trial objectives and should precede
choices relating to data collection and analytic
approaches.
• clinically interpretable, in terms of the population and
endpoint, but also in terms of the intervention effect of
interest and, finally, the summary measure.
• duly justified considering the therapeutic setting and the
treatment goals of the intervention, from which the key
scientific questions of interest can be derived.
• a multi-disciplinary undertaking and should be the subject
of discussion between sponsors and regulators.
10
Impact: regulatory experience
• Current experience from Scientific Advice Working Party
– Company questions
– Plenary discussions
• Future expectations for Scientific Advice requests
– Identify intercurrent events, discuss strategies and
resulting estimands
– Move away from discussing endpoints?
– Discuss design and statistical analysis in relation to an
agreed estimand
11
Impact: regulatory experience
• Current experience from CHMP
– List of Questions; ask about the handling of ‘missing
data’ or treatment effect of interest?
– Plenary discussions
• Potential consequences for future MAAs
– One ‘estimand’ or many?
– Product Information and HTA engagement
– Note on external comparisons; ‘context’; meta-analyses
etc.
• Current expectations for future therapy-area guidelines
12
Thanks for your attention
13