NONMEM Users Network Archive

Hosted by Cognigen

RE: Why population PK approach is good for pooling data compared to Classical statistical analysis?

From: Smith, Mike K <mike.k.smith>
Date: Thu, 23 Oct 2008 10:27:06 +0100

Thanks Nick, for pointing us in the direction of that interesting (and
provocative) article from Sir Michael Rawlins.

QUOTE - "Sir Michael rejects the trend to grade various kinds of
clinical trials and studies on scales of merit which he says has come to
dominate the development of some aspects of clinical decision making."

This is an odd statement coming from the chair of an organisation which
is increasingly using Bayesian Evidence Synthesis and Cost-Effectiveness
arguments to underwrite its decisions. In these approaches weighting
RCTs and Observational studies allows us to trade off the rigour of RCTs
with the "real world" results of Observational trials. To paraphrase a
well known Clinical Pharmacologist "Haven't we learned anything from 100
years of RCT design"? (Actually - please don't respond to that one!) We
*are* moving (slowly) from the confirm-confirm-confirm paradigm of the
"old world" statistical designs into more quantitative drug development
where we *do* learn from previous studies, carry forward information
quantitatively, predict outcomes, select designs which help address gaps
in our knowledge and move on. Drug development in industry is a business
and we are cost and time-constrained. We need to balance good science
with more efficient drug development strategies. If we can't do this
then drug development as a business is unsustainable.

QUOTE - "Generalisability - RCTs are often carried out on specific types
of patients for a relatively short period of time, whereas in clinical
practice the treatment will be used on a much greater variety of
patients - often suffering from other medical conditions - and for much
longer. There is a presumption that, in general, the benefits shown in
an RCT can be extrapolated to a wide population; but there is abundant
evidence to show that the harmfulness of an intervention is often missed
in RCTs."

I agree with some aspects of his critique around the limitations of
RCTs. In drug development we aim to learn about aspects of the drug and
utilise Sheiner's "Learn and Confirm" cycles. However we can't get away
from the fact that robust, quantitative proof of effect needs to come
from a comparison of like with like - and RCTs are the best source to
provide that evidence. We know that once the drug is licensed it will go
out "into the wild": how it is prescribed by the MDs, how it is used by
patients means that more often than not the drug is not taken "per
protocol" in the real world. MDs and patients sometimes do strange
things with medicines. Some of them we can't even dream of... No amount
of Pop PK/PD, RCTs, learning or confirming will be able to address that
issue. I guess the issue is that regulatory proof of efficacy using RCTs
doesn't square with a body like NICE's needs for assessing
cost-effectiveness in the real world setting where observational studies
might be more informative.

As for his comments about cost, unnecessary RCTs or impossible
indications - unfortunately drugs which have a marked effect that appear
quickly and where safety concerns pop out clearly are getting harder and
harder to find. If only 1 in 30 drugs that enter clinical trial testing
make it to registration then we're more likely to find drugs that FAIL
than stop early due to efficacy (sadly). So our job should be to kill
drugs early and cheaply as possible. With increased burden of evidence
on safety from regulatory (and the public) I think it is inevitable that
clinical trial costs will go up.

I'm not sure whether Sir Michael was intending to be provocative, but
his comments are sure to stir up debate, which is probably a good thing.


-----Original Message-----
From: owner-nmusers
On Behalf Of Nick Holford
Sent: 16 October 2008 21:28
To: nmusers
Subject: Re: [NMusers] Why population PK approach is good for pooling
data compared to Classical statistical analysis?



Finally as Sir Michael Rawlins (Chairman of the NICE in the UK) pointed
out yesterday the traditional statistical approach to clinical trials
does not adequately describe the clinical pharmacology and benefits of
medicines. The flexibility of the population approach allows it to used
for 'learning' as well as 'confirming' (Sheiner 1997). This combination
of approaches is in keeping with the broader philosophy posed by

Sheiner LB. The population approach to pharmacokinetic data analysis:
rationale and standard data analysis methods. Drug Metab Rev.
Sheiner LB. Learning versus confirming in clinical drug development.
Clinical Pharmacology & Therapeutics. 1997;61(3):275-91.
Received on Thu Oct 23 2008 - 05:27:06 EDT

The NONMEM Users Network is maintained by ICON plc. Requests to subscribe to the network should be sent to:

Once subscribed, you may contribute to the discussion by emailing: