Dear all,
We are faced to a quite common problem in clinical trials, all suggestions,
recommendations about literature and software are highly appreciated:
In a RCT we are interested to estimate the effect of an intervention on
different continuous outcomes (as surrogate endpoints) and survival (as the
main endpoint).
For evaluation of the efficacy of the intervention we use 3 time points:
baseline and 1,2,3 month after the intervention.
Since the surrogate markers are associated with survival (of course they
are!) we are not able to obtain an unbiased effect of the intervention
(using the markers as outcome) without “integrating out” this risk
associated drop-out process:
When following up patients we observe a “selection effect”, i.e. individuals
with higher risk (and also worse values of the surrogate variables) die
earlier. Since there seems to be an effect of the intervention this
selection effect is stronger in the control group so that we are not able to
observe all 3 post intervention measurements for all patients. By contrast,
in the intervention group, all patients survive until 3 month.
The problem:
- When we use all available measurements for all patients we underestimate
the effect of the intervention because the control group becomes “better”
during the follow-up due to the selection effect mentioned above.
- When we exclude all patients that die earlier, we also do not get an
unbiased estimate of the effect since we exclude high risk patients from the
study population. The bias for this approach might be worse since we expect
effect modification by baseline values.
When we use survival as the outcome we do not have enough power (at this
initial stage of the trial) to show the effect of the intervention.
So we need an approach to analyse surrogate endpoints and survival jointly,
what are the state-of-the-art approaches for this data situation?
Many thanks
Bernd
----------------------------------------------------------------------------
------------------------------------
Dr Bernd Genser
MSc, PhD
Director
BGStats Consulting
Statistical Consulting - Data Analysis - Medical Research Support
email: [log in to unmask]
phone: +44 20 7193 7448
<http://www.bgstats.com/> www.bgstats.com
----------------------------------------------------------------------------
------------------------------------
<skype:bernardinho1969?call>
|