This may seem a simple question but I don't seem to have a good grasp on
the answer.
I have a 5-variable MVN imputation application, specifically 5
sequential years of body mass index (BMI) measurements -- 2 of them are
before an intervention, 1 is during the intervention, and 2 are after
the intervention:
Y1 - pre-baseline
Y2 - pre-intervention baseline
Y3 - intervention
Y4 - post-intervention immediate follow-up
Y5 - later follow-up
The main interest is in Y2 and Y4.
Only a small fraction of the cases (~10%) have fully observed data and
the missingness is non-monotone. We are planning MCMC MI.
There is an argument to exclude from the imputation and main analyses
all cases that are missing both Y2 and Y4. This is based on hesitation
to impute both the main pre- and post- measurements.
There are additional variations on this. For example, exclude any case
that is missing all (Y2,Y3,Y4) and any case that is missing (Y3,Y4,Y5).
This reflects unwillingness to impute Y2 if both the year before and
year after are missing (and the same for Y4).
The counter-argument is that such selective strategies might lead to
biased imputations. And the correlations are high -- ~0.90 for
consecutive measurements and even ~0.75 for the two most distant
measurements (Y1/Y5). So, even without contiguous years, one might get
reasonable imputations.
Can someone more knowledgeable weigh in on this please? And could you
point me to some relevant references?
Many thanks.
CD
--
The documents accompanying this transmission may contain confidential
health or business information. This information is intended for the
use of the individual or entity named above. If you have received
this information in error, please notify the sender immediately and
arrange for the return or destruction of these documents.
Constantine Daskalakis, ScD
Associate Professor,
Thomas Jefferson University, Division of Biostatistics
1015 Chestnut St., Suite M100, Philadelphia, PA 19107
Tel: 215-955-5695
Fax: 215-503-3804
Email: [email protected]
Webpage: http://www.jefferson.edu/clinpharm/biostatistics/
