st: Analysing compound data types from imputed variables

[roland andersson <rolandersson@gmail.com>]

We are analysing the discriminating capacity of a clinical score,
which is constructed from 7 variables. One variable is the ratio of
two variables, two are binary, one is an ordered variabel with 4
levels, and finally we have 4 continous variables which have been
divided into 2-3 intervals. We now want to 1) validate the score on
new data and 2) analyse if some additional variables can add
discriminating capacity to the original score.

Our problem is missing variables. It seems the missingnes is random.
We want to impute these missing values. This is straightforward using
ice or mi and using the passive alternative for the ratio-variable.
The problem is how to calculate and analyse the score for the cases
with missing variables. As we need to divide the continous variables
into intervals after imputation we can not use passive.

I am thinking to construct the imputed dataset, calculate the score
for each imputed dataset according to the intervals of the continous
variables. I suppose I can use mim for the analysis of the
discriminating capacity of the score. Is there any other alternatives?
Should we use the variables after dividing into intervals and then do
the imputation instead? It would then be possible to use passive
imputation for the score. Other proposals or comments are wellcome

Roland Andersson
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/