But ---
this is only correct after a thorough study of the clinical
problem, which is required before a valid predictive model can
be developed. By such study, a list of candidate predictors
can be assembled, and the relevance of the data set judged.