Selection
and Stratification
Predictions
from regression models may help to select patients for randomized
clinical trials through exclusion of patients at low or at high
risk of an adverse outcome.
QUESTION
2.2
Suppose
that a drug with known substantial side effects is studied in
a placebo-controlled trial. Which patients might be excluded?
Keep in mind
that balance is desired in prognosis when treatment groups are
being compared. This can be ensured by stratification.
Stratification may be based on single factors, e.g. some important
risk factors. Statistical models may also play a role here by
providing predictions that are used for stratification. For
example, predictions may be classified into several categories,
with a balance of randomization within each category ("stratified
randomization").