Table.5.

Guidance for conducting covariate-constrained randomization

1. Identify prognostic baseline characteristics a priori using background literature, historical data, or previous trials.

2. Generate all (or at least 1000) simple randomizations to identify baseline characteristics that are always balanced between treatment arms (e.g., ≥95% of the time).

3. Carefully consider the number of variables added to the constraining process or consider using a dimensionality-reduction method for many variables (e.g., principal component analysis).

4. Consider the amount of missingness of constrained baseline characteristics prior to randomization.

5. Enumerate all possible allocation schemes when there are fewer than 20 clusters or at least 100,000 allocations otherwise.