Gold is not always good enough: the shortcomings of randomization when evaluating interventions in small heterogeneous samples.

The three criteria for valid inference in therapeutic intervention evaluation are achieving control, avoiding systematic error, and minimizing random error. The randomized, double-blind, controlled trial has appropriately been accepted as the methodological gold standard because it is the only method with the potential to avoid systematic error resulting from unbalanced distributions of recognized and unrecognized determinants of outcome. This potential is not always realized, however, particularly with small, heterogeneous patient samples--which undermines the rationale for randomization in these circumstances. Minimization is one possible strategy to attain validity in such circumstances, but the acceptability of nonrandomized strategies is currently hampered by deference to the concept of randomization. For each intervention evaluation, research design should be considered afresh, focusing on the criteria determining validity rather than particular methodological elements.