An additional basic science for clinical medicine: III. The challenges of comparison and measurement.

The main scientific virtue of randomized allocation of treatment is an unpredictability that reduces susceptibility bias in the groups created when treatments are assigned preferentially according to prognostic differences. Even in randomized trials, however, prognostic analyses are needed for checking imbalances in the randomization, for precise clinical application of results, and for discerning disparate therapeutic effects. Improved methods of prognostic analyses can enhance effectiveness in randomized trials and credibility for nonrandomized therapeutic comparisons. Regardless of randomization, therapeutic comparisons will also be improved with better analysis of proficiency for the main treatments, better attention to identifying "soft" clinical outcome events, and development of new analytic structures for complex problems in logistics. The challenges will require better scientific methods for the two types of clinical measurement: the mensuration with which descriptive expressions are provided for individual observations; and the quantification with which individual descriptions are grouped, summarized, and compared.