Clinical Data Where Variability Is More Important Than Averages

Abstract

In clinical studies, efficacies of new treatments are usually assessed by comparing averages of new treatment results versus control or placebo. However, averages do not tell the whole story, and the spread of the data may be more relevant. For example, when we assess how a drug reaches various organs, variability of drug concentrations is important, as in some cases too little and in other cases dangerously high levels get through. Also, for the assessment of the pharmacological response to a drug, variabilities may be important. For example, the effects on circadian glucose levels in patients with diabetes mellitus of a slow-release-insulin and acute-release-insulin formula are different. The latter formula is likely to produce more low and high glucose levels than the former formula. Spread or variability of the data is a determinant of diabetic control, and predictor of hypoglycaemic/hyperglycemic events. As an example, in a parallel-group study of n = 200 the former and latter formulas produced mean glucoses of 7.9 and 7.1 mmol/l, while standard deviations were 4.2 and 8.4 mmol/l respectively. This suggests that, although the slow-release formula did not produce a better mean glucose level, it did produce a smaller spread in the data. How do we test these kinds of data. Clinical investigators, although they are generally familiar with testing differences between averages, have difficulties testing differences between variabilities. The current chapter gives examples of situations where variability is more relevant than averages. It also gives simple statistical methods for testing such data. Statistical tests comparing mean values instead of variabilities are relatively simple and are one method everyone seems to learn. It is a service to the readership of this book to put more emphasis on variability.