How best to handle missing data in intent-to-treat analysis of longitudinal data from randomized clinical trials is a complex and controversial area1 with a number of different, valid approaches. However, we doubt that the one Mr Giglio suggests is among them. By considering only those urine screens collected in the last week of treatment in his definition of a "successful outcome," Mr Giglio disregards over 90% of the data, considerably reduces the statistical power of the study, and correspondingly inflates the type II error rate.
Avants SK, Margolin A, Holford TR, Kosten TR. A Randomized Controlled Trial of Auricular Acupuncture for Cocaine Dependence: Treatments vs Outcomes—Reply. Arch Intern Med. 2001;161(6):895. doi: