Two Intent-to-Treat Principles

• Activity Number: 384
• Type: Topic Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Thomas Permutt*+
• Companies: FDA
• Keywords: missing data ; causal inference ; intent to treat

Abstract

Most statisticians think all patients should be analyzed as randomized. Some clinicians think the outcome that should be analyzed is the status of each patient at the planned endpoint, regardless of adherence to assigned treatment. The two principles agree on what to do, in some cases: follow every randomized patient to the end, and analyze as randomized. Both have been called "the intent-to-treat (ITT) principle." But sometimes they conflict. In crossover trials, for example, analyzing completers only satisfies statisticians but may raise questions for clinicians about generalizability.

Furthermore, sometimes clinicians reject the clinical ITT principle, wanting instead a measure of what happens to adherent patients. The statistician's job is to create an analysis that as nearly as possible obeys the statistical principle of comparing like to like while meeting the clinician's need for a relevant measure of outcome. Three tools are available: retrieved dropouts, utility-based composite outcomes, and causal modeling of the effect among the treated.