Intent-To-Treat Analysis

Intent-to-treat analysis (ITT) requires that all study subjects be included in outcome analyses in the study condition in which they were assigned, or were intended to be treated, regardless of actual treatment or adherence to the research protocol. After being assigned to a study condition, subjects may not actually use the treatment or intervention, may use less than intended doses, may drop out from the research program and therefore have indeterminate outcomes, or may even cross over between study conditions. These subjects may differ systematically from those who follow the protocol, and their removal can invalidate random assignment, introduce bias, and lead to inappropriate interpretation of statistical tests. The concept of ITT originated in pharmaceutical and randomized clinical trials but applies to behavioral interventions as well.

[Page 546]ITT is associated with effectiveness trials, which try to emulate real-world circumstances by offering an intervention but not assuming protocol compliance. In contrast to ITT, per-protocol analysis and efficacy subset analysis are associated with efficacy trials, which typically are more rigorous in ensuring that an intervention is not only offered to participants but complied with as well. When conducted on groups with nonrandom biases or errors, efficacy subset analysis can result in a Type I error probability higher than usually accepted alphas of .05 or .10, leading to inappropriate rejection of null hypotheses. The exclusion of such subjects from analysis of intervention outcomes may result in an overestimation of the intervention's effectiveness. On the other hand, ITT analysis can weaken the statistical power of a study and can require the application of methods for handling missing data or crossover subjects. It is possible to conduct both types of analysis on the same data for different research questions: ITT for effectiveness and per-protocol analysis for efficacy in different groups of subjects.

Some have traced the emergence of the debate surrounding ITT versus per-protocol analysis to the Federal Drug Administration and others’ criticism and defense of the Anturane Reinfarction Trial in the early 1980s. In this trial, a number of participants were determined to be ineligible after randomization, and although statistically equal numbers were removed from the experimental and control groups, more deaths were removed from the experimental group, leading to criticism of the reported outcome of the trial. ITT also was controversial in analysis of results from the 3-year Concorde Trial of AZT for asymptomatic HIV-infected patients that began in 1988. In this case, ethical considerations caused the Concorde protocol to change after 1 year of recruitment, so that some participants who had been randomly assigned to the placebo group were able to switch to AZT during the study. Following the ITT rule of ‘as randomized, so analyzed,’ the study suggested that AZT made no difference in progression to AIDS among asymptomatic people with HIV, countering accepted practice and stirring much controversy and concern among researchers and patients.

Research using the ITT principle can be enhanced if efforts are made in study planning and implementation to keep subjects as adherent to the protocol as possible and to minimize study attrition. Documentation of subjects’ actual use of assigned treatment is crucial and often is reported by means of detailed flowcharts and tables of study compliance and attrition at all stages.

...