To assess whether the term "intention to treat" (ITT) predicts inclusion of all randomized subjects in the analysis, we reviewed 100 randomly selected reports of randomized trials that mentioned analysis by ITT. Only 42 of 100 reports included all randomized subjects in the ITT analysis. We could not determine which categories of participants were excluded from the ITT analysis in 13 trials. The most common categories of excluded subjects were patients who, after randomization, received no follow-up (16/100), received no treatment (14/100), or were found not to meet study entry criteria (12/100). We could determine the number of participants in the ITT analysis for 92 studies. Nineteen of the 92 studies excluded more than 5% of randomized participants, and 10 excluded more than 10%. There is considerable variation in how researchers define and apply the principle of intention to treat.