A Pragmatic Intent-to-Treat Analysis Using SAS

Intent-to-treat analyses are warranted in clinical trials, which usually contain right-censored data due to treatment failures or dropouts. An explanatory analysis, easily achieved with PROC MIXED, assumes data missing at random, which is not the case in clinical trials since the dropouts are often associated with the treatment assignments (Little and Rubin, 1987). A pragmatic analysis (Hogan and Laird, 1996) is a mixture model utilizing information from the explanatory part of the analysis and information after censoring. This is done by fitting a piecewise random effects model with both "on-" and "off-treatment" slopes which can be accomplished in SAS by combining the parameters from the mixed effects model using PROC MIXED with the proportion of censoring obtained from PROC LIFETEST. To illustrate, this technique was applied on simulated data from Table 2 in Hogan and Laird (1996) to calculate final IQ ratio values. Bootstrapped standard errors are used to make inferences.