Approximate Causality for Longitudinal Binary Outcomes and Nonadherence

• Activity Number: 265
• Type: Topic Contributed
• Date/Time: Tuesday, August 10, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Thomas R. Tenhave*+ and Dylan Small and Kevin Lynch and David Oslin and Jing Cheng
• Companies: University of Pennsylvania and University of Pennsylvania and University of Pennsylvania and University of Pennsylvania and University of Pennsylvania
• Keywords: random effects ; logistic ; exclusion restriction ; randomization ; encouragement studies ; mental health

Abstract

We present a random effects logistic approach for longitudinal binary outcomes that adjusts for longitudinally measured treatment nonadherence in the context of two randomized behavioral intervention trials. Such an approach is an extension of the Nagelkerke et al. (2000) approximation for the cross-sectional binary outcome case, and is a response to requests for as-treated analyses to supplement intent-to-treat longitudinal analyses based on analogous random effects logistic models. By as-treated, we mean contrasting a condition with treatment versus a condition without treatment. For both trials, naively comparing treated and untreated conditions without consideration for the potential biases associated with treatment nonadherence led to unexpected results: the naïve as-treated contrasts were of smaller magnitude than the intent-to-treat contrasts, which is the converse of the typical case when there is a significant intent-to-treat effect. For both trials, we present an as-treated analysis that adjusts for treatment nonadherence such that the as-treated contrast is greater than the intent-to-treat contrast, while retaining the intent-to-treat inference based on p values.