The focus of this renewal application is the development of a new class of randomized designs for testing longitudinal adaptive treatment strategies in mental health. This extends the work done in the initial grant, which focussed on the analysis of observational data for causal inferences about time-varying treatment. This extension flows naturally from the finding that identifying the treatment strategies latent in the observational study proves to be an even greater challenge than the statistical analysis of the data. We discovered that analytic methods we developed for the observational case would be even more useful in experimental data, providing a sound basis for inference in a flexible randomized design we refer to as Biased-coin Adaptive Within Subject (BAWS). In a BAWS design, the patient's current state (and history of response) is used to influence the future treatment, through a """"""""biased-coin algorithm, similar to that used in adaptive allocation strategies for balancing on baseline characteristics. We propose to continue development of these designs, to make them ready for use. We propose 4 main aims: 1. Provide a detailed theoretical specification of the BAWS design with examples and a coupled modeling strategy that delivers efficient, randomization-based estimates of important causal effects. Specifically, we propose to (a) develop the design and analysis for defining the optimal switching rule when a first-line option is to be followed by a second-line option in """"""""non-responders"""""""" to the first option) develop estimates of the causal effect of the clinical decision process (continual reevaluation of treatment based on current results) that is reified by BAWS, compared to the strategy """"""""treat continuously with the first-line option"""""""". 2. Using mathematical modeling and simulation, develop power and sample size requirements for the BAWS designs. 3. Using data from The Collaborative Depression Study, develop estimates of the kinds of BAWS designs that will result in enhanced compliance while yielding maximally informative results. Specifically we will estimate (a) the way that various summaries of symptom history vary over time and ('))the way that these summaries determine changes in naturalistic treatment choices. 4. Develop language for """"""""informed consent"""""""" in such studies, that explains the design, its ethical consequences, and its risks, to clinicians and patients and pilot test these materials.
Showing the most recent 10 out of 25 publications
