Abstract:
|
Ecological momentary assessments (EMA) deliver repeated measures of a subject's cognition, affect, behavior, and experiences in their natural environment. They provide insight into the mechanisms guiding behavioral processes, such as transitioning between smoking cessation and relapse. EMA data rely on compliance, and consequently, observations are often unbalanced and unequally spaced. This type of data (i.e., panel data) is often analyzed with multistate, continuous-time Markov models. Practical use of these models relies on appropriately including covariates into the model and defining the transition probability matrix. Typically, the transition matrix is specified to test a biological or behavioral theory. Despite vast research on variable selection methods, the selection of covariates associated with transition probabilities is limited to pairwise model comparison tests. Our goal is to develop a variable selection method for this class of models to identify associated covariates in exploratory public health settings. We validate our method through simulation and apply it to EMA data to identify risk factors associated with transitioning between smoking cessation and relapse.
|