Adaptive Repeated Measures Modeling using Likelihood Cross-Validation

G.J. Knafl, K.P. Fennie, and J.P. O'Malley (USA)


Adaptive modeling, likelihood cross-validation, longitudinal data, medication adherence, nonparametric regression, repeated measures


An approach for exploratory repeated measures modeling is presented, based on adaptively selected fractional polynomial models for the expected outcome value combined with the standard compound symmetry covariance structure for representing within-subject dependence. The search strategy for identifying an appropriate model for the expected value is based on likelihood cross-validation scores together with tolerance parameters specifying how much of a change in those scores can be tolerated. These methods extend existing adaptive regression methods for uncorrelated situations to correlated outcomes settings. They also serve as a test case for development of methods to handle more general covariance structures. Example analyses are presented using longitudinal data on self-reported adherence to antiretroviral medications for HIV-positive subjects.

Important Links:

Go Back