Adherence to HIV medications is especially important for preventing partial suppression of viral replication with its enhanced risk of drug resistant HIV and is increasingly being measured in clinical trials with electronic monitoring devices (EMDs). EMD data are rich in longitudinal information often not used to their maximum potential. Summary measures are most commonly used, but do not provide sufficient detail for describing complex medication-taking patterns. We recently developed alternate methods for modeling EMD data at both the individual- and multiple-subject levels providing new insights into adherence patterns and evaluated these methods using MEMS cap data from a clinical trial testing the effectiveness of a nursing intervention for improving adherence to HIV medications. These methods utilize adaptive Poisson regression modeling of grouped EMD data with likelihood cross-validation for model evaluation and rule-based heuristic search through parametric models generating a smooth nonparametric regression fit. We now propose to develop original statistical methods extending adaptive Poisson regression for the purpose of improving its usefulness to HIV researchers and clinicians by addressing the following specific aims: 1) Identify subperiods within the EMD observation period over which a subject or group of subjects exhibits distinctly different adherence patterns. 2) Identify the dependence on time of variability in adherence for a subject or group of subjects. 3) Identify classes of subjects with distinctly different adherence across those classes and similar adherence within those classes for evaluation of possible differential effects of an intervention across those classes as well as the impact of such adherence classes on clinical outcomes. To accomplish these aims, we will develop algorithms for adaptively determining subperiods of distinctly different adherence patterns and for relating these changes to known changes in subjects' treatment and experience; for incorporating changes in variability in adherence using nonparametric quasi-likelihood methods; and for adaptively determining I parsimonious classifications of subjects for predicting change in adherence and its effect on clinical outcomes and for assessing how much of such change can be attributed to specific known factors especially intervention group membership. We will evaluate these methods, using available EMD data for HIV subjects, to assess their usefulness in the understanding, treatment, and prevention of HIV disease/AIDS.
|Knafl, George J; Dixon, Jane K; O'Malley, Jean P et al. (2012) Scale development based on likelihood cross-validation. Stat Methods Med Res 21:599-619|
|Knafl, George J; Bova, Carol A; Fennie, Kristopher P et al. (2010) An analysis of electronically monitored adherence to antiretroviral medications. AIDS Behav 14:755-68|
|Knafl, George J; Grey, Margaret (2007) Factor analysis model evaluation through likelihood cross-validation. Stat Methods Med Res 16:77-102|
|Fennie, Kristopher P; Bova, Carol A; Williams, Ann B (2006) Adjusting and censoring electronic monitoring device data. Implications for study outcomes. J Acquir Immune Defic Syndr 43 Suppl 1:S88-95|