 |
The 130th Annual Meeting of APHA |
|
The 130th Annual Meeting of APHA |
George Knafl, PhD, School of Nursing, Yale University, PO Box 9740, New Haven, CT 06536-0740, 203-785-6280, george.knafl@yale.edu, Kristopher P Fennie, MSc MPH, Yale School of Nursing, Yale University School of Nursing, PO Box 9740, New Haven, CT 06536-0740, Carol Bova, APRN, PhD, School of Nursing, University of Massachusetts, 208 Worchester Road, Box 215, Princeton, MA 01541, Kevin Dieckhaus, MD, Division of Infectious Diseases, University of Connecticut Health Center, 263 Farmington Avenue, Farmington, CT 06030-3212, and Ann Williams, APRN, EdD, School of Nursing, Yalue University, 100 Church Street South, P.O Box 9740, New Haven, CT 06536-0740.
A novel adaptive Poisson regression modeling approach is presented for analyzing electronic monitoring device event rate data. An example analysis is also presented using data on openings of electronic pill bottle caps from a two-year prospective clinical trial investigating the effectiveness of a home-based nursing intervention for increasing adherence among subjects with HIV undergoing highly active antiretroviral therapies (HAART). The modeling approach consists of partitioning the observation period, computing event counts and rates for intervals in this partition, and modeling event counts in terms of elapsed time after entry into the study. Poisson regression models of event counts are augmented with the appropriate offset variable to generate equivalent models of event rates. Event rates can have complex dependence on elapsed time with a variety of forms over all subjects. For this reason, a heuristic search is used to adaptively choose a set of power transforms of elapsed time on which to base the Poisson regression. Rules are used to control a systematic search through arbitrary sets of parametric models, thereby effectively generating a nonparametric regression fit to the data. Models are compared on the basis of likelihood cross-validation (LCV) scoring, a generalization of the commonly-used least squares form of cross-validation which is not sufficiently general to handle Poisson regression. A k-fold cross-validation scheme is employed that deletes one disjoint subset (fold) of the data at a time as determined through uniform random fold assignment, rather than the computationally less efficient leave-one-out approach of deleting one observation at a time.
Learning Objectives: At the conclusion of the session, the participant (learner) in this session will be able to
Keywords: Statistics, Adherence
Presenting author's disclosure statement:
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.