Statistics and Its Interface

Volume 7 (2014)

Number 2

The joint assessment of longitudinal multidimensional functionings in overweight and obese elderly with a time varying covariate

Pages: 297 – 305

DOI: http://dx.doi.org/10.4310/SII.2014.v7.n2.a13

Authors

Hyokyoung Grace Hong (Department of Statistics and Probability, Michigan State University, East Lansing, Mich., U.S.A.)

Satrajit Roychoudhury (BDM Oncology, Novartis Pharmaceutical Company, Florham Park, New Jersey, U.S.A.)

Pulak Ghosh (Department of QM & IS, Indian Institute of Management, Bangalore, India)

Abstract

The occurrence of overweight and obese older adults in the US has increased substantially during the past decades. Toward the goal of overweight or obese elders’ well-being, it is important to detect early functional decline. In contrast to the majority of the previous research, which considers a single dimension of the functioning of the elderly, we consider four-dimensional functionings for daily living- physical, sensory, emotional, and social functioning simultaneously. The challenge of our study is that functionings and some predictors are longitudinally measured and mixed types discrete form; physical functioning is measured by count variable exhibiting the excess zero. The other three functionings are on an ordinal scale. To deal with these complications, our technique utilizes a zero-inflated Poisson regression model to account for the excess zero in the physical functioning. The sensory, emotional, and social functionings are modeled via the ordinal model, and those four functionings are connected by correlated random effects and the model parameters are estimated using a Bayesian approach via Markov chain Monte Carlo algorithms. Analytical results based on the Second Longitudinal Study of Aging show that selfrated health commonly affects our interested dimension of the functionings, sensory functioning most obviously deteriorates with aging and emotional well-being remains relatively high in old age.

Keywords

Bayesian, longitudinal data, multivariate analysis, ordinal response, zero-inflated Poisson

Full Text (PDF format)