Affiliations: College of Pharmacy, The University of Arizona,
Tucson, AZ, USA | Department of Mathematics and Statistics, Faculty of
Science and Agriculture, The University of The West Indies, Trinidad and
Tobago
Note: [] Corresponding author: Grant H. Skrepnek, Ph.D., The University of Arizona, Center for Health
Outcomes and PharmacoEconomic Research, 1295 North Martin Avenue, Tucson, AZ
85721, USA. E-mail: [email protected]
Abstract: Cost and outcomes data within pharmacoeconomic analyses often
possess distributional properties that require advanced statistical approaches
to yield robust findings. An analyst's failure to recognize and control for
these characteristics may result in inappropriate evaluations of statistical
associations or causal effects which may ultimately support incorrect policy
decisionmaking. Given the importance of appropriate analysis and interpretation
in pharmacoeconomics, the purpose of this paper is to address the more common
statistical issues encountered in assessing healthcare costs or outcomes,
emphasizing approaches that may be employed to analyze these data. More
specifically, statistical methods used commonly with retrospective cohort
analyses are presented including least squares (e.g., ordinary least squares,
OLS), logarithmic transformations, log-plus-constant models, two-part models,
maximum likelihood estimation (MLE), and generalized linear models (GLM) and
extensions, among others.
Keywords: Regression, retrospective databases, costs and outcomes
