Bayesian Change-Point Mixed Models Applied to Data on Outpatient Antibiotic Use in Europe
Girma Minalu1, Niel Hens1,2, Marc Aerts1,4, Samuel Coenen3, Arno Muller3, Niels Adriaenssens3, Philippe Beutels2, Geert Molenberghs1,4, Herman Goossens3
1Interuniversity Institute for Biostatistics and Statistical Bioinformatics (I-BIOSTAT), Hasselt University, Belgium; 2Centre for Health Economics Research and Modeling Infectious Diseases, Centre for the Evaluation of Vaccination (WHO & VAXINFECTIO Collaborating Centre), Antwerp, Belgium; 3Laboratory of Medical Microbiology, Vaccine & Infectious Diseases, Institute (VAXINFECTIO), University of Antwerp, Belgium; 4Interuniversity Institute for Biostatistics and Statistical Bioinformatics (I-BIOSTAT), Catholic University of Leuven, Belgium

Introduction: Resistance to antibiotics is a major public-health problem and antibiotic use is being increasingly recognized as the main selective pressure driving this resistance. Quarterly data on outpatient antibiotic use are available from 25 European countries for the period 1997-2008. Antibiotic use was measured as defined daily doses per 1000 inhabitants per day (DID). The main objective is to develop an appropriate statistical model to assess the significance of country-specific trends over Europe and to identify change-points, while accounting for country-specific global use as well as seasonal effects. The application of the model yields new important insights in the evolution of outpatient antibiotic use over several European countries.

Methods: Change-point models have previously been used in several applications to model longitudinal data (see e.g. Ghosh et al., 2007; Dominicus et al., 2008). In this paper, an adaptive Bayesian linear spline model is proposed, where the number of knots (change-points) and their location are data-driven and determined by the deviance information criterion (DIC). Random effects for the global level of use, the trend effects, the amplitude of the seasonal effect, and the location of the change-point are used to account for heterogeneity across countries.

Results: Application of the proposed adaptive model indicates that the data support a model with two unknown fixed change-points. The first change-point in the trend of antibiotic use in DID is 17 (first quarter of 2001) and the second change-point is 30 (second quarter of 2004). The Bayesian highest posterior density credible intervals for the change-points are given by (11.97, 21.86) and (27.01, 33.68), respectively. There is an overall decrease in the trend of outpatient antibiotic use in DID before the first change-point (slope=-0.0278), and a slight decrease in the trend after the first change-point (slope=-0.0059). There is no significant difference in the trend of DID before and after the second change-point. The location of the change-points may be related to points in time where public-health strategies were initiated, aiming to increase the awareness of the public to a more rational use of antibiotics or aiming to reduce overconsumption of antibiotics.


Dominicus, A., Ripatti, S., Pedersen, N. L., and Palmgren, J. (2008). A random change point model for assessing variability in repeated measures of cognitive function. Statistics in Medicine, 27, 5786-5798.

Ghosh, P. and Vaida, F. (2007). Random changepoint modelling of HIV immunological responses. Statistics in Medicine, 26, 2074-2087.

Keywords: Change-point mixed model; Adaptive linear spline model; DIC

Biography: My name is Girma Minalu, and I am a PhD student at the Interuniversity Institute for Biostatistics and Statistical Bioinformatics (I-BIOSTAT), Hasselt University, Campus Diepenbeek, Belgium.