Statistical analysis of temporal and spatial trends of zoonotic agents in animals and food Part I: Critical review of the statistical analysis carried out on the Community Summary Report 2006 data [1]
The Task Force wishes to thank the members of the Working Group for the preparation of this
report: Marc Aerts, Giusi Amore, Billy Amzal, Pierre-Alexandre Beloeil, Frank Boelaert,
Annamaria Conte, José Cortiñas, Renata Del Rosario, Hanne-Dorthe Emborg, Klemens
Fuchs, Uriel Kitron, Alessandro Mannelli, Giovanna Nicolini, Monica Pratesi, Francesca
Riolo, Robin Sayers, Tobias Verbeke, and Therese Westrell.
No abstract available
The European Union (EU) Member States (MSs) collect data on zoonoses, zoonotic agents, antimicrobial resistance and food-borne outbreaks. The European Food Safety Authority (EFSA) is assigned the tasks of examining and analysing the data collected and preparing the Community Summary Report (CSR).
Statistical analyses were carried out for the first time on the 2006 data. The objective was to evaluate the significance of temporal variations in the EU-level prevalence of zoonotic agents in animals and in the EU-level proportion of positive food units. Indeed, trend analysis provides information on developments in the Community and Member States and it may give information on the effects of Community or national control measures to reduce the occurrence of zoonotic agents.
A critical review of the statistical analysis of time trends in 2006 was carried out. Two critical characteristics of the data, which can affect the validity of the analysis, were identified: 1) varying sampling probability of epidemiological units among MSs and across years in the same MS, corresponding to disproportionate stratified sampling, where MSs are considered as strata; 2) correlation among observations on zoonotic agents in the same MS in subsequent years.
The SURVEYLOGISTIC procedure in the SAS system, which was used in the 2006 analysis, is specifically suited to analyse survey data taking into account sampling design. In order to account for disproportionate stratified sampling, weights were calculated, per MS and year, as the reciprocal of the sampling ratio (number of units in the MS population, divided by number of sampled units). The choice of the population to be used to calculate weights is critical since weighting may strongly affect parameter estimates and statistical significance of trends. Moreover, finite population correction could be applied if considered as appropriate (design-based approach).
The CLUSTER statement was used to take into account correlation among observations from the same MS. Consequently, the standard error for the parameter corresponding to the effect of year on the probability of a positive result is inflated and the probability of statistical significance reduced. In SURVEYLOGISTIC, the CLUSTER statement is specifically aimed to deal with a design aspect, namely the random selection of clusters. Therefore, in the absence of a specific sampling design, other techniques, such as generalised estimating equations (GEE, REPEATED statement in the GENMOD procedure) would be more appropriate. However, from a practical point of view, the approach to the analysis of time trends used by EFSA was valid, and results are the same or very similar to those obtained by using GEE. By using techniques for correlated data, the effects of between MS harmonisation problems were reduced, and the analysis was valid as long as harmonisation was constant within MSs, across years.
Further developments of trend analysis are introduced. The Bayesian approach, for example, is a flexible and powerful set of tools in the analysis of time trends, where model parameters are not assumed to be a fixed unknown constant to be estimated, but instead they are seen as random variables. The distribution of parameters (called posterior distribution) quantifies all uncertainty and information present in the model and the data, but also possibly from other prior sources of knowledge available (defining prior distribution of parameters).
Investigating the spatial and temporal distribution of zoonotic agents and diseases, performing specific spatial and temporal analyses is a powerful approach in the study of temporal and spatial trends of zoonoses. Therefore, spatial and space-time clustering analyses of zoonoses is an important field of further development carried out by EFSA. A review of the most commonly used statistical methods in spatial epidemiology is included, together with data needs for meaningful spatial analysis at EU level.
Results of complex statistical analysis and certain details on the methods need to be communicated to risk managers and the general public. In a section of this report, basic principles on how to communicate in a comprehensible way are presented, and a worked-out example shown.
In a part II report the use of different software and statistical packages (including open source) will be compared in the analysis of temporal trends. Moreover, the analysis of spatial trends in zoonotic agents will be developed through a case study.
zoonoses, statistical analysis, trend analysis, spatial analysis, data analysis
