Academic journal article Demographic Research

Geographical Mortality Patterns in Italy: A Bayesian Analysis

Academic journal article Demographic Research

Geographical Mortality Patterns in Italy: A Bayesian Analysis

Article excerpt

Abstract

In this paper, we present a hierarchical spatial model for the analysis of geographical variation in mortality between the Italian provinces in the year 2001, according to gender, age class, and cause of death. When analysing counts data specific to geographical locations, classical empirical rates or standardised mortality ratios may produce estimates that show a very high level of overdispersion due to the effect of spatial autocorrelation among the observations, and due to the presence of heterogeneity among the population sizes. We adopt a Bayesian approach and a Markov chain Monte Carlo computation with the goal of making more consistent inferences about the quantities of interest. While considering information for the year 1991, we also take into account a temporal effect from the previous geographical pattern. Results have demonstrated the flexibility of our proposal in evaluating specific aspects of a counts spatial process, such as the clustering effect and the heterogeneity effect.

(ProQuest: ... denotes formulae omitted.)

1. Introduction

In the field of public health, the analysis of spatial distribution of mortality and disease rates plays an important role in the formulation of valid aetiological hypotheses that help us to better understand the relationships between events of interest and factors contributing to their occurrence. Moreover, the mapping of the geographical variation in risk may help to identify areas with lower or higher rates of incidence with respect to a reference average. This can be useful in enabling us to better plan and manage the resources of a public health system (for an in-depth review on disease mapping methods and applications see Marshall 1991a and Lawson et al. 1999, 2000).

In recent years, there has also been considerable interest in these kinds of studies from a demographic perspective, and a great deal of effort has been put into developing and applying statistical methods which take into account mortality information about small geographical areas (see for example Benach et al. 2003, 2004; Caselli and Lipsi 2006).

Within this framework, statistical results become more or less pertinent depending on the degree to which the areas display homogenous environmental, demographic, social, and economic characteristics, as these characteristics may influence the phenomenon of interest. In this way, the average value of the mortality incidence observed with respect to any area effectively represents the risk that the majority of the corresponding population is exposed to. Moreover, the phenomenon of interest may be more correctly connected to possible determinants.

In geographical studies of mortality, such a fine specification of the territorial domain, alongside the need to consider data that is also distributed over several factors (e.g., sex and age), clashes with the challenge of dealing with events that are becoming increasingly rare. As a consequence, the territorial distribution of the mortality risks appears strongly influenced by the random variation enclosed in the observable data.

The most common solutions adopted to deal with these problems involve two main approaches. The first consists of stabilising the statistical estimates through the introduction of multiple-years averages (Caselli and Egidi 1980, Pickle et al. 1996, Frova et al. 1999, Jougla et al. 2002, Dupré et al. 2004, De Simoni and Lipsi 2005, Egidi et al. 2005). Underlying this approach is the strong assumption that the territorial units are independent of each other. Moreover, this approach also neglects the possibility of considering the temporal dynamic of the phenomenon within the period considered.

The second approach, which was developed more recently, considers geographical smoothing through the use of spatial interaction models. These kinds of methods were first developed in the fields of spatial epidemiology and disease mapping, both within a classical framework (see for example Breslow and Clayton 1993 for a generalisation of the linear mixed models approach, Ferrandiz et al. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.