Statistical Modelling 6 (2006), 159–173

Space-cohort Bayesian models in ecological studies

Dolores Catelan
Department of Statistical Sciences,
University of Udine,
Via Treppo 18,
I–33100 Udine
Italy
eMail: catelan@dss.uniud.it

Annibale Biggeri
Department of Statistics 'G. Parenti', University of Florence
and
Biostatistics Unit, CSPO
Florence, Italy

Emanuela Dreassi
Department of Statistics 'G. Parenti', University of Florence
Florence, Italy

Corrado Lagazio
Department of Statistical Sciences, University of Udine, Udine, Italy

Abstract:

The ecological association between ‘low educational level’ and lung cancer mortality, both recorded atmunicipality level, is investigated. Six birth cohortswere retained from 1905 to 1940. Education data were derived from censuses of the period 1921–91. The education score was defined as prevalence of less educated people and was measured on a relative scale, defining a different threshold for ‘low educational level’ at each census. Four potentially relevant ages at first exposure were defined (20, 30, 40, 50) to explore the temporal pattern of the disease. Thus, mortality in each cohort was matched to relative education at different periods corresponding to different ages at first exposure. The relevance of each age at first exposure and the degree of association between education and lung cancer mortality (males, Tuscany, 1971–99) were evaluated, defining a set of hierarchical Bayesian models, each corresponding to a different aetiologic hypothesis. Results show an inverse relationship between low education score and mortality for lung cancer, whose intensity decreases by cohort and becomes positive in the last one. This association was more evident for age at first exposure in the range 20–30 years. These results are consistent with the epidemiological transition of risk factors among socioeconomic classes and are coherent with the biological model of initiating carcinogen agents.

Keywords:

birth cohort; disease mapping; lung cancer; hierarchical space–time Bayesian models; timedependent covariates

Example data and software in zipped archive


back