Municipal distribution of breast cancer mortality among women in Spain.(Research article)(Clinical report)

Authors: Marina Poll�n (corresponding author) [1,2]; Rebeca Ramis [1,2]; Nuria Aragon�s [1,2]; Beatriz P�rez-G�mez [1,2]; Diana G�mez [1]; Virginia Lope [1,2]; Javier Garc�a-P�rez [1,2]; Jose Miguel Carrasco [1]; Maria Jos� Garc�a-Mendiz�bal [1,2]; Gonzalo L�pez-Abente [1,2]

Background

Breast cancer is the leading malignant tumour in women, accounting for 27% of cancers in European women [1]. Spain has one of the lowest rates in Europe, in terms both of incidence (estimated age-standardised rate of 51 per 100,000 population) more especially, of mortality (16 per 100,000)[1]. The major risk factors seem to be genetic susceptibility, reproductive behaviour, obesity and, less consistently demonstrated, diet[2]. Ecological studies have shown an association with fertility, body weight or fat consumption, but these variables explained only a minor component of the overall variation[3]. A positive association between breast cancer and socio-economic level has been consistently reported [4, 5, 6, 7, 8, 9, 10]. Although socio-economic status can be a surrogate of several risk factors (reproductive behaviour, diet or physical activity) [11, 12, 13], these variables do not completely explain the excess risk observed in the more affluent groups[8].

One of the classic approaches in epidemiology is the study of geographical distribution. In administrative terms, Spain is divided into Autonomous Regions known as Comunidades Aut�nomas. These are in turn divided into provinces and, at the lowest level, into municipalities. Breast cancer mortality has been previously studied at a provincial level. The highest standardised rates observed in Las Palmas Province (Canary Islands) were double those registered for Orense, the province with the lowest rate[14].

Currently, spatial epidemiology allows for a greater level of disaggregation. One of the advantages of this approach is to highlight local effects that might be linked to specific geographic, social or environmental characteristics[15]. This study reports on municipal distribution of breast cancer mortality in Spain and the variability associated with socio-economic level and other explanatory variables. Furthermore, given that pre- and postmenopausal tumours have somewhat different risk factors (i.e. obesity seems to act as protective exposure in younger women and is a well-established risk factor among postmenopausal women), the geographical pattern is independently explored in women aged under 50 years or 50 years and over.

Methods

As our case source, we used all Spanish individual death entries for the period 1989-1998 corresponding to breast cancer (International Classification of Diseases, 9th Revision (ICD-9) code 174) broken down by municipality. A municipality is an administrative unit, made up of a clearly demarcated territory and its population, governed by a municipal council. This is the smallest aggregated division for which mortality and population data could be obtained at a national basis for the study period. Mortality data were furnished by the National Statistics Institute. Municipal populations, broken down by age group (18 groups) and sex, were obtained from the 1991 census and 1996 municipal rolls. These years correspond to the midpoints of the two quinquennia that comprise the study period (1989-1993 and 1994-1998). The person-years for each five-year period were obtained by multiplying these populations by 5.

As an indicator of socio-economic level, the index provided by the Spanish Credit Bank (Banco Espa�ol de Cr�dito) for 1991 was used[16]. This index classifies municipalities into 10 levels, according to different markers of economic activity, namely, the number of holiday homes, bank branch offices and telephones, and estimated average family income. An indicator of rurality was drawn up, based on the number of inhabitants, as classified by the National Statistics Institute in the following 10 categories: > 500,000; 100,001-500,000; 50,001-100,000; 20,001-50,000; 10,001-20,000; 5,001-10,000; 2,001-5,000; 1,001-2,000; 501-1,000; and 101-500 and <100 inhabitants. Finally, the percentage of people over the age of 64 years living in each municipality was deemed to be a surrogate of life-style factors that might have changed and were less prevalent in older generations.

Standardised mortality ratios (SMRs) were computed as the ratio of observed to expected deaths. Expected cases were computed, taking Spanish breast cancer mortality rates, broken down by age and five-year period, as reference. SMRs were also calculated by province and category of explanatory variable, and confidence intervals for these categories were duly computed using Byar's approach[17].

Breast cancer mortality was separately studied among women aged under 50 years and comprised deaths from cancers diagnosed mainly in premenopausal women and among women aged 50 years and over. This latter group was made up of a mixture of pre- and postmenopausal cases but was nevertheless dominated by the second group.

Smoothed municipal relative risks (RRs) were calculated using the conditional autoregressive model introduced by Clayton and Kaldor[18], and further developed by Besag, York and Molli� [19]. This model has been applied in the field of ecological studies[20]. It is a Poisson spatial model with observed cases as the dependent variable, expected cases as offset, and two random effects terms that take the following into account: a) municipal contiguity (spatial term); and b) municipal heterogeneity. Socio-economic level, rurality and percentage of people over the age of 64 years were introduced into the model as continuous explanatory variables. The purpose was twofold: 1) to ascertain their influence on breast cancer mortality; and 2) to smooth relative risk, taking into account the variability associated with these factors rather than merely the spatial relationship among municipal areas. The model took the following form

[math omitted]

where:

[lambda]i is the relative risk in area I; Oi is the number of deaths in area I; Ei are the expected cases; [beta]j is the coefficient representing the effect (log(RR)) of the explanatory variable J, xji refers to the value of the explanatory variable J in the area I, hi is the municipal heterogeneity term; and bi is the spatial term.

Models were fitted using Bayesian Markov Chain Monte Carlo simulation methods[21]. Posterior distributions of RR were obtained using WinBugs[22]. The criterion of contiguity used was adjacency. Convergence was verified using the BOA (Bayesian Output Analysis) R programme library[23]. Given the great number of parameters, the convergence analysis was performed on a randomly selected sample of 10 towns and cities, taking 4 strata defined by municipal size. …