Nội dung

López-Abente et al. BMC Cancer 2014, 14:535 http://www.biomedcentral.com/1471-2407/14/535 RESEARCH ARTICLE Open Access Time trends in municipal distribution patterns of cancer mortality in Spain Gonzalo López-Abente1,2*, Nuria Aragonés1,2, Beatriz Pérez-Gómez1,2, Marina Pollán1,2, Javier García-Pérez1,2, Rebeca Ramis1,2 and Pablo Fernández-Navarro1,2 Abstract Background: New disease mapping techniques widely used in small-area studies enable disease distribution patterns to be identified and have become extremely popular in the field of public health. This paper reports on trends in the geographical mortality patterns of the most frequent cancers in Spain, over a period of 20 years. Methods: We studied the municipal spatial pattern of stomach, colorectal, lung, breast, prostate and urinary bladder cancer mortality in Spain across four quinquennia, spanning the period 1989-2008. Case data were broken down by town (8073 municipalities), period and sex. Expected cases for each town were calculated using reference rates for each five-year period. For map plotting purposes, smoothed municipal relative risks were calculated using the conditional autoregressive model proposed by Besag, York and Mollié, with independent data for each quinquennium. We evaluated the presence of spatial patterns in maps on the basis of models, calculating the variance in relative risk corresponding to the structured spatial component and the unstructured component, as well as the proportion of variance explained by the structured spatial component. Results: The mortality patterns observed for stomach, colorectal and lung cancer were maintained over the 20 years covered by the study. Prostate cancer and the tumours studied in women showed no defined spatial pattern, with the single exception of stomach cancer. The trend in spatial fractional variance indicated the possibility of a change in the spatial pattern in breast, bladder and colorectal cancer in women during the last five-year period. The paper goes on to discuss ways in which spatio-temporal data are depicted in the case of cancer, and review the risk factors that may possibly influence the respective tumours’ spatial patterns. Conclusion: In men, the marked geographical patterns of stomach, colorectal, lung and bladder cancer remained stable over time. Breast, colorectal and bladder cancer in women show signs of the possible appearance of a spatial pattern in Spain and should therefore be monitored. Keywords: Disease mapping, Cancer mortality, Epidemiology, Spatial epidemiology Background New disease mapping techniques widely used in small-area studies enable disease distribution patterns to be identified and have become extremely popular in the field of public health [1,2]. Cancer and other disease mortality atlases [3-5] have shown that many risk factors of a territorial (and thus of an environmental) nature, influence geographical * Correspondence: glabente@isciii.es 1 Environmental and Cancer Epidemiology Unit, National Centre for Epidemiology, Carlos III Institute of Health, Avda, Monforte de Lemos 5 28029, Madrid, Spain 2 Consortium for Biomedical Research in Epidemiology and Public Health (CIBER en Epidemiología y Salud Pública - CIBERESP), Madrid, Spain patterns. New methods of analysis make it possible to smooth selected disease indicators and so reveal their geographical structure [6]. Estimators obtained in less populated areas share information with neighbouring areas, assuming that they are exposed to common environmental factors. Yet, disease incidence and mortality are dynamic processes and therefore variable in space and time. As a result, different spatio-temporal disease mapping techniques have recently been proposed without any broad consensus as to how to describe spatio-temporal disease trends [7-11]. The aim of this study was to report on trends in geographical mortality patterns of the most frequent cancers © 2014 López-Abente et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. López-Abente et al. BMC Cancer 2014, 14:535 http://www.biomedcentral.com/1471-2407/14/535 in Spain across four quinquennia, and in so doing, update the information published in previous studies [12-14]. Methods As our case source, we used individual death entries for the period 1989-2008 corresponding to stomach cancer (International Classification of Diseases, Ninth Revision (ICD-9) code 151, ICD-10 C16), colorectal cancer (ICD- 9 codes 153-154, 159.0, ICD-10 code C18-C21), lung cancer (ICD- 9 code 162, ICD-10 code C33-C34), breast cancer in women (ICD- 9 code 174, ICD-10 code C50), prostate (ICD- 9 code 185, ICD-10 code C61) and bladder cancer (ICD- 9 code 188, ICD-10 code C67). These data were furnished by the National Statistics Institute (Instituto Nacional de Estadística - INE). The observed case data were broken down by town (8073 municipalities) and sex. Municipal populations, with a breakdown by age group (18 groups) and sex, were obtained from the 1991 and 2001 censuses and the 1996 and 2006 municipal rolls, official information provided by INE. These years corresponded to the midpoints of the four quinquennia comprising the study period (1989-1993, 1994-1998, 1999-2003 and 2004-2008). The person-years for each five-year period were obtained by multiplying these populations by 5. To calculate expected cases, the overall Spanish mortality rates for the above four 5-year periods were multiplied by each town’s person-years, by age group and quinquennium. Standardised mortality ratios (SMRs) were calculated as the ratio of observed to expected deaths. For map plotting purposes, smoothed municipal relative risks (RRs) were calculated using the conditional autoregressive model proposed by Besag, York and Mollié (BYM) [6]. This model is based on fitting a Poisson spatial model with observed cases as the dependent variable, expected cases as offset, and two types of random effect terms which take the following into account: a) municipal contiguity (spatial term υi); and b) municipal heterogeneity (νi). α is the intercept quantifying the average mortality rate in all the towns. In this model the linear predictor is:
ηi ¼ log ρi ¼ α þ υi þ vi The smoothed RRs for plotting purposes are (ζ i = exp (υi + νi)). A separate model was fitted for each period. Lastly, in order to have an indicator of the presence of a spatial pattern in the mortality plotted for the respective cancers, we estimated the variance in relative risk corresponding to the structured spatial component and the unstructured component, as well as the proportion of variance explained by the structured spatial component [10]. Page 2 of 15 Integrated nested Laplace approximations (INLAs) were used as a tool for Bayesian inference. For this purpose, we used R-INLA [15] with the option of simplified Laplace estimation of the parameters, a package available in the R environment [16]. A total of 8073 towns were included, and the spatial data on municipal contiguities was obtained by processing the official INE maps. Examples of R scripts describing the models used to obtain the relative risk smoothed maps and the calculation of the components of the variance of spatial terms have been published elsewhere [10]. Results Table 1 shows the number of deaths for the tumours studied, by sex and five-year period, and the trend in age-adjusted rates (European standard population) [17]. Data on the number of deaths are of interest for assessing the figure used to estimate spatial patterns, and the adjusted rates are of interest for assessing the general mortality trend for each tumour. Most of the cancers studied registered a decline in mortality in the last fiveyear period, except for: colorectal cancer in men (which rose 1.2% in 4th versus the 3rd quinquennium); and lung and bladder cancer in women (which rose 2.6% and 0.6% respectively in 4th versus the 3rd quinquennium). Figure 1 shows the situation of the different provinces in Spain. Stomach cancer During the period 1989-2008, there were 144,561 stomach cancer-related deaths in Spain (75,750 in men and 48,811 in women), accounting for 8% of all deaths due to malignant tumours. Figure 2 shows the maps depicting the municipal distribution of stomach cancer mortality for each quinquennium, using the respective reference rates for men and women. The pattern proved similar for both sexes. These maps clearly show that the geographical pattern changed very little over the course of the 20 years. In general, there was a reduction in the number of towns in the highest RR category. When compared to the average for Spain, areas that maintained an excess risk of dying of gastric cancer continued to be large areas of Castile-Leon and towns along the Atlantic coast of Galicia, with the latter registering higher RRs than those recorded for the provinces of Burgos and Palencia. Colorectal cancer During the period 1989-2008, there were 224,305 deaths due to this cause in Spain (123,965 in men and 100,340 in women), accounting for 14.5% of all deaths due to malignant tumours. Figure 3 shows the maps depicting the municipal distribution of colorectal cancer mortality for each quinquennium, using by way of reference the overall mortality in each period among men and women respectively. The geographical pattern was not very López-Abente et al. BMC Cancer 2014, 14:535 http://www.biomedcentral.com/1471-2407/14/535 Page 3 of 15 Table 1 Deaths & age-adjusted rates (x 100,000) by sex, period and cancer site in Spain Cancer site Stomach cancer 1989-1993 1994-1998 1999-2003 2004-2008 Total 1989-2008 Deaths 20176 19596 18367 17611 75,750 Rate 21.16 18.22 15.14 12.83 Deaths 13681 12616 11606 10908 Rate 9.86 8.04 6.59 5.54 Men Deaths 23196 28958 33542 38269 Rate 24.24 26.68 27.03 27.36 Women Deaths 21378 24976 26036 27950 Rate 15.75 16.15 17.79 14.08 Deaths 65853 74395 79704 84398 Rate 68.95 70.04 67.52 64.22 Deaths 6827 8076 9967 13335 Rate 5.35 5.84 6.85 8.64 Deaths 27638 29117 28866 29459 Rate 24.53 23.12 20.27 18.38 Deaths 22310 27029 27998 27618 Men Women Colorectal cancer Lung cancer Men Women Breast cancer Women Prostate cancer Bladder cancer Men Women Rate 23.32 24.27 21.37 18.43 Deaths 13343 14851 16833 18686 Rate 13.81 13.48 13.30 13.07 Deaths 2975 3138 3574 4012 Rate 2.00 1.77 1.75 1.76 pronounced for this cancer and displayed many similarities between the sexes. The most characteristic feature was that the first five-year period in men and the first two five-year periods in women were marked by excess mortality in towns in Catalonia and in the province of Leon, a pattern that became attenuated in subsequent quinquennia. Lung cancer From 1989 to 2008 there were 342,555 lung cancerrelated deaths in both sexes (304,350 in men and 38,205 in women). The municipal distribution of lung cancer mortality in men and women was very different (Figure 4). The geographical pattern in men changed relatively little over the 20 years of study, though the excess observed in the province of Cadiz became attenuated. The areas with highest mortality were the Region of Extremadura, extensive areas of west Andalusia (Huelva, Seville and Cadiz) and towns along the sections of the Cantabrian coast corresponding to Asturias and Cantabria. In women, there was hardly any discernable geographical pattern but some towns in the provinces of Pontevedra and Ourense registered excess mortality across all four quinquennia. Breast cancer Between 1989 and 2008, there were 115,080 deaths due to breast cancer in women in Spain, which accounted for 18% of female cancer-related deaths and 4% of overall 48,811 123,965 100,340 304,350 38,205 115,080 104,955 63,713 13,699 female mortality. There was no unduly pronounced breast cancer mortality pattern (Figure 5 above), indicating that the risk factors were uniformly distributed throughout the territory. During the first two five-year periods, breast cancer in women plotted an already known pattern, marked by towns in Catalonia and the Balearic Isles with higher mortality. Although this excess mortality became somewhat attenuated with time, excess mortality nevertheless emerged across wide swathes of west Andalusia, with towns in Huelva, Seville and Cadiz registering RRs of over 1.10. The lowest mortality was recorded for Galicia, south-west Castile-Leon and east Andalusia. Prostate cancer During the period 1989-2008, there were 104,955 deaths due to prostate cancer in Spain, accounting for 10% of mortality due to all malignant tumours in men. The prostate cancer mortality pattern was not at all pronounced (Figure 5 down), indicating -as in the case of breast cancer among women- territorial uniformity in exposure to possible risk factors. A south-north pattern was in evidence during the first three quinquennia. Galicia was the area having the highest number of towns with excess risk. In Andalusia, which had registered lower than expected mortality in the first three quinquennia, a change seemed to be taking place in the period 2004-2008 in the provinces of Huelva and Cadiz, which began to display RRs of over 1.05 in many towns. López-Abente et al. BMC Cancer 2014, 14:535 http://www.biomedcentral.com/1471-2407/14/535 Page 4 of 15 Figure 1 Geographical situation of Spain’s provinces and Autonomous Regions (Comunidades Autónomas). Bladder cancer From 1989 to 2008, there were 77,412 bladder cancer deaths in Spain, 63,713 in men and 13,699 in women. Although the pattern among men and women was different (Figure 6), with less variability among women, in men the towns with the highest mortality were situated in west Andalusia (Cadiz, Seville, Huelva) and in the province of Barcelona, where excess risk was observed in towns in the Bages district (Suría, Sallent, Balsareny, Manresa and Cardona), which seemed to become attenuated with the passage of time. Bladder cancer mortality was five times lower among women than among men, with the spatial pattern showing a rather singular trend: in west Andalusia, excess mortality among women was confined to the first five-year period and then disappeared; and, while the pattern displayed by women in towns in Catalonia appeared to level off in the 3rd quinquennium, in the period 2004-2008 there was a resurgence in RRs of over 1.10 in Catalonia and in areas lying in the Pyrenean foothills (Pre-Pyrenees). Spatial patterns and changes over time Table 2 shows the variance in the spatial component, the variance in the component of heterogeneity, and the proportion of the variance in mortality explained by the spatial structure of the models. The magnitude of these values in the spatial component in the model indicates the importance of the spatial pattern. Hence, cancers with the most defined patterns have a proportion of variance explained by the highest spatially structured component. This was the case of lung and bladder cancer in men and stomach cancer in both sexes, with variances in excess of 0.05 in the different periods, which accounted for over 70% of the variance. The presence of a change in the spatial pattern can be evaluated by changes in spatial fractional variance over time. Breast, bladder and colorectal cancer in women display a possible change in the spatial pattern in the last five-year period, a change that will have to be monitored by subsequent studies. Additional file 1 includes the following information: 1) Probability density functions showing distribution of smoothed relative risks used in each map and 2) Plots showing the evolution of the variance of model spatial and heterogeneity components and the proportion of the mortality relative risk variance explained by the spatial structure of the models. López-Abente et al. BMC Cancer 2014, 14:535 http://www.biomedcentral.com/1471-2407/14/535 Page 5 of 15 Figure 2 Municipal distribution of relative risk of stomach cancer mortality in men (above) and women (down) for each five-year period, Independent maps for each quinquennium. Spain 1989-2008. Discussion This study describes the mortality geographical patterns trend across four quinquennia, and so updates the municipal mortality maps previously published for the most frequent malignant tumours in Spain. Generally speaking, the mortality patterns observed for the most frequent cancers of the digestive system (stomach and colorectal) and respiratory tract, appear to be stable over time. In women, the distribution of mortality due to the tumours studied showed no defined spatial patterns, with the single exception of stomach cancer. However, time trends in spatial fractional variance enable the presence of changes in the spatial pattern to be assessed. Among the tumours analysed, breast, bladder and colorectal cancer in women suggest the possible appearance of a spatial pattern in the last quinquennium, a circumstance that will have to be monitored by subsequent studies. Prostate cancer mortality displayed no clear geographical pattern. Cancer mortality is a dynamic process that is variable in time and space. Yet, the prolonged latency periods that exist between exposure and disease, mean that changes in incidence and/or mortality indicators may occur gradually over time, and their examination requires the incorporation of instruments of spatio-temporal analysis. In recent López-Abente et al. BMC Cancer 2014, 14:535 http://www.biomedcentral.com/1471-2407/14/535 Page 6 of 15 Figure 3 Municipal distribution of relative risk of colorectal cancer mortality in men (above) and women (down) for each five-year period, Independent maps for each quinquennium. Spain 1989-2008. years, a number of spatio-temporal analysis techniques have been proposed [7-11]. The implementation of these methods is not simple, however, in that their theoretical development is far in advance of that of specific applications. Furthermore, there is no broad consensus as to how to describe spatial and temporal trends simultaneously and appropriately. These techniques’ performance might possibly be very different if they were applied to diseases without long latency periods. In this study we opted to use separate spatial models for each period. Accordingly, each map shares the spatial data, and the temporal data are solely subject to the constraint of selection of reference rates for each period. This decision was taken after exploring other possibilities, such as plotting the maps by periods obtained on the basis of spatio-temporal models having a common reference rate or specific reference rates for each of the four periods. Each of these approaches has its advantages and disadvantages. The best solution is probably to adapt each method of presentation and modelling to the disease being studied, or to draw conclusions from the results of the different methods of analysis. In Spain, cancer, both overall and by principal site (lung, colon-rectum, prostate, stomach, pancreas, breast, uterus, López-Abente et al. BMC Cancer 2014, 14:535 http://www.biomedcentral.com/1471-2407/14/535 Page 7 of 15 Figure 4 Municipal distribution of relative risk of lung cancer mortality in men (above) and women (down) for each five-year period, Independent maps for each quinquennium. Spain 1989-2008. brain, leukaemia, lymphomas and myeloma), is well reflected in death certificates. A complete description of the entire process of certification and registration of deaths in Spain and an assessment of the quality of the information have been published elsewhere [18]. Since the development of a malignant tumour is a slow process, with a long latency period, influenced by many factors, the epidemiology of these tumours cannot be understood without taking into account the interaction between different environmental factors and genetic factors. On the other hand, spatial patterns could reveal differences in risk factors and make a significant contribution to establish policies for the fight against cancer in different areas and population groups. Stomach cancer In Spain (2011), stomach cancer caused 5597 deaths, which accounted for 5.3% of all cancer-related deaths. It is estimated that 7810 new cases were diagnosed in 2012. In Spain, the adjusted incidence rate (European standard population) estimated for 2012 was 16.4 cases per 100,000 population in men (10.8 mortality) and 7.5 in women (4.8 mortality) [19]. In terms of mortality, the rates estimated for this same year were 10.8 per 100,000 population in López-Abente et al. BMC Cancer 2014, 14:535 http://www.biomedcentral.com/1471-2407/14/535 Page 8 of 15 Figure 5 Municipal distribution of relative risk of breast cancer mortality in women (above) and prostate cancer mortality (down) for each five-year period, Independent maps for each quinquennium. Spain 1989-2008. men and 4.8 in women. The small difference between the incidence and mortality rates is due to the low survival recorded for gastric cancer, which is estimated in Spain to be 27.8% at 5 years, a little above the mean relative survival for Europe as a whole (24.1) [20]. Within Europe, the highest incidence is registered by Central and East European countries, while Nordic countries register the lowest rates for both sexes [21]. Until a few decades ago, gastric cancer was the most frequent cause of cancer-related death world-wide. Subsequently, albeit at different points in time, incidence and mortality due to this tumour began to decline in all countries during the second half of the 20th century. Despite this decline, this tumour continues to rank fourth in incidence and the second in mortality [21]. Currently, the highest rates are registered in developing countries, and in East Asia in particular [21]. In Spain, there is a geographical gastric-cancer mortality distribution pattern that is characterised by its persistence in time and, in particular, by its similarity in both sexes. This study shows that the pattern has not changed substantially over the 20 years studied. Although the implication of H. pylori infection in the pattern observed in this country is uncertain, one cannot rule out the implication López-Abente et al. BMC Cancer 2014, 14:535 http://www.biomedcentral.com/1471-2407/14/535 Page 9 of 15 Figure 6 Municipal distribution of relative risk of bladder cancer mortality in men (above) and women (down) for each five-year period, Independent maps for each quinquennium. Spain 1989-2008. of different H. pylori strains which circulate in different regions or which circulated in the first half of the 20th century when the generations currently studied were infected (during their infancy). Furthermore, the marked coast-interior pattern might be associated with different dietary habits in more rural areas of the interior, where more cured/smoked foodstuffs and less fruit and vegetables might be consumed than in coastal areas, or with territorially-related environmental exposures. In order to understand the epidemiology of this tumour the interaction among the factors responsible for the virulence of the micro-organisms implicated, environmental factors and genetic factors [22] have to be taken into account. The great geographical variability in this tumour’s mortality in Spain indicates the need to continue studying the factors implicated in the excess risk that is present in some areas, to enable its frequency to be reduced. Colorectal cancer In Spain, colorectal cancer accounts for 13.7% of cancerrelated deaths in men and 15.8% in women, according to data for the year 2011. Estimates for 2012 put the number of new cases in both sexes at 32,240, and the number of López-Abente et al. BMC Cancer 2014, 14:535 http://www.biomedcentral.com/1471-2407/14/535 Page 10 of 15 Table 2 Variance in spatial and heterogeneity components, and proportion of the variance in mortality relative risk explained by the spatial structure of the models Men Cancer site Women 1989-93 1994-98 1999-2003 2004-2008 1989-93 1994-98 1999-2003 2004-2008 Spatial 0.083 0.067 0.059 0.053 0.091 0.081 0.063 0.063 Heterog 0.001 0.003 0.009 0.013 0.003 0.007 0.001 0.001 sfv 0.99 0.95 0.86 0.81 0.97 0.92 0.98 0.98 Spatial 0.040 0.036 0.033 0.017 0.026 0.021 0.011 0.011 Heterog 0.022 0.012 0.020 0.017 0.003 0.009 0.012 0.001 sfv 0.64 0.76 0.62 0.50 0.88 0.70 0.48 0.91 Spatial 0.099 0.099 0.076 0.063 0.027 0.038 0.037 0.054 Heterog 0.004 0.002 0.001 0.001 0.003 0.005 0.013 0.015 sfv 0.97 0.98 0.98 0.98 0.90 0.89 0.74 0.79 Spatial 0.039 0.028 0.017 0.020 Heterog 0.013 0.008 0.007 0.001 sfv 0.75 0.76 0.70 0.96 Stomach Colorectal Lung Breast Prostate Spatial 0.014 0.013 0.014 0.007 Heterog 0.017 0.012 0.011 0.011 sfv 0.46 0.52 0.56 0.39 Spatial 0.067 0.054 0.043 0.045 0.014 0.004 0.001 0.025 Heterog 0.020 0.001 0.013 0.023 0.002 0.032 0.002 0.002 sfv 0.77 0.98 0.77 0.66 0.88 0.12 0.30 0.93 Bladder Spatial: spatially structured component variance. Heterog: unstructured component variance. Spatial fractional variance (sfv): var.RRspatial/(var.RRspatial + var.RRhet). deaths at 14,700 [19]. The two sites are normally analysed jointly due to frequent errors of classification of tumours of the rectosigmoid junction [18]. Mortality is very high, with this constituting the second leading tumour site among men and women alike. From the second half of the 1990s onwards, mortality and incidence in both sexes were seen to level off [23]. In these tumours, the mortality data do not reflect the true incidence of the disease, since survival has improved in recent years, mainly among the young. In Spain, age-standardised relative survival at five years of diagnosis stands at 53.6% in both sexes, 54.9% if the tumour is located in the colon, and 51.7% if it is situated in the rectum [20]. In most cases, the aetiology of colorectal cancer is unknown. The frequency of these tumours has been linked to economic development, i.e., unlike the remaining tumours of the digestive system (buccal cavity and pharynx, oesophagus and stomach), it is highest in the most developed countries. Incidence and mortality rates in Spain are similar to the average rates for Europe and West European countries, with the highest rates corresponding to East European countries. In 2006, mortality and incidence were substantially lower in Spain than in North European countries, with Spanish rates being below the average for Europe. Among the known aetiological factors is genetic predisposition, which would give rise to the presence of familial polyposis with tumours that more frequently turned malignant. Hereditary factors are present in 10% to 15% of cases. Risk factors described in the literature include higher consumption of meat and animal fats, and lower fibre consumption. In addition, a number of cohort studies, case-control studies and meta-analyses published in recent years have confirmed that smoking increases the risk of developing colorectal cancer [24,25]. Although the results are not very consistent, alcohol consumption has been reported as being a possible risk factor