Time trend and Bayesian mapping of multiple myeloma incidence in Sardinia, Italy

A few reports have described increasing trends and spatial distribution of multiple myeloma (MM). We used a validated database including the 1606 cases of MM diagnosed in Sardinia in 1974–2003 to explore its time trend, and we applied Bayesian methods to plot MM probability by administrative unit on the regional map. Over the 30 years of observation, the MM standardized incidence rate (standard world population, all ages) was 2.17 × 10–5 (95% CI 2.01–2.34), 2.29 (95% CI 2.06–2.52) among men, and 2.06 (95% CI 1.83–2.28) among women. MM incidence increased by 3.3%/year in 1974–2003, in both males and females, particularly among the elderly and in the high incidence areas. Areas at risk tended to cluster in the north-eastern part of the region. A higher proportion of elderly in the resident population, but not socioeconomic factors, nor livestock farming, was associated with higher incidence rates. The steep upward time trend and the spatial clustering of MM suggest interactions between genetic and environmental determinants that might be more efficiently investigated in the areas at risk.


Results
Time trend in MM incidence. In 1974In -2003In , 1606 MM cases occurred among the Sardinian population.
Based on the standard World population, the incidence rate over these three decades was 2.17 × 10 -5 (95% CI 2.01-2.34) for the total population (all ages), 2.29 (95% CI 2.06-2.52) among males, and 2.06 (95% CI 1.83-2.28) among females, with a male/female ratio of 0.95. The crude rate over the study period was 2.73 × 10 -5 (95% CI 2.57-2.89) among the total population, 3.17 (95% CI 2.94-3.40) among males, and 2.30 (95% CI 2.07-2.53) among females. With Poisson regression analysis, the average annual increase in MM incidence was 2.19% (95% CI 1.60 -2.78, p < 0.001). Graphs in Fig. 1 show the trend of MM incidence from linear regression analysis, by gender: the upward trend was similar among the female (0.189 × 10 -5 per year, p = 0.007) and the male population (0.113 × 10 -5 per year, p < 0.001), with a slope 10 times steeper above the age of 65 (0.479 × 10 -5 per year, p < 0.001) than below 65 (0.049 × 10 -5 per year, p < 0.001) (analysis of covariance: F = 307. 37  Geographic map of MM incidence. Figure 2 shows the map of the posterior probability (P) of a MM incidence rate above the critical rate in each of the 356 Sardinian communes, overall and by gender. Seven communes stand out exceeding the 95 th percentile distribution of the likelihood ratio. www.nature.com/scientificreports/ ratio 46.8, P = 0.979), Bitti (11 cases, likelihood ratio 59.2, P = 0.983), Oschiri (11 cases, likelihood ratio 70.0, P = 0.986), Perfugas (10 cases, likelihood ratio 75.2, P = 0.987), and Seneghe (9 cases, likelihood ratio 177.9, P = 0.994). Another 25 communes have a posterior probability ranging 80-94%, based on 3 or more cases. The map shows a tendency of communes with a high posterior probability of an increased MM incidence to concentrate in the north-eastern area of the island, with the low probability areas located in the southern areas and in the two major urban areas, one, Cagliari, in the south, and the second, Sassari, in northwest Sardinia. Supplementary Table 2 shows the age and gender standardized incidence rates aggregated by health district.

Analysis of MM risk factors.
We explored with weighted multivariate regression analysis whether socioeconomic conditions, or size of livestock, or an aging population might have generated the wide geographic variability in MM incidence we observed. As an indicator of socioeconomic conditions, we used the Italian National Institute of Statistics (ISTAT) deprivation index, which combines the following: proportion of the resident population who attained elementary education at most; proportion of the resident population aged 15 years or older searching for an occupation; proportion of the active population engaged in manual work; proportion of rentals over the total residences; persons per room in the household 24 . Independent covariates were male/female ratio of the resident population, deprivation index, presence and size of livestock farms (cattle, sheep, and goats) 25 , and proportion of the population aged ≥ 65 years. We used the commune population sizes relative to the overall regional population as the weights. None of the covariates showed a significant association with MM incidence (Supplementary Table 3).

Discussion
Our results show that in 1974-2003 MM incidence increased in both genders among the Sardinian population. Cancer Registry data confirmed such finding limited to the northern area of the region and the last decade covered by the database we used. There was no further increase in the subsequent years. The steeper slope of the regression line among the elderly suggests that the increasing aging of the Sardinian population and the increasing access of the elderly to specialized medical care over the study period might have contributed to a more frequent diagnosis of the disease along the years. However, the same upward time trend was also observed among the population aged < 65, indicating that other factors might have played a role, possibly interacting with genetic susceptibility to generate the steeper upward time trend we observed in high-risk areas. Seven communes stand out with the highest probability of their rate exceeding that of the overall population of the region, four located in the north-eastern area (Benetutti, Bitti, Oschiri, and Perfugas) and three scattered in the north-west and central areas. These are mainly agricultural areas, six with archaeological remains indicating their origin dating back to prehistory, and one, Arborea, built in 1928 over a reclaiming land, previously covered by marshes in a malaria endemic area. Most of the population hosted in this new city came in the early years of its foundation from Veneto and Friuli, two regions in north-east Italy, and created a flourishing livestock and agricultural crop economy. This town is also known for the large size of its livestock, with two thirds of Sardinian cattle raised in its land. Although we are unaware of publications exploring the association between zoonoses and risk of multiple myeloma among livestock breeders, several papers investigated the association between risk of lymphoma and contact with livestock 26 . However, we did not find a relationship between presence of large cattle farms and MM incidence over the whole region. The analysis of cancer mortality and hospitalisation in areas at risk because of industrial or military settlements, reported an isolated excess of mortality from MM in an urban area in north-eastern Sardinia, and no excess in any of eight areas including major industrial settlements 27 . Our results confirm no excess of MM incidence in the communes surrounding the industrial areas in the Sardinian territory.
The finding of an excess risk associated with having a first-degree relative affected by MM, particularly among men, and African Americans, supports a role of genetic factors 28 . On the other hand, about 17% of MM heritability seems explained by the known gene variants 29 . Besides, based on results from molecular biology studies, aberrant class switch recombination occurring early in the natural history of MM suggests that environmental factors, such as high doses of ionizing radiation, and occupational exposure in the farming and petrochemical industries, might also contribute to increase risk 30 . The DNA damage resulting from environmental exposures would interact with the class switch recombination process to increase the risk of chromosomal translocations, oncogene deregulation, and malignant transformation 30 . In an analysis of MM risk related to occupation, a moderate increase in risk was reported in association with contact with livestock 31 . Also, gardeners and nursery workers combined, but not other farming jobs, metal processors, female cleaners, and occupations with high level exposure to organic solvents showed a moderate increase in risk 32 . Among lifestyle factors, a moderate alcohol intake might would reportedly convey protection 33 . We could not detect a role of deprivation index, an indicator of socioeconomic level, nor did the prevalence of elderly or the male/female ratio or the presence and size of livestock farming affect MM incidence. However, in our ecological analysis, the population of each commune was the unit, and not the individual. Unless the exposed represent a large proportion of the resident population and a strong association exists between the environmental exposure and the disease, the so-called ecological fallacy is likely to occur and to mask possible associations or to generate spurious increases in risk 34 .
We are not aware of genetic investigations aiming to identify the varying prevalence of gene polymorphisms implicated in MM among the Sardinian population. The small town of Arborea, with his peculiar modernist architecture, is home for about 4000 inhabitants, a large fraction of whom preserved their original language, diet, and habits. This population has a different ethnic origin than the rest of the Sardinian population, but it is unclear whether this might be related to the excess incidence of MM therein observed. Nonetheless, the incidence for the resident population (both genders), standardized based on the world population, was 4.6 × 10 -5 , 5.05 among men, and 4.2 among women. The corresponding rates in the Veneto region cancer registry were 4.5 for men The imprecise correspondence with Cancer Registry data might be due to the fact that the database we used includes the cases resulting from an exhaustive active search of the MM diagnoses in the registers of all clinical departments, followed by a double check of the clinical records of each case for the Durie and Salmon diagnostic criteria 36 to come up with the diagnostic certainty of MM. These might not include all the incident cases reported to the Cancer Registries; besides, underdiagnosis of MM might have occurred among the elderly, particularly in the early years of creating the database we used. However, this would affect mostly small villages far apart from the specialized haematology units, located in the major urban centres; still, the elevated risk was mainly observed exactly in small towns, which would contrast this hypothesis.
For the same reason, post-diagnosis relocation of the families seems unlikely to have occurred. The exact address at the time of diagnosis was missing for 58/1606 patients (3.6%); it seems also unlikely that this might have affected the overall pattern.
An advantage of our study is that the diagnoses were all reviewed by the same expert haematologist (GB), thus preventing bias due to the varying diagnostic ability by time and geographic area and minimizing and spreading equally the probability of misdiagnosis over the whole region and along the study period. Although not exactly matching the Cancer Registry data limited to the overlapping period and to the northern part of the island for the reasons explained above, the similar figures calculated from our database confirm the completeness of its records.

Conclusion
Our results describe for the first time increasing time trends of multiple myeloma over several decades among the population of the island of Sardinia. Multiple myeloma incidence increased in both genders, and particularly among the elderly and in high-risk areas. We also observed a clustering of high MM incidence in the northeastern area, which might be of interest for future gene-environment interaction studies, with special focus on agricultural factors, such as use of pesticides, exposure to endotoxin, and contact with livestock and zoonotic agents. We could not identify a role of socioeconomic status, as indicated by the deprivation index, nor of livestock farming. Finally, our results might support the extension of the Cancer Registry coverage to the whole Sardinian population, which would be of paramount importance not only for fostering further research, but also for early detection of risk areas, so to promote effective preventive intervention, and for a rational planning of cancer treatment resources.

Material and methods
A detailed description of the database of haemolymphatic malignancies we used in this study can be found elsewhere 20 . Briefly, it includes 14,744 incident cases of any haematological cancer, in both genders, and at any age diagnosed in the Italian region of Sardinia in 1974-2003. For the purposes of this analysis, we selected the 1606 MM cases, 781 males and 825 females.
For each commune (the smallest administrative unit in Italy), we calculated the total person-years for each gender and age group (0-24, 25-34, 35-44, 45-54, 55-64, 65-74, and 75 +) over the study period, from January 1974 through December 2003. The standardized incidence rate, annual and over the whole study period, of MM was calculated using the 1971, 1981, 1991, and 2001 census data of the regional population as the standard. Census figures were extended four years onwards and five years backwards to estimate the resident population in the intercensal years. The time trend along the study period was explored with the linear regression equation, and with Poisson regression, adjusting by age and gender. To compare results with those from the IARC CI5 volumes, we also standardized the regional, gender specific incidence rates using the standard world population. We used analysis of covariance to test the chance probability associated with the different slope of regression coefficients by gender, by age at diagnosis (below or above 65 years old), and by residence in an area with low vs high probability of MM occurrence, using the median as the cut point 37 .
The statistics to explore the spatial distribution of the probability of MM occurrence have been previously described in detail 38 . Briefly, we used a Bayesian approach, which allowed us to combine information on MM incidence over the island with that from the individual communes with the following equation: where P(η|d,I) is the posterior probability distribution of MM incidence rate η for a given commune, after combining the data d for that commune with those from the whole region. P(η|I) is the prior MM standardized incidence rate, η, from the background regional information; P(d|η, I) is the probability of getting d for the commune assuming η is true, and P(d|I) sets to one the integral of the posterior probability P(η|d,I) over all possible values of η , so to obtain a probability density function.
To detect communes at high risk, we set the critical value of p = 0.001 in the prior probability distribution of MM incidence over the 356 Sardinian communes, and, for each commune and each gender and age subgroup, we calculated the likelihood ratio between the probability of a MM incidence rate higher than the critical probability level and that of a MM incidence rate consistent with what observed at the regional level.
Finally, we plotted on the regional map the probability associated with the likelihood ratio for each commune, using the following colour scale for the area of each commune, based on the quintiles of the probability distribution: white < 0.165, light grey 0.166-0.335, medium-light grey 0.336-0.50, medium-dark grey 0.501-0.80, dark grey 0.801-0.95. The few communes associated with a probability higher than 95% had the darkest black shade.