Leukemias are the most frequent childhood neoplasm worldwide, corresponding to approximately one third of all malignancies in this age group. Acute lymphocytic leukemia (ALL) is the most common subtype.1 Low incidence rates are usually observed in developing countries, but in Brazil, rates are similar to those found in developed regions (4.1/100,000 and 3.6/100,000 for males and females, respectively).2
Acute leukemia is a progressive clonal disorder determined by mutations. The majority of the chromosomal and molecular alterations which are found in leukemic cells are limited to the leukemic clone, suggesting that these changes are acquired.3 Although the etiology of childhood leukemia remains undefined, research studies have established a causal association with ionizing radiation,4 Down syndrome5 and for acute myeloid leukemias, exposure to chemotherapeutic agents.6 Other factors such as parental occupation,7, 8 magnetic fields,9 pesticides,10, 11 vaccination history,12, 13 breastfeeding,14–16 dietary factors,17–20 parental smoking,21 chemical exposures,22 and genetic susceptibility23–25 have also been described as risk or protective factors.
Infections have long been implicated as possible etiologic factors for childhood leukemias. The theory proposed by Greaves3 assumes that delayed exposure to common infections results in a deficient immune system and a consequent higher risk of common childhood leukemias, particularly ALL. Kinlen26 has suggested that through population mixing, i.e., influxes of people to previously isolated communities, nonimmune children are exposed to infectious agents and consequently are at an increased risk of ALL. Population mixing, as a result of the residential mobility, has been utilized as a proxy of a mixing of infections, since crowded and fast-growing communities have high prevalence of infections.27
Over the last years, the interest in assessing social inequalities and health has increased substantially. Socioeconomic characteristics have been associated with morbidity and mortality discrepancies in many developed countries.28–30 The relationship between social inequalities and cancer has been well studied for adults31 but less extensively for childhood cancer. Childhood leukemia seems to be unique in this aspect, as earlier it was reported that childhood leukemia was more frequent among individuals of low socioeconomic status (SES).32
A comprehensive review on the association between SES and childhood leukemia was recently published and the authors have pointed out that this association is likely to vary according to time, place and study design.33 There are more positive associations (i.e., higher incidence rates among high SES) in older studies and negative associations in newer ones. Overall, studies carried out in Europe have shown a positive association between childhood ALL and SES, while North-American studies reported negative ones. Registry-based and record-based studies generally presented positive associations, whereas interview-based case–control studies showed negative associations.33, 34 The authors concluded that more studies are necessary, including studies with several types of individual and ecological SES measures, which could help to evaluate the strength of this association.33, 34
We propose to evaluate the association between SES and childhood leukemia, using population data from São Paulo, Brazil. We will also investigate the impact of surrogate measures of childhood infections on the development of the disease.
Materials and methods
We collected data on children aged less than 15 years diagnosed with ALL between 1997 and 2002 from the population-based Cancer Registry of the São Paulo, Brazil. The diagnosis of ALL was confirmed microscopically for 96% of the children. This registry covers the geographic area of the city of São Paulo, capital of the state of São Paulo, located at the upland of Serra do Mar mountain range, 860 m above mean sea level. The city area is 1,509 km2 and the childhood population (2,715,519 people under 15 years of age) represents almost 30% of the city population.
Each child of the study was assigned to a socioeconomic group, based on the Social Exclusion Index35 for the district of residence at the time of diagnosis. The 96 administrative districts of São Paulo were divided into 4 socioeconomic groups (A, B, C and D from the richest to the poorest) on the basis of Social Exclusion Index (Fig. 1). This index was built on the basis of 3 dimensions: decent standard of living, knowledge and youth vulnerability. The first dimension includes a poverty indicator (percentage of household heads with insufficient income in each district), an employment indicator (percentage of people older than 10 years working legally) and an inequality indicator (ratio of the percentage of family heads with monthly income > US 1,000 to the percentage of family heads with monthly income ≤ US 1,000). Second dimension comprises 2 indicators: literacy (percentage of population ≥5 years of age that is able to read and to write) and years of study (mean number of years of study for family head), and the last dimension includes a youth indicator (percentage of the population ≤19 years) and a violence indicator (homicides rate/100,000 habitants). All indicators were obtained from the 2000 Brazilian Demographic Census. The standardization of these indicators to proper indices utilized the same methods applied for the Human Development Index (HDI), where a raw variable, say X, is transformed into a unit-free index between 0 and 1 (allowing that different indices can be added), according to the following formula:
p = indicator under study (poverty, literacy, etc.); i = unit of analysis (district, city, state, country, etc.); Xi = value of the indicator used for calculation; MIN (Xi) = minimum value found in the indicator's distribution. MAX (Xi) = maximum value found in the indicator's distribution.
The population size (children under 15 years of age) of districts ranges from 2,186 (Barra Funda) to 104,351 (Grajaú). Twenty-three districts have less than 10,000 children, 34 have population size 10,001–25,000 children, 25 districts have a population size between 25,001 and 50,000 children and 14 districts have more than 50,000 children. Approximately 70% (26/39) of the districts with a population larger than 25,000 children are classified as low SES (Group D). Furthermore, 85% of these districts (33/39) have more than 50% of their population with a monthly household income less than 350 dollars.
Using the district of residence at diagnosis, cases also were classified according to the percentage of households with 7 or more persons in their district, using 2000 Brazilian Census information. This percentage ranged from 0.7% (Jardim Paulista) to 10.2% (Marsilac).
Population density (habitants/km2) for each district was derived from census data (2000) and 5 categories were created based on quintiles (< 5,000, 5,000–8,219, 8,220–11,500, 11,501–14,499 and ≥14,500). Population growth (percentage/year) was also divided into quintiles (≤ −1.27, −1.26 to −0.58, −0.57 to 0.31, 0.32 to 1.34, ≥1.35).
Age-specific (0–4, 5–9 and 10–14 years) as well as age-standardized incidence rates (ASIR), stratified by gender and social exclusion level, were calculated by the direct method using the 1960 world standard population. Rate ratios (RR) with corresponding 95% confidence intervals were estimated.
Chi-square test for linear trend was used to verify the existence of a gradient in risk according to SES levels and Spearman's rank correlation coefficient was employed to test the correlation between SEI and household crowding. For all statistical tests, an alpha error equal to 5% was fixed.
A total of 574 cases of ALL were reported to the São Paulo Cancer Registry from 1997 to 2002. Sixty-seven cases (11.7%) were excluded because of missing address. Male-female ratio was 1.3:1. As expected, a peak was observed in the 0- to 4-year age group (38.1% of all cases). Crude incidence rates were 3.64/100,000 and 2.81/100,000 for males and females, respectively. Age-adjusted incidence rate was 3.68/100,000 for males and 2.87/100,000 for females. For both genders the highest rates were found in the district of Santo Amaro (27.3 and 28.3/100,000, for males and females, respectively). Lowest rates were found in the districts of Jardim São Luís (0.87/100,000 for males) and Campo Limpo (0.69/100,000 for females). During the study period, no new cases of ALL were registered among children living in the following districts: Anhanguera, Barra Funda, Belém, Marsilac, Perus, Vila Guilherme and Vila Leopoldina.
Children living in areas with high indexes of social exclusion (SEI 0.0–0.4) presented a significantly lower risk of ALL compared with those living in the most wealthy districts (SEI 0.6–1.0) (RR = 0.34; 95% CI 0.28–0.44) (Table I) and the same effect was observed for males and females.
Table I. Age-Standardized Incidence Rates of All Per 100,000 Children (0–14 Years) and Rate Ratios (RR) According Social Exclusion Strata
When the analysis was restricted to children aged 2–5 years, the decrease in risk of ALL for those living in areas of high social exclusion (low SES) was even more significant (Group B–RR = 0.33, 95% CI 0.24–0.61; Group C-RR = 0.27, 95% CI 0.21–0.49; Group D-RR = 0.28, 95% CI 0.24–0.48; chi-square for linear trend = 29.98; p < 0.001). Lower incidence rates of childhood ALL were also found in those districts with high percentages of households with 7 or more persons, achieving a 70% reduction in risk for children living in areas in which this percentage was ≥5.7% (RR = 0.32, 95% CI 0.26–0.43) compared to areas with a proportion ≤2.2% (Table II).
Table II. Age-Standardized Incidence Rates of All Per 100,000 Children (0–14 Years) and Rate Ratios (RR) According Percentage of Households with 7 or More Persons
The analysis of variation in childhood ALL incidence by population density did not demonstrate risk differences in the different categories for total population and males only. However, in the female group, a statistically significant decreasing trend in leukemia incidence is observed concurrent to an increase in population density (p-value for linear trend = 0.023) (Table III).
Table III. Age-Standardized Incidence Rates of All Per 100,000 Children (0–14 Years) and Rate Ratios (RR) According Population Density
A strong correlation between SEI and crowding (rho = −0.95, p < 0.001) was also observed.
Several studies have shown evidence of an association between childhood leukemia and high SES.36–41 Overall, studies utilizing area-based socioeconomic measures have demonstrated an increased risk of ALL among people with high SES.32, 42 In Yorkshire, England, an ecological analysis performed on 248 children (<15 years) revealed a marked reduction in incidence of ALL for the highest quintile of deprivation (risk ratio = 0.67, 95% CI 0.44–1.01).43
On the other hand, the results of the studies of SES and childhood leukemia using individual-level assessment are controversial.38–40, 44–48 High levels of family income and parental education, measured individually, have been consistently associated with a lower risk of childhood leukemia while the association of paternal occupational class with childhood leukemia demonstrates a contrary association, i.e., high rates are correlated with high SES,49–51 including findings from 2 cohort studies.52–53 A recent case–control study conducted in United Kingdom did not show any difference in childhood ALL risk according to deprivation levels, whether using area- or individual-based measure of SES (father's occupation), at the time of birth or diagnosis.54 On the basis of their findings, the authors suggest that SES in the United Kingdom does not have influence on the development of ALL in children and that the previous findings could be artefactual.54
There is also no unanimity regarding the advantages or shortcomings of using composite scores versus single-variable SES indicators.55 In United Kingdom, 2 area-based deprivation measures have been widely used: the Carstairs-Morris index56 and the Townsend index.57 The first one combines percentages of crowding, unemployment, no car ownership and home rental, while the second specifies male unemployment and adds the percentage of low social class. More recently, the Public Health Disparities Geocoding Project has developed an area-based composite measure based on United States Census data, which comprises occupational class, income, poverty, wealth, education, and crowding.55 The social exclusion index used in our study used a method similar to that applied by the Human Development Index (HDI),58 which has been used worldwide since 1990.
The lower risk of ALL among children living in low SES areas found in the current study is further corroborated by the inverse association between ALL incidence and the percentage of households with 7 persons or more and also by the strong correlation between the latter variable and the Social Exclusion Index. Two previous studies have shown similar results: a population-based historic cohort study carried out in Northern Ireland from 1971–1986 demonstrated a strong inverse association between household density and the risk of ALL, when comparing children born into households with ≥1 person/room compared to those born into less crowded households (adjusted Relative Risk = 0.56, 95% CI = 0.35–0.91).53 Previously, Dockerty et al.46 described an association between ALL and household crowding although results were not statistically significant for the highest category of crowding, when adjusted by age, sex, and other variables (child's social class, mother's marital status, mother's education, child's infection with influenza in the first year of life, age child was when solid food was first introduced).
Recently, Poole et al. pointed out the difficulties in making quantitative comparisons between studies, since many different types of SES measures were utilized and their distinct social implications can vary by place and time.33 Berkman and Macintyre reported that it is usually assumed that SES measures that produce extreme gradients are the best ones. In fact, the adequacy of the measure is also dependent on the study aims and availability of information in each particular country.59
Our finding of lower childhood ALL rates among females in areas of higher population density is not compatible with results reported by Muirhead in 1995 for 3 metropolitan areas of United States: San-Francisco-Oakland (CA), Detroit (MI) and Atlanta (GA), who described a statistically significant increasing trend in childhood leukemia incidence rates concurrent with an increase in population density (RR = 1.4 95% CI 1.0–2.0 comparing highest to lowest category of total population density).60 However, a geographical analysis of the childhood ALL cases from north-west England has demonstrated a monotonic relationship between incidence rates and population density, with a higher risk in more densely populated areas.61 Furthermore, we can not ignore the fact that São Paulo is one of the most densely populated cities in the world (6993 habitants/km2)62 and even the lowest quintile (<5,000 hab/km2) encompass districts with higher densities than cities like Bangkok, Berlin, Toronto and Los Angeles. In addition, the level of urbanization in São Paulo is very high (93.5% for the city): 78% of the districts are completely (100%) urban.62
The inverse association between ALL incidence and population growth observed in our study is compatible with results reported in Ontario, Canada, where ALL risk for children 0–4 years decreased with increased levels of population growth in urban areas.63 Similar findings were also described by Kinlen et al., when analyzing mortality from leukemia under age 25 in British New Towns 1946–1985.64
We are aware that an “ecological fallacy” can not be ruled out in analyses of this type, since individual assessment of SES was not performed and the smaller units analyzed (administrative district) could be too large to represent a neighborhood effect. Epidemiologic studies have demonstrated that the performance of area-based socioeconomic measures, based on census tracts, is comparable to individual-based ones,55, 65–68 but results were not consistent for postal code.55, 65
Another limitation was the chance of selection bias, since it could be assumed that missing addresses correspond more frequently to cases of children living in poor areas. However, when the analysis included cases with missing addresses, assuming that all of them belonged to the lowest SES, the direction and magnitude of the association were unchanged (data not shown). Furthermore, in a Canadian study, Borugian et al. have noted the possibility of the underreporting cancer cases in the poorest neighborhoods, which they believe could be mitigated by the existence of a universal health care system and also by a cancer registry with high coverage.42 Since 1988, Brazil has had a universal health care system (SUS-Sistema Único de Saúde) and the city of São Paulo has 18 health care facilities (public or non-profit hospitals and clinics), which help oncologic patients through this system.69 Furthermore, the population-based cancer registry of São Paulo, which has acceptable quality indices as defined by the International Agency for Research on Cancer (IARC), and has been included in the last edition of the Cancer Incidence in 5 Continents.
Our study using surrogate measures of childhood infection (i.e., SES and household density) supports the hypothesis of an infectious etiology for childhood ALL. Greaves stated that the peak incidence of common ALL at 2–5 years of age, observed in regions with high SES, could be explained by a “delayed infection” hypothesis, in which an anomalous or deregulated immunological response to post-natal infection, would generate secondary genetic events and leukemia development.70–71 The presumed agent (s) could be bacterial or viral and the mechanism of action is indirect or non-transforming, but until now, no particular microbial agent has been implicated. It is well known that the opportunity and timing of infection is influenced by individual factors (breastfeeding, birth order, vaccinations) as well as community factors (mobility, population density, socioeconomic context). Therefore, it is anticipated that the conduct of additional studies will enhance the understanding of the relationship of genetic and environmental factors, and the identification of environmental factors that will provide a basis for reduction of childhood leukemia risk.
The authors gratefully acknowledge the Population-based Cancer Registry of São Paulo, especially to Dr. Antônio Pedro Mirra, for the collaboration.