Secondary Logo

Journal Logo

Original Article

Childhood Leukemia and Socioeconomic Status in Canada

Borugian, Marilyn J.*; Spinelli, John J.*; Mezei, Gabor; Wilkins, Russell; Abanto, Zenaida*; McBride, Mary L.*

Author Information
doi: 10.1097/01.ede.0000164813.46859.63



Click on the links below to access all the ArticlePlus for this article.

Please note that ArticlePlus files may launch a viewer application outside of your web browser.

Early studies of childhood leukemia have reported a higher incidence with higher socioeconomic status (SES).1–3 In more recent case–control studies of extremely low-frequency magnetic fields and childhood leukemia, however, cases tended to be of lower SES than controls, suggesting a possible inverse association between SES and childhood leukemia.4–6

The inverse association observed in these recent epidemiologic studies may represent a real association between SES and childhood leukemia, which may have shifted over time from the previously observed positive association. If this shift is real, then it might help to identify an etiologic pathway for a disease that is poorly understood. It is also possible, however, that a shift to an inverse association between SES and childhood leukemia may result from bias due to either a nonrepresentative source of controls or differential participation of cases and controls by SES.7–9 A positive association would also be more consistent with the hypothesis of viral etiology, which suggests that early exposure to viral agents, potentially associated with lower SES, could be protective for acute lymphoid leukemia.10

In addition, if SES is a risk factor for childhood leukemia, regardless of the direction of the association, then SES could also behave as a confounder. Many epidemiologic studies of leukemia routinely control for one or more aspects of SES without a clear rationale. In this study, we analyzed age-standardized incidence rates of all childhood leukemia cases throughout Canada in relation to SES using population-based quintiles of neighborhood income adjusted for household size.


Childhood Leukemia Cases

All incident leukemia cases in children (ages 0–19 years) diagnosed from 1 January 1985 through 31 December 2001 were included in the study. Cases were identified from population-based provincial cancer registries. In Canada, the provincial cancer registries use multiple sources of notification, including reports from hematology and pathology laboratories, cancer treatment centers, and provincial death certificate data. The provincial registries cover at least 95% of all Canadian cancer cases.11 The following cancer registries participated: British Columbia, Alberta, Saskatchewan, Manitoba, Ontario, Quebec, New Brunswick, Nova Scotia, Prince Edward Island, and Newfoundland and Labrador. We selected 1985 as the initial year to minimize the problem of missing postal codes and underreporting of cases before that time. One province, New Brunswick, was unable to provide data for 1985–1988. We selected 2001 as the final year to ensure completeness of reporting for the total period. One province, Quebec, was unable to provide data for 2001. Population denominators were adjusted to remove data for these province and year combinations in which cases were not reported.

The leukemia subtypes were categorized according to the International Childhood Cancer Classification (ICCC), 1996 revision.12 This classification is based on diagnostic morphology and topography codes of the second edition of the International Classification of Diseases for Oncology (ICD-O-2)13 with some additions for new codes. The ICD-O-2 site/morphology codes that make up each diagnostic group are given in Appendix I, available with the electronic version of this article. We used conversion programs from the National Cancer Institute's Surveillance, Epidemiology and End Results web site14 to convert from ICD-O-1 or ICD-O-3 coding to ICD-O-2 coding. In cases in which there was inconsistent coding, morphology took precedence generally, but we took ICD-9 coding over morphology when ICD-9 coding was provided.

No personal identifiers were used, and the only individual-level data used were already present in the cancer registries. The study was approved by the joint research ethics boards of the University of British Columbia and the British Columbia Cancer Agency. Additional approvals were obtained from each provincial cancer registry, as well as from the Manitoba Research Ethics Board and the Nova Scotia Research Ethics Board.

Data Sources and Included Variables

Population data by census year (1986, 1991, 1996, 2001), sex, and 5-year age group (0–4, 5–9, 10–14, 15–19) were obtained from Statistics Canada for the smallest geographic area for which Canadian population data are released: enumeration areas for 1986–1996 and dissemination areas for 2001. Both kinds of areas have a typical population of approximately 400 persons and roughly correspond to U.S. Census block groups. For noncensus years, we used the population data for the closest census year (eg, for 1994–1998, the 1996 census data were used).

For each cancer case, we obtained the following data from the cancer registry records: age at diagnosis, year of diagnosis, histology (or morphology), site (or topography), sex, and postal code of place of residence at the time of diagnosis.

Neighborhood Income Quintiles

Neighborhood income quintiles were defined for enumeration or dissemination areas according to methods developed at Statistics Canada. Quintile values were determined for each census during the study period, as detailed subsequently. We ascertained the postal code of the subject's usual place of residence at the time of diagnosis and assigned the neighborhood quintile value derived from the nearest census. Income deciles were also constructed and analyzed, with results similar to those for quintiles; results for deciles are not reported here due to the small size of some categories and wider confidence intervals.

Using Statistics Canada postal code conversion software (PCCF+ Version 3J),15 the postal code of the subject's residence at diagnosis was linked to the appropriate 1996 census enumeration area. Additional files were used to determine the corresponding 1991 and 1986 census enumeration areas, and 2001 census dissemination areas, based on nearest centroids of those areas with respect to the 1996 enumeration area centroids. Neighborhood income data were obtained from the census nearest in time to the event.

Neighborhood income quintiles were based on the average income per single-person equivalent in the enumeration area or dissemination area. This measure uses the person-weights implicit in the Statistics Canada low-income cutoffs to derive “single-person equivalent” multipliers for each household size. For example, for 1996, a single-person household received a multiplier of 1.0, a 2-person household received a multiplier of 1.25, and a 3-person household received a multiplier of 1.55. The total income of the enumeration area or dissemination area (average household income times the number of households) was then divided by the total number of single person equivalents, yielding income per single-person equivalent. This is a way of adjusting for household size, because more sophisticated variables (such as the percentage of population under the low-income cutoff) are not available at the enumeration area or dissemination area level.

Quintiles of population by neighborhood income were constructed within each area (census metropolitan area, census agglomeration, or residual areas within each province) and then pooled across areas. The reason for creating the quintiles within each area is that housing costs vary enormously across Canada. Rents and house prices in some places (such as most of Quebec and the Atlantic provinces) have historically been much lower than those in Toronto or Vancouver. Quintiles calculated by area have revealed greater disparities than quintiles by national standards.16

For rural postal codes and for urban postal codes of outlying suburban and rural areas, the same postal code is generally used for multiple enumeration areas or dissemination areas (average number of areas = 3; range = 2–11). The selection of a single such area for coding purposes is random but with probabilities respecting the proportions of population with that postal code in each of the possible small areas. Thus, the coding is far less precise than for centralized urban postal codes, which are usually only linked to a single enumeration area or dissemination area. For this reason, we also performed supplementary analyses of the data excluding rural and small-town areas (with community population <10,000, where rural postal codes predominate) from the numerator and denominator.

A total of 801 cases had only the first 3 digits of the 6-digit Canadian residential postal code available, thereby yielding less precise estimates of income quintile. Excluding them did not affect our results, and so they are included in all results presented here.


Of 5411 reported cases, we excluded 171 (3.2%). Reasons for exclusion were missing values for postal code (n = 106), missing values for income quintile (n = 51), or residential postal code referring to a hospital, school, or university residence (n = 14). This left 5240 cases (96.8%) for analysis.

Statistical Analysis

Using Orius 98 Client software,17 we calculated incidence rates for each ICCC category, 5-year age group, sex, calendar period (1985–1988, 1989–1993, 1994–1998, 1999–2001), and neighborhood income quintile. Age-standardized incidence rates (ASIRs) were computed by the direct method with the Canadian 1991 population as the standard. Rate ratios (RRs) with 95% confidence intervals (CIs) based on the Poisson distribution were computed from the ASIR for each income quintile relative to the highest quintile. Tests for heterogeneity and trend were calculated using the likelihood ratio test from Poisson regression.18 Differences in income quintile RR across sex, age, and calendar period were tested using interaction terms in the Poisson regression model.


The distribution of leukemia cases by diagnostic group, sex, and age is shown in Table 1. As expected, a higher incidence of leukemia was seen among younger children, with a peak in the 0- to 4-year age group, and incidence was slightly greater among boys than girls. Three fourths (77%) of the cases were ICCC group Ia (acute lymphoid leukemia).

Distribution of Incident Cases of Childhood Leukemia, by Age at Diagnosis, Sex, and Diagnostic Group

For children diagnosed with acute lymphoid leukemia, the rate ratio for the poorest neighborhood income quintile compared with the richest was 0.86 (95% CI = 0.78–0.95) (Table 2). Risk estimates for acute nonlymphocytic leukemia and for chronic myeloid and other leukemias were slightly weaker and less precise. When the data were restricted to urban areas (community population >10,000), the risk in the poorest quintile was reduced slightly (eg, 0.83 [0.74–0.93] for acute lymphoid leukemia).

Age-Standardized Incidence Rates (ASIRs) of Childhood Leukemia per 100,000 Population and Rate Ratios (RRs) by ICCC Group and Income Quintile

No evidence for an interaction was observed between calendar period and income quintile. As shown in Figure 1, the leukemia risk tended to increase with income for each of the time periods. We found one interaction between sex and income quintile; for chronic myeloid and other leukemias, boys in the poorest income quintile were at somewhat lower risk but the girls were not.

Time trends of incidence of (A) all leukemia and (B) acute lymphoid leukemia.


This study examined the association between an SES surrogate, neighborhood income, and incidence of childhood leukemia using data on more than 96% of reported Canadian cases for 1985–2001. We found a slightly lower risk of childhood leukemia in the poorest neighborhood income quintile compared with the richest. These findings do not support the idea of a shift over time to higher risk among the lower SES groups.

A comprehensive review by Linet et al19 summarizes potential causal factors of childhood cancer. The report mentions SES in reference to only one cancer type (soft tissue sarcomas) and even then as a factor with limited evidence. Some studies have examined the association of SES and childhood cancers, but few have done so based on large geographically defined populations. This is due in part to the low risk of childhood cancer and the need for population-based cancer registries of good quality and wide coverage. Dockerty et al,20 in a large case–control study conducted in England and Wales, reported an adjusted odds ratio of 0.91 (0.75–1.1) for acute lymphoblastic leukemia among children in the poorest quintile compared with the richest. These results are similar to ours, both in direction and magnitude. An earlier British ecologic study used an amalgamation of census small area-based measures of social class and economic status21 and found an increased incidence of childhood leukemia with higher SES.

Several well-designed studies and a few reviews have considered SES as a potential source of bias or confounding in studies of the association of childhood leukemia with other risk factors such as traffic density,22 birth characteristics,23 peak age of onset,24 and residential exposure to magnetic fields.7,25–27 Reynolds et al took advantage of the statewide population-based California Cancer Registry to conduct 2 large case–control studies examining the role of exposure to traffic density22 and birth characteristics23 in the incidence of childhood cancer, particularly leukemia. Although the risk of leukemia in the latter study appeared elevated in children with more highly educated parents, this weakly positive association was not confirmed in the study of traffic density and neighborhood median family income of birth residence. Reynolds concluded that SES of the neighborhood of birth is not a sufficiently strong risk factor for leukemia to explain the effect of another variable such as wire codes. Our results, based on neighborhood income for the place of residence at diagnosis, were also weakly positive and unable to explain the effect of other stronger variables.

The very large geographically defined area of our study and the very high rate of coverage are strengths. The use of an ecologic measure of income, however, means that confounding factors (such as residential mobility or ethnicity) may account for some or all of the observed results. There could also be selective underreporting of cancer cases in the poorest neighborhoods, although the provincial registries cover at least 95% of all Canadian cancer cases.11 Universal health care in Canada should reduce or eliminate a case-finding or treatment effect that might bias studies done under other types of healthcare systems, especially in light of more frequent total physician and hospital visits among the poor in Canada.28,29 However, little research has been done to characterize unregistered cancer cases with respect to SES. Finally, there could be a real association between risk of childhood leukemia and some exposure related to neighborhood income. One possible example of such an exposure is early contact with infectious agents, which is thought to confer some protection and might be more prevalent in neighborhoods with lower income and poorer living conditions.30,31 Smith et al31 reported that early exposure of the mother may be protective for the development of acute lymphoid leukemia in the child by preventing in utero exposure when there is a lack of protective maternal antibodies. More recently, Ma et al32 reported that day care attendance was associated with decreased risk of acute lymphoid leukemia; such an association may be explained by the protective effect of early exposure to common infections among day care attendees.

Although neighborhood income is a good indicator of SES, the results may not generalize to other SES measures based on individual data. In a study of one Canadian province, Mustard et al33 examined the measurement validity of ecologic measures as proxies for individual-level measures. They reported that, contrary to their hypothesis, risk estimates derived from ecologic income measures were not attenuated relative to risk estimates from individual measures of household income, thus providing evidence supporting the use of ecologic-level income measures. They also contrasted the results of neighborhood-level and household-level income measures in urban and rural populations, finding no consistent evidence that neighborhood-level estimates of risk were attenuated relative to household-level estimates of risk for rural populations.

In a review of socioeconomic measures in U.S. public health research, Krieger et al34 reported that neighborhood-based measures of social class merit greater use because they can be used for persons of all ages and may capture aspects of living conditions not measurable at the individual level. In our study, a census small area-based measure of social class allowed us to construct population-based incidence rates by neighborhood income level, because our population denominators were also census-based and could be classified in the same way. Krieger also reported that use of neighborhood-level measures is more likely to underestimate socioeconomic effects than to overestimate them, and validity may be improved by using the smallest and most homogeneous census groups available (like we did using census enumeration and dissemination areas).

In summary, our results showed a moderately lower risk of childhood leukemia among children in the poorest neighborhood income quintile, mostly limited to acute lymphoid leukemia, in a population-based study group.


We are grateful for the helpful support provided by the provincial cancer registry representatives: Alberta (Heather Bryant, Douglas Dover), British Columbia (Wendy Robb), Manitoba (Erich Kliewer, Jeri Kostyra), New Brunswick (Christofer Balram, Suzanne Leonfellner), Newfoundland and Labrador (Bertha Paulse, Susan Ryan), Nova Scotia (Maureen MacIntyre, Ron Dewar), Ontario (Eric Holowaty, Carole Herbert), Prince Edward Island (Dagny Dryer, Kim Vriends), Quebec (Michel Beaupré), and Saskatchewan (Diane Robson, Karen Robb). We are also grateful to the University of British Columbia Data Services staff for assistance with population data. We thank Robert Semenciw of the Surveillance and Risk Assessment Division of Health Canada for his generous contribution of time, expertise, and technical knowledge of the Orius software. Special thanks to Kim McGrail for her invaluable assistance in the early stages of the study.


1. Greenberg RS, Shuster JL Jr. Epidemiology of cancer in children. Epidemiol Rev. 1985;7:22–48.
2. Linet MS. The Leukemias: Epidemiologic Aspects, vol 6. New York: Oxford University Press; 1985.
3. Linet MS, Cartwright RA. The leukemias. In: Schottenfeld D, Fraumeni JF, eds. Cancer Epidemiology and Prevention, 2nd ed. New York: Oxford University Press; 1996:841–892.
4. Linet MS, Hatch EE, Kleinerman RA, et al. Residential exposure to magnetic fields and acute lymphoblastic leukemia in children. N Engl J Med. 1997;337:1–7.
5. McBride ML, Gallagher RP, Theriault G, et al. Power-frequency electric and magnetic fields and risk of childhood leukemia in Canada. Am J Epidemiol. 1999;149:831–842.
6. Schuz J, Grigat JP, Brinkmann K, et al. Residential magnetic fields as a risk factor for childhood acute leukaemia: results from a German population-based case–control study. Int J Cancer. 2001;91:728–735.
7. Hatch EE, Kleinerman RA, Linet MS, et al. Do confounding or selection factors of residential wiring codes and magnetic fields distort findings of electromagnetic fields studies? Epidemiology. 2000;11:189–198.
8. Spinelli JJ, McBride ML, Abanto ZU, et al. Assessing response bias as an explanation for the observation of low socioeconomic status (SES) as a risk factor for childhood leukemia. Am J Epidemiol. 2001;153:
9. Selection bias in epidemiologic studies of EMF and childhood leukemia. Proceedings of the 2001 Whistler workshop. Technical Report 1008149. Palo Alto, CA: Electric Power Research Institute; 2003.
10. Mezei G, Kheifets LI. Clues to the possible viral etiology of childhood leukemia. Technology. 2002;9:3–14.
11. Parkin DM, Whelan SL, Ferlay J, et al., eds. Cancer Incidence in Five Continents. Lyon, France: IARC; 1997:VII.
12. The new classification. In: Kramarova E, Stiller CA, Ferlay J, et al., eds. International Classification of Childhood Cancer. Lyon, France: IARC; 1996.
13. Percy C, Van Holten V, Muir C, eds. International Classification of Diseases for Oncology, 2nd ed. Geneva: World Health Organization; 1992.
14. SEER conversion programs [National Cancer Institute web site]. Available at: Accessed March 28, 2004.
15. Wilkins R. PCCF+ Version 3J User's Guide. Automated Geographic Coding Based on the Statistics Canada Postal Code Conversion Files, Including Postal Codes to May 2002. Ottawa: Health Analysis and Measurement Group, Statistics Canada; 2002:61.
16. Wilkins R, Berthelot JM, Ng E. Trends in mortality by neighbourhood income in urban Canada from 1971 to 1996 (Catalogue 82-003-SIE). Health Reports. 2002;13(suppl):45–71.
17. On L, Semenciw RM, Mao Y. Orius software: calculation of rates and epidemiologic indicators, and preparation of graphical output. Chronic Dis Can. 2000;21:134–136.
18. Breslow NE, Day NE. Statistical Methods in Cancer Research, Volume II: The Design and Analysis of Cohort Studies. Lyon, France: IARC; 1987.
19. Linet MS, Wacholder S, Zahm SH. Interpreting epidemiologic research: lessons from studies of childhood cancer. Pediatrics. 2003;112:218–232.
20. Dockerty JD, Draper G, Vincent T, et al. Case–control study of parental age, parity and socioeconomic level in relation to childhood cancers. Int J Epidemiol. 2001;30:1428–1437.
21. Alexander FE, Cartwright RA, McKinney PA, et al. Leukaemia incidence, social class and estuaries: an ecological analysis. J Public Health Med. 1990;12:109–117.
22. Reynolds P, Von Behren J, Gunier RB, et al. Residential exposure to traffic in California and childhood cancer. Epidemiology. 2004; 15:6–12.
23. Reynolds P, Von Behren J, Elkin EP. Birth characteristics and leukemia in young children. Am J Epidemiol. 2002;155:603–613.
24. Swensen AR, Ross JA, Severson RK, et al. The age peak in childhood acute lymphoblastic leukemia: exploring the potential relationship with socioeconomic status. Cancer. 1997;79:2045–2051.
25. Gurney JG, Davis S, Schwartz SM, et al. Childhood cancer occurrence in relation to power line configurations: a study of potential selection bias in case–control studies. Epidemiology. 1995;6:31–35.
26. Savitz DA, Poole C. Do studies of wire code and childhood leukemia point towards or away from magnetic fields as the causal agent? Bioelectromagnetics. 2001;Suppl 5:S69–S85.
27. Wartenberg D. The potential impact of bias in studies of residential exposure to magnetic fields and childhood leukemia. Bioelectromagnetics. 2001;Suppl 5:S32–S47.
28. Dunlop S, Coyte PC, McIsaac W. Socio-economic status and the utilisation of physicians’ services: results from the Canadian National Population Health Survey. Soc Sci Med. 2000;51:123–133.
29. Veugelers PJ, Yip AM. Socioeconomic disparities in health care use: does universal coverage reduce inequalities in health? J Epidemiol Community Health. 2003;57:424–428.
30. Kinlen LJ. Infection and childhood leukemia. Cancer Causes Control. 1998;9:237–239.
31. Smith MA, Simon R, Strickler HD, et al. Evidence that childhood acute lymphoblastic leukemia is associated with an infectious agent linked to hygiene conditions. Cancer Causes Control. 1998;9:285–298.
32. Ma X, Buffler PA, Selvin S, et al. Daycare attendance and risk of childhood acute lymphoblastic leukaemia. Br J Cancer. 2002;86:1419–1424.
33. Mustard CA, Derksen S, Berthelot JM, et al. Assessing ecologic proxies for household income: a comparison of household and neighbourhood level income measures in the study of population health status. Health Place. 1999;5:157–171.
34. Krieger N, Williams DR, Moss NE. Measuring social class in US public health research: concepts, methodologies, and guidelines. Annu Rev Public Health. 1997;18:341–378.

Supplemental Digital Content

© 2005 Lippincott Williams & Wilkins, Inc.