Incidence trends and projections for childhood cancer in Ontario


  • Mohammed Agha,

    1. Pediatric Oncology Group of Ontario, Toronto; ON, Canada
    Search for more papers by this author
  • Bruna DiMonte,

    1. Pediatric Oncology Group of Ontario, Toronto; ON, Canada
    Search for more papers by this author
  • Mark Greenberg,

    Corresponding author
    1. Pediatric Oncology Group of Ontario, Toronto; ON, Canada
    2. Division of Haematology/Oncology, The Hospital for Sick Children, Toronto, ON, Canada
    3. Department of Paediatrics, University of Toronto, Toronto, ON, Canada
    • 480 University Avenue, Suite 1014, Toronto, ON, M5G1V2, Canada M5G1V2
    Search for more papers by this author
    • Fax: +416-592-1285.

  • Corin Greenberg,

    1. Pediatric Oncology Group of Ontario, Toronto; ON, Canada
    Search for more papers by this author
  • Ronald Barr,

    1. Pediatric Oncology Group of Ontario, Toronto; ON, Canada
    2. Service of Hematology–Oncology, McMaster Children's Hospital, Hamilton, ON, Canada
    3. Department of Pediatrics, McMaster University, Hamilton, ON, Canada
    Search for more papers by this author
  • John R. McLaughlin

    1. Division of Epidemiology and Biostatistics, Samuel Lunenfeld Research Institute, Toronto, ON, Canada
    2. Department of Public Health Sciences, University of Toronto, Toronto, ON, Canada
    Search for more papers by this author


Studies of cancer incidence patterns and trends can provide useful measures of health burden and possible disease etiology, which can aid the planning of cancer care services. This report aims to characterize trends in incidence of childhood cancer, and to assess the implications of these trends by generating incidence projections to 2015. Cancer incidence data were obtained from the database of the Pediatric Oncology Group of Ontario (POGO), which has registered all cancer cases in Ontario since 1985. Annual incidence rates were calculated with census-based population estimates for the 1986–2001 period. Poisson regression models were used to analyze trends, and to calculate projected numbers of cases up to the year 2015. From 1986 to 2001, 5,163 cancer cases occurred among children aged 0–14. Leukemia, CNS tumors and lymphomas were the most common cancers. The number of incident cases increased by 14%, from 296 in 1986 to 336 in 2001. For all cancers, average annual age-standardized rates increased from 147 per million in 1991 to 157 per million in 2001. Over the next 15 years, the 0–14 year population is expected to decrease from 2.28 million in 2000 to 2.13 million in 2015. A marginally statistically significant trend in incidence was projected for all cancers combined (0.5% increase per year p < 0.10) and a statistically significant increase for lymphomas, (1.2% per year 95% CI = 0.0–3.9%). During this period, the number of cases of leukemia and CNS tumors is expected to remain relatively stable. The number of cases of all cancers is expected to increase by 8%, from the average of 320 in 1995 to ∼347 in 2015. Understanding of these projections will facilitate health care resource planning. © 2005 Wiley-Liss, Inc.

Although mortality rates for childhood cancer have declined steadily over the past 3 decades,1 cancer remains the most common cause of disease-related death among children in Canada.2 Despite its rarity and the gains achieved in terms of mortality reduction, cancer is an important health problem among children, because of the severity of the disease and the immediate consequences of its treatment, and also because of the long-term sequelae.3

Descriptive epidemiologic studies of incidence patterns and trends provide useful measures of the health burden and potential impact on health care delivery because of childhood cancer, which can aid the planning of cancer care services. In addition, these studies may guide investigations of disease etiology, if cancer trends are found to be associated with a concomitant change in risk factors. This report aims to establish the nature of trends in incidence rates and the number of new cases of childhood cancer in the province of Ontario, Canada, and to assess the implications of these trends by generating incidence projections up to the year 2015. Trends and projections are considered for all cancers combined, and for the 3 most common types of cancer among children: leukemias, central nervous system (CNS) tumors and lymphomas. By estimating the future caseload of childhood cancer in the population, this article demonstrates a process that can guide resource allocation and other considerations in health care delivery.

Material and methods

Cancer incidence data for Ontario were obtained from the database of the Pediatric Oncology Group of Ontario (POGO). Ontario is the most populous province in Canada, with an estimated total population of 12.3 million in 2003.4 Health care is available to all residents through a comprehensive government-sponsored health insurance program. POGO is an organization funded by the provincial government to develop and operate a co-ordinated system of childhood cancer services. As part of this mandate, POGO has maintained a specialized database since 1985, which refers to children with cancer, who are treated at the province's 5 pediatric oncology centers. We have demonstrated previously that this database captures 96% of children, 0–14 years of age.5

Information obtained from the POGO database consisted of the total number of incident cancer cases newly registered at POGO centers annually from 1986 through 2001, among children residing in Ontario. While the POGO database captures data on children aged 0–17, the focus of this report is on children 0–14 years of age for 2 reasons. First, the reports on childhood cancer usually focus on cases aged <15 or <20 years, and second, an evaluation of the quality of incidence data indicated under-reporting of cases in the 15–17 year age group in the POGO database, due to some adolescents receiving care outside pediatric centers.6 Data were available by age group (0–4, 5–9, and 10–14), sex, year of diagnosis and type of cancer. Diagnoses were classified according to the standard international classification scheme for childhood cancer,7 which is based on the morphological classification system developed by the Manchester Children's Tumor Registry.8 Data were obtained for all cancers combined and for the leading individual types of cancer among children, which are the leukemias (class I: mostly acute lymphoblastic leukemia), lymphomas and CNS tumors (class III: mostly astrocytomas, primitive neuroectodermal tumors, ependymomas, other gliomas and other intracranial and intraspinal tumors). Although the revised Manchester system refers primarily to malignant tumors, some neoplasms of uncertain behavior are included in the system, in particular, in group III (CNS tumors) and group II (lymphoma and other reticuloendotheliosis), because the behavior of several childhood tumors, identified morphologically as benign, is more compatible with malignant biology.

Annual incidence rates were calculated with census-based and intercensal population estimates (1986–2001) developed for Ontario by Statistics Canada. Annual projected populations up to 2015 for the 5-year age groups were developed by Statistics Canada, using medium assumptions of fertility, mortality and migration. With the focus being a detailed depiction of patterns as seen at pediatric oncology centers across the province, this study referred to an age range from 0 to 14 years, and employed Ontario's population in a census year (1991) as the basis for age-standardization of rates. To place the cancer trends for Ontario in a context that enabled international comparisons, estimates were recalculated by using the Standard World Population9 to calculate age-standardized rates.

To reduce the effect of random variations on age-adjusted incidence trends, we used a 5-year moving average technique. Based on this approach, the moving average in each year is based on an averaging incidence, in that year, 2 years after that and 2 years prior to it. To reduce the impact on tail years, we used a “weighted” moving average. Instead of simple averaging of these 5 years, we assigned higher weights to the years closer to the index year. Highest weight was assigned to the index year (0.5). The year immediately before and after received a weight of 0.2 and the next 2 years 0.1. For the years 1987 and 2000, we used 0.5, 0.25 and 0.25 for the index year and the years prior and after it. For the tail years, we only used the rate for the index year.10

Poisson regression models were used to analyze trends and to calculate projected numbers of cases, based on adaptations of previously described age-period modeling approaches.11, 12, 13, 14, 15 This statistical model applies to count data of rare events, whereby the annual number of incident cases in each age group is assumed to be distributed as an independent Poisson variable. The model was log E(dj− Yr) = log (nj− Yr) + α + βjXj + γYr, where dj− Yr, is the number of events in each of the 4 (j) age groups in each calendar year (Yr = year − 1986), E(dj− Yr) is its expected value and n is the population size (used as an offset in the model).16 The regression analysis estimated the disease rate [exp(α)] at baseline (i.e., the youngest age group in the first year] and the rate ratios (RR) associated with age group [exp(βj)] and year [exp(γ)]. Model fitting and tests of significance and goodness of fit were performed using SAS. Tests of significance were based on the likelihood ratio test (χ2 distributed), and Wald 95% confidence intervals (CI) were calculated. The Year variable was modeled initially as a simple continuous variable, and then greater complexity was considered by testing the significance of terms that represented a quadratic form, and an interaction with age. Gender was also included in the initial regression model, and it was a statistically significant factor for all cancers combined, but due to the small sample size issue for data analysis at the level of specific cancer types, gender was not considered in the model.

In addition, the method used by Smith et al.17 in the analysis of CNS tumor rates was adapted to assess whether there was a change-point in the trend estimates. The model used was the “jump-model,” based on the assumption of a step increase in early 1990s. This regression method fits 2 trend lines to the incidence data for the 2 time periods (i.e. before and after 1990). Unlike Smith's model, we did not assume 2 flat trend lines (independent of time) before and after the jump point. We also used Joinpoint statistical software18 to examine the jump model and compared the results between 2 models.

The trends for all sites combined, leukemia and lymphoma were best represented by the simple linear trend variable, whereas for CNS tumors, the best fit was achieved by adding a term that indicated a change in 1990. The expected number of cases in future years was obtained by calculating the projected age-specific incidence rates from the best fitted regression model for each cancer type and for all cancers combined, and then multiplying by age-specific (in 5-year age groups) and year-specific population projections.


In Ontario, from 1986 to 2001, 5,163 cancer cases occurred among children aged 0–14 (Table I). The most common childhood cancers were leukemia, CNS tumors and lymphomas, which accounted for 32, 23 and 10% of all cases, respectively. The number of incident cases seen at pediatric oncology centers increased from 296 in 1986 to 336 in 2001, an increase of 14% over 15 years. The number of cases of each specific tumor type increased over time with wide fluctuations annually for leukemia and lymphoma. Simultaneously, the population (aged 0–14) increased steadily from 1.9 to 2.3 million (an increase of 19%). Rates declined with increasing age for both leukemia and CNS tumors, whereas lymphoma rates increased with age (Table II). For all cancers combined, average annual age-standardized rates (5 years moving average) increased from 148.5 per million children in 1991 to 158.2 per million in 2000.

Table I. The Number of New Cases for Leading Types of Childhood Cancer and Populations in Ontario, Canada, 1986–2001, by Year and by Age Group (0–14)
 Number of new casesPopulation (millions)
All cancersLeukemiaCNSLymphoma
  1. CNS, central nervous system tumors.

Year of diagnosis
Age group (years)
Table II. Age–Standardized Incidence Rates (Per Million) for Leading Types of Childhood (Ages 0–14) Cancer in Ontario, by Year of Diagnosis (Standardized to the 1991 Ontario Population)
YearAll cancersLeukemiaCNSLymphoma
AnnualMoving avg
  1. CNS, central nervous system tumors.


To enable international comparisons, rates in Table II were recalculated by age adjusting using the Standard World Population (Table III). The age-standardized rates for ages 0–14 based on the Standard World Population are slightly higher than are the rates shown in Table II, because the world standard assigns greater weight to the younger age groups that have relatively higher rates.

Table III. Age–Standardized Incidence Rates (Per Million) for Childhood Cancer in Ontario (All Sites), for Ages 0–14, using 2 Different Standard Populations (The Ontario 1991 Population and the Standard World Population1), by Year of Diagnosis
YearRate standardized to:
Ontario 1991 populationStandard world population
  • 1

    The proportion of the population in the 0–4, 5–9, and 10–14 year age groups, respectively, are 34%, 34% and 32% for Ontario in 1991, and 39%, 32% and 29% for the World Standard Population.


For all cancers combined, the incidence rate in the baseline year (1986) and for the baseline age group (ages 0–4) was estimated by the Poisson regression model (Table IV) as 209 cases per million (95% CI = 195–223). Age was statistically significantly associated with incidence (95% CIs excluded 1.0), with rates being highest for the 0–4 year age group, about half that value for ages 5–14 (RR = 0.5). The model indicated that there was a 0.5% average annual increase in incidence for all cancers combined, with marginal statistical significance (p < 0.10). Incidence rates for leukemia and CNS tumors in the 0–4 year age group were statistically significantly higher than for older children, whereas the opposite was true for the lymphomas (Table IV). For leukemia, there was no statistically significant annual trend as the RR for “year” was very close to 1.0, but for lymphoma the model indicates a 1.9% annual increase in incidence. The best fitting model for CNS tumors indicated that rates increased significantly in 1990 (RR = 1.37; 95% CI = 1.12–1.68), and after accounting for this shift there was no further indication of an annual trend. This was confirmed by the Joinpoint model. Figure 1 shows the comparison of trend observed based on the 2 models.

Figure 1.

Trend in age-adjusted incidence rates for CNS tumors in Ontario 1986–2001. Results from two models: regression and Joinpoint model.

Table IV. Childhood Cancer Incidence Rate in Ontario at Baseline (Per Million Children, Ages 0–4 Years, in 1986), RR Estimates and CIS Obtained by Poisson Regression Analysis, by Type of Cancer
Type of cancerBaseline rate for ages 0–4 in 1986Rate ratio
Year (annual trend indicator)Period (shift indicator for 1990–2001)Age group
Ages 5–9Ages 10–14
  • CNS, central nervous system tumors; NS, not significant.

  • 1

    Values in parentheses indicate 95% confidence intervals.

All cancers208.6 (195.4–222.7)11.005 (0.999–1.011)NS0.55 (0.51–0.58)0.52 (0.48–0.55)
Leukaemia77.9 (69.5–87.1)1.002 (0.992–1.013)NS0.49 (0.44–0.55)0.31 (0.27–0.36)
CNS35.1 (30.1–40.9)0.986 (0.968–1.004)1.371 (1.12–1.68)0.87 (0.76–0.99)0.76 (0.66–0.88)
Lymphoma5.7 (4.3–7.5)1.019 (1.000–1.039)NS2.05 (1.56–2.69)3.40 (2.64–4.38)

Goodness of fit statistics indicated that each model fitted the data adequately, even for the lymphomas, which can be seen in Figure 2 to have the widest relative variability in rates (e.g., goodness of fit χ2 test = 44.53 on 44 degrees of freedom, p = 0.86). For both CNS tumors (after accounting for the shift that occurred in 1990) and leukemia, the relative stability of rates is apparent in Figure 2, whereas all cancers and lymphoma rates increased slightly, but steadily.

Figure 2.

Actual (1986–2001) and projected (2002–2015) age-standardized incidence rates (per million) for cancer among children (aged 0–14) in Ontario, for all sites combined and for leading types of cancer (standardized to the 1991 Ontario population).

Projected estimates of age-adjusted incidence rates obtained from the models were applied to the change in population expected over the next 13 years, which indicated that the number of cases of all cancer types is expected to demonstrate a slight decline from the observed value of 349 in 2000 to ∼341 (95% CI = 282–409) in 2010, and then, start increasing thereafter to about 351 (95% CI = 290–418) in 2015 (Fig. 3).

Figure 3.

Trends in the age-standardized incidence rate (per million; standardized to the 1991 Ontario population) and the number of cases for all types of cancer, among children (aged 0–14) in Ontario, showing actual (1986–2001) and projected (2002–2015) values.

Over this period, the population in the 0–14 year age range is expected to decrease from 2.28 million in 2000 to 2.11 million in 2012 and to rise again to 2.13 million in 2015. The decrease in the population, which is expected to be mainly in the 0–4 year age group, could explain the pattern observed for the projected number of cases for all cancers and specific cancer types in the next 13 years. While the incidence rate for leukemia is expected to rise from 49 per million in 2000 to 51 (95% CI = 34–68) per million in 2015, the projected numbers of leukemia cases are expected to decrease from 116 observed cases in 2000 to 103 expected in 2005 and 105 (95% CI = 72–142) cases in 2015. Since leukemia is the most frequent cancer type in children and 0–4 is the most common age group affected by this cancer, it is anticipated that the decrease in the population will play the more significant role in the expected number of cases of leukemia.

The projected number of CNS tumors (based on the jump model accounting for shift in 1990) is expected to decrease from 86 in 2001 to 65 and 60 cases in 2010 and 2015, respectively (Fig. 4). For lymphoma, the statistically significant trend in rates resulted in a projected increase from 39 cases in 2001 to 43 cases in 2010 and to 47 (95% CI = 33–84) cases in 2015.

Figure 4.

Trends in the number of cases for leukemia, CNS tumors and lymphoma, among children (aged 0–17) in Ontario, showing actual (1986–2001) and projected (2002–2015) values.


Incidence rates for childhood cancers in Ontario were similar to or slightly higher than those reported for several other populations, including Great Britain,19 the United States20 and elsewhere.21 For example, Zahm and Devesa20 reported an age-standardized incidence rate of 141 per million for the United States (white, aged 0–14) in 1987–1990, during which time the Ontario rate ranged from 158 to 164 per million (Table III). On the basis of cancer registries around the world, Parkin et al.21 reported wide international variation in childhood cancer incidence rates, with rates in Ontario being generally similar to those for white populations in the United States, Australia and western Europe, but higher than in Asia, Africa and South America. In addition, the observed relative frequencies of leukemia, CNS tumors and lymphomas as the most common types of childhood cancer, and the age-specific patterns, depicted by regression models, were consistent with incidence patterns reported in other westernized countries.

The analyses of population-based incidence rates for childhood cancer over a 16-year period in Ontario indicated that trends have been modest overall, even though both rates and counts have increased substantially (Tables I and II). A marginally statistically significant trend was detected for all cancers combined (0.5% increase per year with p value < 0.10) and for lymphomas, which have increased by an average of 1.2% per year (95% CI = 0.0–3.9%). Over this period, childhood leukemia rates were stable, changing on average by <0.5% per year. The best fitting model for CNS tumor incidence indicated that rates in the 1990s were 37% (95% CI = 12–68%) higher than in the 1980s, and that there was no further statistically significant yearly trend. When the significant shift at 1990 was considered for all childhood cancers, the resultant pattern for all sites combined showed no statistically significant increase in 1990 (RR = 1.096, 95% CI = 0.98–1.01).

In interpreting these incidence patterns, consideration must be given to data quality and its impact. Advantages of examining trends in Ontario include that it refers to a stable and well-characterized population over a 16-year period, such that the source data were relatively uniform. One strength of this setting is that the co-ordinated, government-funded health services system allows patients across the province to be managed similarly in terms of the detection, diagnosis and histologic classification of their disease. POGONIS is an active registry for cancer diagnosed in Ontario children. All childhood cancer cases diagnosed and treated in pediatric centers in Ontario are registered in POGO and actively followed during and after treatment. An evaluation of the quality of the incidence data was conducted by comparing incidence data from the POGO database with that in the Ontario Cancer Registry (OCR), which is the other major independent source of cancer incidence data in the province. These data sources differ in that the POGO database focuses on children and contains detailed clinical information actively recorded at pediatric oncology centers, whereas the OCR refers to cases of all ages and contains information collected using a passive, voluntary method that links cancer clinic data, pathology reports, hospitalization information and death certificates. For the age group 0–19 years, the Canadian Childhood Cancer Surveillance and Control Program reported an 88% agreement between incidence data for 1995–1996 supplied by the OCR and POGO,5 largely as a result of lower capture rates in the age range 15–19, since a significant proportion of adolescents receive care outside pediatric centres.6 However, analysis of comparative data from the OCR and POGO sources for the age group 0–14 revealed 96% capture of incident cases in the POGO database.5 A more recent comparison of POGO and OCR data bases for the 0–14 age group indicated that between 1986 and 2003, POGO captured about 99% of cases registered in OCR cancer registry (personal communication). The previous evaluations of data quality resulted in new strategies for improving completeness and data quality at both POGO and the OCR. It is noteworthy that the leading categories for cases missed in the POGO database in 1991 and before were the 15–17 year age range, and a diagnosis of lymphoma, whereas the observed incidence rate shift was particularly strong for CNS tumors across all age groups. Improvements in medical services may account for some of the observed patterns. Although computerized tomography (CT) facilities were available widely prior to 1985 in Ontario, their use and the subsequent development of other high-resolution imaging procedures (e.g., magnetic resonance imaging (MRI)) may have resulted in increased case finding, and potentially a redistribution of cases that previously had a nonspecific diagnosis. In an analysis of pediatric brain tumor incidence trends in the United States from 1973 to 1994, Smith et al.17 demonstrated that the trend was best explained statistically by a sharp increase in the mid-1980s, and that rates before and after that time were relatively stable. Smith et al.17 concluded that this pattern was quite likely due to earlier detection and reporting enabled by the dissemination of MRI technology. In Ontario, MRI facilities became available later than in the United States, with initial programs being established by 1990, province-wide availability by 1995 (with 12 MRI facilities), and more complete coverage at present, with 44 facilities in 2001.22 The finding in Ontario that the shift parameter was associated with CNS tumors but not with leukemia and lymphoma is consistent with the fact that advances in imaging have been of particular value in the diagnosis of CNS tumors. Thus, the incidence rates for all cancers combined and for CNS tumors in Ontario are consistent with the pattern observed for brain tumors alone by Smith et al.17

Incidence rates for all sites combined among children have increased slightly, by almost 1–2% per year in Great Britain over a 30-year period,23 and in the SEER registry regions in the United States from 1974 to 1991,13 the Greater Delaware Valley,24 Queensland, Australia25 and specific regions of Spain.15 A similar trend was detected in a simple linear model of the Ontario rates. Childhood leukemia incidence rates were quite stable in most previous reports, as was observed in Ontario,23, 24, 26, 27, 28 although slight increases of about 1% per year have been reported in the United States13, 20, 29 and Queensland.25 Regarding childhood lymphoma trends, statistically significant but small annual increases in nonHodgkin's lymphoma were reported in the Greater Delaware Valley24 and Queensland,25 whereas other US studies found stable rates.13, 29 Thus, the incidence patterns seen in Ontario for childhood leukemia and lymphoma are consistent with those reported by others.

Several studies have reported trends of increasing childhood CNS tumor incidence rates. In the USA, Chow et al.27 found a 1.4% average annual increase from 1969 to 1989, Gurney et al.13 reported an average annual increase of 2% for astroglial tumors from 1974 to 1991, and Bunin et al. detected a 3% annual increase for all CNS tumors combined and for specific major histologic types. Similar significant increases in CNS incidence were seen in other studies in the USA,20, 29 Denmark,30 Sweden,31, 32 Finland,33 Spain15 and elsewhere.26 We did not observe a similar pattern in CNS tumor incidence for Ontario, but for most studies, it is not possible to judge whether a change-point contributed to the trend, such as that detected in Ontario and as reported by Smith et al.17 Important implications of these results include that comparisons between studies must be made with caution, because changes in incidence will be even more sensitive to the period under investigation, as the timing of a change-point may vary between populations that have different health services delivery systems, and thus, different times of introduction of new diagnostic modalities.

Further implications of these results include the projection of the future caseload expected in Ontario, under the assumption that past trends will continue. The number of cases for leukemia and CNS tumors is expected to remain relatively stable, whereas the number of lymphoma cases is expected to increase by 45%, from the 5-year moving average of 31 in 1995 to ∼45 in 2015. This large increase in lymphoma is expected mostly as a result of a 1.9% average annual increase (e.g., RR over 21 years = 1.02121 = 1.46). Similarly, the number of cases of all cancers combined is expected to increase by 8%, from the 5-year moving average of 320 in 1995 to ∼347 in 2015.

In summary, these analyses of childhood cancer incidence rates for the population of Ontario over a 16-year period demonstrated that, overall, time trends were generally modest, except for the lymphomas, which increased by about 1.9% per year. The concept of a change-point, which was demonstrated previously in CNS tumor incidence in the United States, was found to be relevant in Ontario and after accounting for it there was no further evidence of a steady increase over time.

In addition to providing targets for the planning and development of pediatric cancer care programs, these projections serve as benchmarks for detecting whether current incidence diverges from previously established patterns. Although the cause of these patterns cannot be determined directly in a descriptive study such as this, potential contributors include changes in data quality and changes in diagnostic practices, reporting and classification. This reinforces the primacy of data quality, the need for rigorous and standardized processes for collecting and processing data, and the value of continuing evaluation of surveillance systems, which aim to detect determinants of risk in the population.


The assistance of E. Walker, E. McLaughlin, R. Parkes and G. Fehringer is gratefully acknowledged. We would also like to acknowledge the contribution of the 5 POGO partner institutions: The Hospital for Sick Children in Toronto, McMaster Children's Hospital in Hamilton, Children's Hospital of Western Ontario in London, Kingston General Hospital, and Children's Hospital of Eastern Ontario in Ottawa. We are grateful to the POGO data managers/clinical research associates, who gathered the data and ensured its quality.