Health Reports
Trends in paediatric cancer survival in Canada, 1992 to 2017

by Larry F. Ellison, Lin Xie and Lillian Sung

DOI: https://www.doi.org/10.25318/82-003-x202100200001-eng

Each year in Canada, approximately 1,000 children aged 0 to 14 years are diagnosed with cancer, and 110 die from the disease.^{Note 1} Worldwide, impressive gains in survival have been made over time.^{Note 2Note 3Note 4} These advances can primarily be attributed to a deeper understanding of paediatric cancer biology, combined with successive, multi-institutional clinical trials.^{Note 5} As a result, much of what is known about expected outcomes and risk factors is derived from clinical trials. Clinical trials provide excellent insight into how regimens perform within the context of close monitoring, strict application of eligibility criteria and adherence to protocol therapy. However, these results may not be generalizable to children not enrolled in clinical trials. For example, observational studies have suggested that children enrolled in clinical trials differ in terms of demographic characteristics and cancer-specific features from those not enrolled. Consequently, clinical trial results may not reflect outcomes at the population level.^{Note 6Note 7}

One way to overcome the limitation of generalizability is through the use of population-based cancer registries such as the Canadian Cancer Registry (CCR). A previous analysis using the CCR found that paediatric cancer incidence rates were stable while rates of death decreased after 1985.^{Note 8} While this baseline evaluation provided important information from a national perspective, several important questions remain. Gaps include the description of population-based survival for a contemporary cohort—the last comprehensive evaluation was published in 2007^{Note 2}—and the evaluation of whether improvement in survival has been constant over time and by diagnosis. In addition, conditional survival, or the probability of continued survival given an initial survival period, has not been well described for paediatric cancer, whereas such analysis has been performed for adult cancer.^{Note 9} Such a description would be meaningful to both patients and health care providers.

Consequently, the objectives of this study were to describe survival, improvement in survival over time and conditional survival for paediatric cancer patients in Canada. Using data from the CCR, the study presents short- and long-term predicted survival estimates for the five-year period from 2013 to 2017, for all childhood cancer diagnostic groups and for selected subgroups. Trends in survival, including those for all childhood cancers combined, are examined from the 1992-to-1996 period to the 2013-to-2017 period. Five-year survival estimates conditional on having already survived 1 to 10 years are also provided.

Data and methods

Data sources and definitions

The data source was a pre-existing analytic file created by linking CCR cases diagnosed from 1992 to 2017 to mortality information complete through December 31, 2017, via Statistics Canada’s Social Data Linkage Environment.^{Note 10} The CCR is a population-based database composed of cases diagnosed among Canadian residents since 1992. Each provincial and territorial cancer registry provides demographic and cancer-specific information to Statistics Canada in a standard format. Annual submissions by jurisdictions include additions and revisions to data submitted in previous years.^{Note 11} The mortality information was obtained from the Canadian Vital Statistics Death database (CVSD), whose scope is all deaths in Canada,^{Note 12} and from the T1 personal master file (as reported on tax returns). The use of death information from tax returns facilitated the identification of additional death events of patients in the CCR that may not have been included in the CVSD, such as out-of-country deaths. This source was also used to validate the date of death when discrepancies arose between dates in the CCR and the CVSD. The analytic file followed the multiple primary coding rules of `the International Agency for Research on Cancer.^{Note 13} Survival time was measured in days. More information on the linkage process and on the resulting death-linked analytic file is available upon request.

Cancers in children aged 0 to 14 years were classified according to the Surveillance, Epidemiology, and End Results (SEER) Program update^{Note 14} of the International Classification of Childhood Cancer, Third Edition (ICCC-3).^{Note 15} The update was in response to new morphology codes introduced by the World Health Organization.^{Note 16} For 19 cases with a histology code of 8963 (malignant rhabdoid tumour) and a topography code of C71 (brain) that would otherwise not have been mapped to a diagnostic group, the histology code was edited to 9508 (atypical teratoid/rhabdoid tumour) and the cases included in diagnostic subgroup IIIc.

Inclusion and exclusions

New malignant primary cancers diagnosed in children aged 0 to 14 years were initially included. Cases from Quebec were excluded because cancer incidence data from this province had not been submitted to the CCR since the 2010 data year. Exclusions then proceeded in a stepwise manner, starting with cases for which the diagnosis was established through autopsy only or death certificate only (0.5% excluded). The year of death, if applicable, was known in each case. The dataset was further restricted to first primary cancers per person per diagnostic subgroup diagnosed from 1992 to 2017 (0.4% excluded).^{Note 17Note 18Note 19Note 20} For children with multiple primary cancers within the same diagnostic group, only the first cancer was included in analyses at the diagnostic group level (0.1% excluded). These exclusions were incorporated to avoid including two deaths for a single person in the same survival analysis.^{Note 20} Finally, 15 remaining malignant cancer cases that did not map to a diagnostic group were excluded.

Diagnostic group and subgroup reporting

Results were reported for all 12 ICCC-3 diagnostic groups and for 21 of the 47 diagnostic subgroups, because of the rarity of diagnoses in many subgroups. Specifically, results were reported for subgroups when the standard error associated with their five-year observed survival proportion (OSP) for 2013 to 2017 was equal to or less than 0.05 (rounded), and the total number of cases diagnosed from 2008 to 2017 was at least 50. Consequently, in addition to the total cohort that included all eligible cancer cases, results for 24 independent individual ICCC-3 groups or subgroups were provided (i.e., the 21 subgroups plus 3 groups for which no subgroups were reported).

Statistical analysis

OSPs were derived using an algorithm developed by Dickman^{Note 21} and reported as percentages. Cases with the same date of diagnosis and death (excluding those diagnosed through autopsy only or death certificate only) were assigned one day of survival so they would be included in the survival estimates. Standard errors of OSP estimates were determined using Greenwood’s method.^{Note 22} OSPs for all childhood cancers combined were calculated as a weighted average of sex- and diagnostic-group-specific estimates. The weights used were based on the sex and diagnostic group case-mix distribution of people aged 0 to 14 diagnosed with cancer in Canada, excluding Quebec, from 2010 to 2014 (see Appendix Table A1 for weights). Standard errors for estimates of this index were estimated by taking the square root of the sum of the squared, weighted, diagnostic-group-specific OSP standard errors. The most recent five-year period for which actual estimates can be calculated depends on the follow-up duration being examined (e.g., 2003 to 2007 for 10-year survival).

To describe changes in survival over time, survival estimates for three periods were presented: 1992 to 1996, 2003 to 2007 and 2013 to 2017. The period method^{Note 23} was used to predict OSPs for 2013 to 2017, while non-predictive (actual) estimates for 1992 to 1996 and 2003 to 2007 were calculated using the cohort method. Actual long-term survival estimates for people diagnosed in the most recent period will not be known for some time. The most recent cohort of cancer patients with complete five-year survival information was diagnosed in the 2008-to-2012 period. Empirical evaluations of period analysis have shown that this method provides estimates that closely predict the survival that is eventually observed for people diagnosed in the period of interest, particularly when survival is fairly constant.^{Note 24Note 25Note 26} When survival is generally increasing (or decreasing), a period estimate tends to be a conservative prediction of the survival that is eventually observed.^{Note 25Note 27}

The underlying methodology of both the cohort and the period methods is essentially the same, except the follow-up information used in the period method necessarily does not relate to a fixed cohort of people. Rather, estimates of period survival are based on the assumption that people diagnosed in the period of interest will experience the most recently observed conditional probabilities of OSPs.

The percentage point increase in five-year OSPs was used as the measure of improvement in survival. Differences in OSPs were calculated before rounding to one decimal place. The Z-test was used to determine P-values for between-time-period differences; the standard errors of differences were estimated by the square root of the sum of the variances associated with the two OSP estimates. P-values correspond to two-sided tests of the null hypothesis that the change in OSPs is zero, with a significance level of 0.05.

Five-year observed conditional survival proportions (OCSPs) were calculated as per five-year OSPs using only the data of people who had survived given selected times.^{Note 9Note 28} That is, they are the survival estimates for an additional 5 years among people who had already survived 1, 3, 5 or 10 years.

Results

Distribution of cases

From 1992 to 2017, 18,056 new cancer cases diagnosed in children 0 to 14 years of age were eligible and included in the survival analyses. Table 1 describes the demographics of the cohort and distribution of cases by ICCC-3 diagnostic group and selected subgroup. More boys (54%) than girls were diagnosed, slightly less than half of the children were diagnosed before the age of 5 (46%), and most of the cases were diagnosed in Ontario (52%). Among eligible cases, 87% were histologically verified, as were 94% of eligible cases for which the method of diagnosis was known. The most common diagnostic group was leukemias, myeloproliferative diseases and myelodysplastic diseases (33%), and the most common subgroup was lymphoid leukemias (26%).

Survival varied by cancer type

Table 2 presents 1-, 3-, 5- and 10-year predicted survival for cases diagnosed from 2013 to 2017. Five-year OSPs were at least 90% for 10 of the 24 individual ICCC-3 diagnostic groups or subgroups reported, and less than 80% for 9 others. Five-year survival was highest for thyroid carcinomas, at 100% (95% confidence interval [CI] undefined). This was followed by Hodgkin lymphomas, at 99% (95% CI = 95 to 100); malignant gonadal germ cell tumours, at 97% (95% CI = 85 to 100); and nephroblastoma and other nonepithelial renal tumours, at 96% (95% CI = 92 to 98). It was lowest for other gliomas, at 42% (95% CI = 33 to 51), followed by acute myeloid leukemias (AMLs), at 65% (95% CI = 57 to 71), and osteosarcomas, at 65% (95% CI = 53 to 74). Median diagnostic-group-level OSPs were 95.5% (1-year), 88.5% (3-year), 84.0% (5-year) and 81.0% (10-year).

Survival did not vary by sex among leading cancer types

No significant sex-specific differences in five-year survival were observed among any of the three most commonly diagnosed cancer subgroups (Figure 1). For lymphoid leukemias, astrocytomas, and neuroblastoma and ganglioneuroblastoma, corresponding survival estimates among males and females differed by 1 percentage point or less. For lymphoid leukemias, five-year survival was higher among children diagnosed before the age of 10 than among those diagnosed between the ages of 10 and 14.

Survival improved over time both overall and for selected cancers

Figure 2 depicts changes over time in 1-, 5- and 10-year OSPs for all childhood cancers combined, after adjustment for changes over time in the sex and diagnostic group distribution of cancer cases. Predicted OSPs for all childhood cancers combined for 2013 to 2017 were 93% (1-year), 84% (5-year) and 82% (10-year). Increases in the 5- and 10-year OSPs were virtually the same from the 1992-to-1996 period to the 2013-to-2017 period (7.5 to 7.6 percentage points), and 2.7 percentage points for 1-year survival. Increases were statistically significant for all three durations (p < 0.001).

A statistically significant increase in the five-year OSP from the 1992-to-1996 period to the 2013-to-2017 period was observed for 8 of the 24 individual ICCC-3 groups or subgroups reported (Tables 3-1 and 3-2). Diagnostic groups or subgroups experiencing the largest increases over time were chronic myeloproliferative diseases (35.4 percentage point increase), and ependymomas and choroid plexus tumour (32.1 percentage point increase).

Improvement in survival greatest in earliest period

For all childhood cancers combined, much of the 7.5 percentage point improvement in five-year survival occurred in the first half of the study period. The 1.3 percentage point increase from the 2003-to-2007 period to the 2013-to-2017 period was not statistically significant (p = 0.134). A similar pattern was also noted for six of the eight individual cancers exhibiting an increase over the entire period, including all three subgroups in the leukemias, myeloproliferative diseases and myelodysplastic diseases diagnostic group. The split was particularly noteworthy for AMLs, for which a 23.5 percentage point increase from the 1992-to-1996 period to the 2003-to-2007 period was followed by a non-significant 7.0 percentage point decrease from the 2003-to-2007 period to the 2013-to-2017 period. Conversely, all of the 9.5 percentage point increase for nephroblastoma and other nonepithelial renal tumours, and three-quarters of the 16.6 percentage point increase for intracranial and intraspinal embryonal tumours, occurred in the most recent period. A statistically significant increase since 2003 to 2007 was observed only for one other category (miscellaneous lymphoreticular neoplasms). For the other gliomas subgroup, a significant decrease since 2003 to 2007 followed a very similar significant increase from the 1992-to-1996 period to the 2003-to-2007 period.

Favourable prognoses for those surviving early years

The five-year OCSPs among children surviving the first year after their diagnosis met or exceeded 95% in 9 of the 24 individual ICCC-3 groups or subgroups reported (Table 4). When the analysis is restricted to those surviving the first five years after diagnosis, all cancers achieved this standard, with the exception of ependymomas and choroid plexus tumour, at 88% (95% CI = 74 to 94); osteosarcomas, at 91% (95% CI = 79 to 96); and intracranial and intraspinal embryonal tumours, at 92% (95% CI = 84 to 96). While the five-year OSP for other gliomas of 42% (95% CI = 33 to 51) was the lowest observed from diagnosis, the five-year OCSP among children surviving this cancer for three years was 92% (95% CI = 81 to 96). Similarly, the five-year outlook after three years among those diagnosed with AMLs was 95% (95% CI = 88 to 98), whereas at diagnosis it was 65% (95% CI = 57 to 71).

Discussion

In Canada, five-year predicted OSPs for the period from 2013 to 2017 were at least 90% for 10 of the 24 childhood cancer diagnostic groups or subgroups reported. A significant 7.5% increase in the five-year OSP from the 1992-to-1996 period to the 2013-to-2017 period was observed for all childhood cancers combined. The greatest increase was for chronic myeloproliferative diseases, at 35.4 percentage points. However, no improvement in five-year survival was observed for a number of diagnostic groups with relatively poor prognoses at baseline. Once children survived five years, the probability of surviving another five years exceeded 95% across most diagnoses.

Overall, reasons behind increases in survival are likely multifactorial and include factors such as better supportive care, improved treatment protocols derived from successive randomized trials and better risk stratification. The biggest increase in survival occurred in chronic myeloproliferative diseases. This finding is likely attributable to the development of targeted therapy with tyrosine kinase inhibitors,^{Note 29Note 30} which has enabled better survival, combined with a substantial reduction in the intensity of therapy, including the use of hematopoietic stem cell transplantation. The improvement in ependymoma and choroid plexus tumour survival may be attributable to better surgical and radiotherapy approaches.^{Note 31Note 32}

Significant progress in survival was observed for children diagnosed with AMLs in Canada, though the five-year estimate remained relatively poor, at 65%. However, the progress appeared entirely limited to the first half of the study period. Similar results were reported in the United States, except progress appeared more consistent over time.^{Note 33} Improvements in survival for children with AMLs have been attributed to better supportive care,^{Note 34} including routine administration of prophylactic antibacterial^{Note 35Note 36Note 37} and antifungal therapy,^{Note 38Note 39} as well as better approaches to allogeneic hematopoietic stem cell transplantation. However, these approaches should have been similar in Canada and the United States. The different patterns seen in the two countries may be spurious or may also be related to the use of therapy based on the United Kingdom Medical Research Council in some Canadian centres^{Note 40} during the earlier period, which was associated with very good outcomes.^{Note 41Note 42Note 43}

There are two broad reasons why some diagnostic groups or subgroups may not have experienced significant increases in survival over time. Some paediatric cancers were associated with high survival during the baseline period (i.e., 1992 to 1996), making it difficult to achieve and demonstrate significant increases over time. For example, while no significant increases in five-year survival were observed for either germ cell tumour subgroup, the five-year OSP for 2013 to 2017 exceeded 90% for both. However, failure to observe increases in survival may also mean that research has not translated into better outcomes for paediatric patients with relatively poor survival. For example, patients diagnosed with hepatic tumours, malignant bone tumours, and soft tissue and other extraosseous sarcomas all had relatively poor prognoses at baseline that remained virtually unchanged 20 years later. Future research should focus on identifying innovative approaches for diagnoses with poor outcomes and in which progress has been stalled. This highlights the importance of this study.

It has been emphasized that the characteristics of childhood cancer, including low incidence rates and favourable prognosis, argue for collecting paediatric cancer data separately from adult cancer data.^{Note 44} Therefore, some may be concerned that estimates of paediatric cancer incidence may be biased in the CCR. However, paediatric cancer incidence rates in the CCR closely approximate those in Cancer in Young People in Canada, a paediatric-cancer-specific national population-based cancer registry,^{Note 45} suggesting that biased ascertainment of paediatric cancer in the CCR is not a concern. Another issue is that registries differ in their inclusion of non-malignant tumours, thus raising questions about comparability. The present study focused only on malignant cancers and classified cancers according to the SEER update of the ICCC-3, thus improving interpretability of the data. Lastly, this study did not include patients aged 15 and older. Future efforts could focus on evaluation of adolescent and young adult cancer patients.

During the COVID-19 pandemic, diagnosis of cancer in children may be delayed.^{Note 46} Parents may be less likely to engage with the health care system for cancer-like symptoms in their children. Delays in investigating symptoms not initially thought to be related to cancer may also be an issue, as children can be diagnosed with cancer incidentally. Children diagnosed with cancer during the pandemic, and those diagnosed shortly before, may also experience atypical delays in receiving treatment.^{Note 46Note 47} The impact of delays in the diagnosis and treatment of childhood cancer is unclear,^{Note 48Note 49} but delays are likely to have an emotional impact on the family, particularly where outcomes are poor. While the results of this study are based on data collected prior to the pandemic, they provide a baseline against which the impact of the pandemic on childhood cancer survival outcomes can eventually be gauged.

Strengths and limitations

Strengths of this study include its population-based nature, which reduces the possibility of selection bias. Furthermore, the ability to ascertain almost all deaths irrespective of time from diagnosis provides confidence in the long-term survival estimates. This aspect is particularly important since clinical trials and institutional reports may be limited in their ability to identify late deaths, particularly after children transition to adult institutions or if they move to other jurisdictions. Lastly, the description of conditional survival is clinically meaningful and will be useful for reassuring and counselling families. However, the data from this study should be interpreted in light of its limitations. The CCR lacks detailed diagnostic and treatment information; therefore, possible factors associated with both survival and improvements in survival over time could not be evaluated. More annotated databases, such as the Cancer in Young People in Canada registry,^{Note 45} could potentially be used to evaluate factors such as treatments associated with improvements in survival or worsening outcomes. Finally, the absence of cases diagnosed in the province of Quebec is an important limitation that should be addressed in future research.

Conclusion

Significant improvements in both short- and long-term paediatric cancer survival have been made in Canada since the early to mid-1990s. However, there has been little improvement over time for some cancer types with poor prognosis, including hepatic tumours, malignant bone tumours, and soft tissue and other extraosseous sarcomas. For children who survived the initial few years after diagnosis, the subsequent long-term outlook was very favourable. This finding is clinically meaningful and will be useful for reassuring and counselling families.

Appendix