Chapter 5: Cancer incidence in Canada: trends and projections (1983-2032) - HPCDP: Volume 35, Supplement 1, Spring 2015

Chapter 5: Discussion

Main findings

In this monograph we show that the ASIRs for all cancers combined in Canada are not projected to change substantially from 2003–2007 to 2028–2032. The rates are expected to decrease by 5% in males, from 464.8 to 443.2 per 100 000, and to increase by 4% in females, from 358.3 to 371.0 per 100 000 (Figure 3.3). The decrease in lung cancer rates in males 65 or older and of prostate cancer in those 75 or older will contribute to the overall decrease in cancer rates in males, given that these 2 cancers account for about 40% of all new cancer cases in Canadian males. The predicted overall increase in cancer rates for females is primarily the result of increasing lung cancer rates in women aged 65 or older; it also represents the expected increase in cancers of uterus, thyroid, breast (in females under 45), leukemia, pancreas, kidney and melanoma.

The annual number of newly diagnosed cancer cases is projected to increase by 84% in males, from 80 810 in 2003–2007 to 148 370 in 2028–2032, and by 74% in females, from 74 165 to 128 830 over the same period. Our decomposition analysis of the drivers of change illustrated that the projected rise in the number of new cancer cases will primarily come from the structural aging of the Canadian population and, to a lesser extent, the increase in population size. Changes in the risk of cancer will contribute little to the increase in new cases, especially for males.

Between 2003–2007 and 2028–2032, significant risk reductions are projected for major common tobacco-related cancers, albeit with relatively lower reductions or delayed downturns in females. The incidence of smoking-related cancers is projected to decrease by 2% to 59% for oral cancer in males, cervical and esophageal cancer in females, and larynx, lung, stomach and bladder cancers in both sexes, while an increase of 0.6% to 7% is expected for kidney cancer, leukemia and pancreas cancer in both sexes, oral cancer in females, and esophageal cancer in males. The differences in incidence trends of the tobacco-related cancers between males and females reflect the differences in histories of tobacco use.^{Endnote 42, Endnote 43} Given the lag of 20 years or more between the drop in smoking rates and the decrease in cancer incidence rates, it is likely that incidence rates in females will begin to drop more noticeably for the tobaccorelated cancers with projected stable or marginally decreased trends over the longer term.

Over the 25-year projection period, the incidence rates for cancers associated with excess weight and physical inactivity are estimated to rise by 0.6% to 16% for cancers of uterus, kidney, pancreas, female breast and male esophagus, in descending order. Incidence rates for colorectal and female esophageal cancer, also associated with excess weight and physical inactivity, are estimated to fall by 2% to 6%. Increased obesity prevalence in Canada may contribute to the increased incidence trends of these cancers. ^{Endnote 51, Endnote 72, Endnote 73, Endnote 110} Weight control and physical activity may represent opportunities for modifying the risk of developing these cancers.

The most common cancers caused by chronic infections are cancers of cervix, caused by human papilloma virus (HPV), stomach, caused by Helicobacter pylori (H. pylori), and liver, caused by hepatitis B virus (HBV) and hepatitis C virus (HCV). The ongoing increasing trend of liver cancer incidence in Canada is possibly linked to the historical increase and continued high incidence in HCV infection, ^{Endnote 98} the aging of the population previously infected and increasing immigration from areas where risk factors, such as HBV, are prevalent.^{Endnote 75, Endnote 104, Endnote 105} The persisting decrease in incidence of stomach cancer may be explained by improved healthy behaviours, such as decreased smoking and changes in diet,^{Endnote 82} and increased recognition and treatment of infection with H. pylori.^{Endnote 36, Endnote 83} The continuing downward trend in the rates of cervical cancer is mainly attributable to general population screening with the Papanicolaou (Pap) test and successful treatment of screening-detected precancerous lesions. The immunization of school-aged children with HPV vaccine is anticipated to further reduce incidence of cervical cancer.

The incidence rates for all cancers combined are projected to continue to be highest for males in the Atlantic region and for females in Quebec in 15 years but in Ontario thereafter, and lowest in British Columbia. The cancer-specific analysis shows that the highest incidence rates in males are projected to be in the Atlantic region for cancers of oral, esophagus, stomach, colorectum, pancreas, larynx, melanoma, prostate, kidney and non- Hodgkin lymphoma (NHL); in Ontario for melanoma, thyroid, NHL and leukemia; and in Quebec for cancers of larynx, lung, bladder, central nervous system (CNS), Hodgkin lymphoma and multiple myeloma. For females, the elevated incidence rates are predicted in the Prairies for cervical cancer; in Ontario for multiple myeloma, leukemia, oral, stomach, breast, uterus, ovary and thyroid cancers; in Quebec for cancers of pancreas, lung, bladder, CNS and Hodgkin lymphoma; and in the Atlantic region for cancers of colorectal, larynx, melanoma, kidney and NHL.

While British Columbia is projected to continue to have the lowest incidence rates for majority of cancers in both males and females, this province will continue to experience the highest rates of esophageal cancer in females, liver cancer in both sexes, and testis cancer. The Atlantic region is projected to have the lowest rates for breast, uterus and ovarian cancers, and for liver and leukemia in both sexes, but elevated rates in males for about half the cancers studied (listed above).

The differences in incidence rates in regions are influenced in part by variation in the past prevalence of risk factors across the country, in keeping with lengthy time lags between exposure and cancer outcomes. Cancer risk factors include cigarette smoking, alcohol consumption, obesity, physical inactivity, diet/nutrition, radiation, some chronic infections, medicinal drugs, immunosuppression, occupational and environmental contaminants, and genetic susceptibility. The historically higher smoking rates in Quebec and Atlantic Canada likely account for the higher incidence rates of lung cancer in these regions. The higher rates of liver cancer in British Columbia is possibly linked to the higher HCV rates in this province.^{Endnote 297} In addition, the high incidence rates of liver cancer and of esophagus cancer in females in British Columbia could partially be explained by high number of immigrants from South Asia and China where HBV is endemic.^{Endnote 75, Endnote 76, Endnote 77, Endnote 78} Significantly higher incidence rates of liver cancer have been found in immigrants from South-East Asia and North-East Asia in Canada.^{Endnote 104} Compared to Canada, the rate of esophageal cancer is significantly higher in Asia, including China and Central Asia, especially in females.^{Endnote 312} The geographical variation in cancer incidence may also be due to the availability of screening and diagnostic services for breast, colorectal, prostate and cervical cancers, and the different rates of participation in formal screening programs (e.g. mammographic screening for breast cancer) or other screening procedures (e.g. prostate-specific antigen [PSA] testing for prostate cancer). Finally, the variation in cancer registry practices could also explain some of the geographical differences in cancer distribution (see the section “Data quality issues” below). Low rates of prostate cancer and melanoma in Quebec are likely the result of the registry relying on hospitalization data and missing cancers diagnosed and treated outside the hospitals.^{Endnote 130}

Prostate, colorectal, lung and bladder cancers figure among the top 4 most common cancers newly diagnosed in males in 1983–1987, 2003–2007 and 2028–2032. However, prostate cancer replaced lung cancer as the most frequent in 2003–2007, and colorectal cancer is projected to overtake lung cancer as the second most frequently diagnosed cancer in males by 2028–2032. For females, breast, lung, colorectal and uterine cancers are the leading incident cancers in these 3 periods, but colorectal cancer—the second most common type of cancer in 1983–1987—is ranked third as of 2003–2007. Thyroid cancer will replace NHL as the fifth most common cancer in females by 2028–2032.

Mistry et al.^{Endnote 4} projected cancer incidence in the UK for 2008–2030 based on 1975–2007 data and using a method similar to the Nordpred package. The years 2007 and 2030 are used here to approximate the periods 2003–2007 and 2028–2032, and correspondingly, to compare projections between Canada and the UK. As in Canada (Table 5.1), almost no change is projected in the ASIRs of all cancers combined from 2007 to 2030 in the UK. Sites with similar projected decreases in ASIRs in both countries included cancers of stomach and CNS. Oral cancer incidence rates in Canada are projected to decrease in males by 6% from2007 to 2030 and remain stable in females, in contrast to the increases of 25% in males and 21% in females in the UK. However, the rates for 2007 in Canada were higher than those in the UK by 15% in males and 13% in females. The predicted changes in ASIRs for colorectal cancer are below the medians in all cancers, with a decrease of 6% for both sexes in Canada, whereas the rates in the UK are expected to decrease by a similar amount in males but increase by 2% in females. The incidence rates of laryngeal cancer are estimated to decrease about 2 times faster in Canada than in the UK in both sexes. For lung cancer, the ASIRs are projected to decrease by 34% in males and 16% in females in Canada, compared with a predicted decrease of 8% in males and increase of 7% in females in the UK. The ASIR of ovarian cancer is expected to decrease to a greater extent in the UK than in Canada (28% vs. 4%). Breast cancer incidence is expected to show the smallest change (increase of less than 1%) over the entire 25-year forecasting horizon in all cancer sites in Canadian females. Similar to Canada, breast cancer incidence rate in females is not predicted to change substantially in the next 25 years in the UK.

The ASIRs of liver cancer in males are projected to rise by 43% in Canada and 27% in the UK from the similar level in 2007, whereas the rates in females are estimated to increase by 15%in Canada but level off with a 2% fall in the UK. However, the UK rate in females for 2007 was 69% higher than the Canadian rate. A 52% increase in melanoma rates is projected for males and females in the UK from 2007 to 2030, while Canadian rates are predicted to decrease by 6% in males and increase by 5% in females. Melanoma incidence rates in the UK were similar to Canadian rates in males and slightly higher in females in 2007. The ASIR of uterus cancer is projected to rise more rapidly in Canada (16% vs. 4%) from the same level in 2007. Increases in kidney cancer rates in females are projected in both the UK and Canada to about the same level, but the rate in 2007 in Canada was 10% higher than that in the UK. Conversely, the rates in males were the same in the both countries in 2007, but are estimated to increase 4 times faster in the UK between 2007 and 2030. The ASIRs of thyroid cancer are estimated to increase at 55% for Canadian males and 65% for Canadian females, but are not shown in the UK study. In both sexes, the rates of multiple myeloma and leukemia are projected to decrease by 14% to 23% from 2007 to 2030 in the UK but increase by 4% to 11% in Canada. Recent relatively stable rates of multiple myeloma in Canada have been mapped to these future trends.

Data quality issues

Although the standardization of case ascertainment, definition and classification has improved the registration of cancer cases and comparability of data across the country, reporting procedures, accuracy and completeness still vary.^{Endnote 1} International Agency for Research on Cancer (IARC) rules^{Endnote 313} for multiple primaries were used for cases from the Canadian Cancer Registry (CCR), whereas during the period covered by the National Cancer Incidence Reporting System (NCIRS), registries other than Quebec and Ontario used multiple primary rules that allowed a small percentage of additional cases.

Non-melanoma skin cancer is difficult to register completely because it is quite plentiful and may be diagnosed and treated in a variety of settings. Most provincial and territorial cancer registries do not register these cases. For this reason, non-melanoma skin cancer is excluded from our analysis.

For the observed data years covered by this analysis, death certificate only (DCO) cases were not reported to CCR by Quebec and Newfoundland and Labrador, with the exception of the 2000–2006 Quebec data and 2007 Newfoundland and Labrador data. The number of DCO cases for 2007 in Quebec was estimated by averaging the numbers in 2002–2006. This missing reporting has likely led to underestimates of the incidence rates in these provinces, especially for highly fatal cancers such as lung and pancreas. In Canada, the number of DCO cases is less than 2% of the total new cancer cases. In addition, the incidence of some cancers in Quebec, particularly for those that rely more heavily on pathological diagnosis, are underestimated as a result of the registry's dependence on hospitalization data. Prostate cancer, melanoma and bladder cancer estimates are affected.^{Endnote 130} Owing to changes to the Quebec registry that increase registration for data after 2007, the number of melanoma cases is underestimated in the current report.

Comments on methods and results

Our model comparison exercise, based on the more recent observed data (2008–2010) which were not available when present study was undertaken, addressed the accuracy of the projection methods used in this monograph. For example, Table 5.2 presents the medians of the absolute relative differences between the observed and projected average annual number of cases at the national level only and across the provinces in 1992–2010 by length of projection for the combinations of cancer site (excluding prostate cancer), sex and province (excluding Quebec, see Chapter 2 for details), not listed in Table 2.3. The projected numbers were calculated by the projection method used in this monograph (denoted as PHACpred, which therefore only includes the Nordpred APC models (NP_ADPC) with the Nordpred standard drift (D) reduction and its modifications), and the 3 versions of NP_ADPC with its standard drift reduction: using the average trend over the whole observation period for projections (M0F); using the slope between the 2 most recent periods for projections (M0T); and automatically determining whether the recent trend (or the average trend) is projected based on a significance test for departure from a linear trend (M0A). The medians are shown with and without all male cancers combined. The table shows that the medians from PHACpred are the smallest in the 4 models for any length of projection period. The differences in the medians among the 4 models or between PHACpred and M0A are not statistically significant when across the provinces (each p ≥.05), but are statistically significant for national-level 15- and 20-year projections. The performances of M0F and M0T were published for the population of the 4 Nordic countries.Endnote 15 In this study, Moller et al.Endnote 15 made projection model comparisons for 20 cancer sites in each sex for Denmark, Finland, Norway and Sweden for 1983–1997 based on 1958–1977 data. The respective median deviations (over the combinations of site, sex and country) of M0F and M0T are 13% and 12% for 10-year projections, and 20% and 18% for 20-year projections. The median numbers are similar to ours for M0T model in the scenario from across the provinces, but M0F seems to perform better for our specified data. Consequently, we can see that our PHACpred multiple modelling approach produced more accurate projections than the default Nordpred method applied uniformly.

Validation of the predicted incidence counts and rates is critical. Incidence data for some cancers were subject to changes in classification/coding practices, introduction or expansion of screening programs, and use of new diagnostic technologies. A model created on cohorts in early periods for these cancers may give inaccurate predictions when applied to contemporary cohorts. Because the datasets for model creation and application in this study were largely from different periods, we examined the projections from the selected models using our knowledge of data quality, trends in cancer rates, risk factors or interventions, which guided selection of the final models.

Our results were compared with the projections using the default Nordpred model, M0A, for each of the combinations of cancer site, sex and geographical area. The medians of the absolute relative differences between the projected average annual numbers of cases from PHACpred and M0A models (relative to the M0A) across all the combinations are 1.9% and 3.8% for 10-year and 25-year projections, respectively. In 25-year projections, the respective medians for breast cancer in females, colorectum and lung are 10.3%, 0.8% and 0%. The largest medians of the disagreement of the 2 methods are found in cancers of prostate (40.2%), thyroid (21.2%) and stomach (21.4%). However, M0A produces extreme increases in prostate cancer rates and new cases and therefore is not applicable to our projections (see discussions in the third paragraph below).

The principal projection models used are based on decomposition of the observed incidence data into 3 time dimensions of age, period and cohort. While the effects of risk factors, screening and intervention were not incorporated into the models because of insufficient data in most circumstances, they have been modelled indirectly to some extent, through the period and cohort effects in the model.^{Endnote 3} However, the models will be insensitive to any recent changes not foreshadowed in the observed time series of cancers because of the long latency between exposure and cancer outcomes.

The observed incidence rates for the cancers of female genital system also reflect the fact that many females who underwent a hysterectomy or bilateral salpingo-oophorectomy were not at risk of developing the disease. Table 5.3 shows that the prevalence of hysterectomy was high in the Atlantic provinces and Quebec based on the 2003 Canadian Community Health Survey (Cycle 2.1). Using all females as the denominator in the rate calculation can result in artefactual differences in regional rates. In addition, changes in trends of the rates of these procedures can impact the cancer projections. For example, if surgery rates decrease more than expected based on current trends, the incidence rates of cervical, uterine and ovarian cancers would be greater than our projections.

It is useful to acknowledge that forecasting prostate cancer incidence is subject to some uncertainty as a result of over-diagnosis of this cancer because of the PSA test. The common Nordpred approach would predict extreme increases in prostate cancer incidence rates, so this necessitated a model adjustment and/or exclusion of the observed data for certain periods. We used the 2-step approach of the short-term modelling projection following by the longterm constant-rates projection for projecting prostate cancer incidence in this report (see Chapter 2 for details). Several publications have adopted the projection method where future numbers of prostate cancer are affected only by demographic changes.^{Endnote 28, Endnote 35, Endnote 314} Quon et al.^{Endnote 314} assumed that the age-specific incidence rates of prostate cancer in the current year would remain in the future in their “best-case” scenario and predicted that the number of new prostate cancers will increase to 35 121 cases by 2021 in Canada. This is consistent with our estimate of 34 460 new cases annually in 2018–2022. Moller et al.^{Endnote 28, Endnote 35} used the 5-year average method for their projections of prostate cancer incidence in England and Norway. These constant-rate projection methods would result in underestimates of the future burden of prostate cancer if the prevalence of screening is increased or the diagnosis is improved. The future use of the PSA test will principally determine the accuracy of our projections for prostate cancer incidence.

Projections for a cancer with low frequency (whether rare or from a small population) may be subjective and unreliable. Although our projections are based on comparisons of the various models (see Chapter 2 for details) for each of such cancers, they are limited in that the number of cases only met the minimum requirements for some models.

Long-term cancer incidence projections inherently carry some uncertainty as they depend on an assumption about the continuity of past trends. Although this assumption seems reasonable based on historical data, it is likely that increasing focus on lifecourse cancer prevention, especially primordial and primary prevention through reducing risk factors while promoting protective ones, and secondary preventions through screening and early detection, will exert an influence on future incidence rates of preventable cancers. On the other hand, the projections are useful in evaluating the effects of preventive interventions. If rates observed in the future differ from those projected, this suggests that the risk/preventions influencing the rates have changed. The reliability of projections also depends on the accuracy of population forecast. The predicted populations were based on the assumptions on rates of fertility, mortality, interprovincial and international migration, and so on.^{Endnote 10} Elements of subjectivity may enter the population projection method. The justifiability of these assumptions can only be decided when the data are available.

Projection, a way to map out possible future cancer scenarios, naturally associates with uncertainty. However, we believe that the results of this study, which are the most reasonable for our present data and the limitations discussed, will provide a useful source for future health planning and evaluation of interventions in Canada.

Implications for future cancer control strategies

The projected aging and growth of the population are expected to cause a progressive and significant increase in the total number of new cases of cancer in Canada over the next 25 years. Consequently, these data indicate the need to continue to strengthen cancer control strategies and leverage resources to meet future health care requirements and reduce the burden of cancer in Canada. Although incidence rates are projected to decrease for many cancers, the rates for some cancers, for example, thyroid, liver, uterus, pancreas, kidney and leukemia, are estimated to increase. Additional etiological research to better understand risk factors and guide prevention efforts is needed.

This monograph underscores the increasing importance of nutrition/diet, physical activity and obesity in relation to cancer prevention as well as the need for continuing efforts to tackle smoking, improve uptake of cancer screening, and increase use of HPV vaccination. The expected effect of future changes in our demographic profiles and cancer trends should be addressed from multidisciplinary perspectives, embracing prevention and early detection, research and surveillance, treatment and psychosocial, palliative and medical care.

Acknowledgements

We would like to acknowledge the contribution of the following people and organizations:

• the Health Statistics Division of Statistics Canada, for providing data from the Canadian Cancer Registry (CCR), and the Canadian provincial/ territorial cancer registries, for providing data to the CCR;
• Dr. Freddie Bray and Dr. Bjorn Moller of the Cancer Registry of Norway and Dr. Michael Otterstatter of the Public Health Agency of Canada for providing methodological support;
• Dr. Eric Holowaty of the Dalla Lana School of Public Health at the University of Toronto, Dr. Hannah K. Weir with the Centers for Disease Control and Prevention (US), and Ms. Amanda Shaw and Dr. Michael Otterstatter of the Public Health Agency of Canada for providing a content review;
• Ms. Lori Anderson, for providing an editorial review;
• The staff members of CancerCare Manitoba who reviewed the risk factor information for the cancer incidence atlas project. This includes Dr. Deepak Pruthi and Dr. Alain Demers along with their site-specific contributors: Dr. Piotr Czaykowski, Dr. Steven Latosinsky, Dr. Robert Lotocki, Dr. Marshall Pitz, Dr. Richard Nason, Dr. Mathew Seftel, Dr. Harminder Singh and Dr. Marni Wiseman.