Health Reports
Progress in net cancer survival in Canada over 20 years

View the most recent version.

Archived Content

Information identified as archived is provided for reference, research or recordkeeping purposes. It is not subject to the Government of Canada Web Standards and has not been altered or updated since it was archived. Please "contact us" to request a format other than those available.

Archived

This page has been archived on the Web.

by Larry F. Ellison

Deriving cancer survival estimates on a population basis permits the monitoring of progress in outcomes over time. Net survival (NS) is the preferred method for comparing cancer survival in population-based cancer studies because it adjusts for the fact that different populations may have different levels of background risk of death.^{Note 1} This measure of cancer survival may be thought of as the survival probability that would be observed in the hypothetical situation where the cancer of interest is the only possible cause of death.

The recent creation of an analytic file (1992 to 2014), wherein incidence data from the Canadian Cancer Registry (CCR) was linked with death information, has provided an excellent opportunity to re-examine progress in cancer survival in Canada across many types of cancer. In comparison to its predecessor, the new file includes six more years of data and was created using additional sources of death information. Moreover, with the maturation of the CCR, it is now possible to report on changes in five-year survival over a 20-year period.

This report provides predicted estimates of NS for the 2012 to 2014 period. NS estimates for durations of 1, 5 and 10 years are derived for 30 of the most commonly diagnosed cancers in people aged 15 to 99. Five-year age-standardized and age-specific estimates for 2012 to 2014 are compared with corresponding figures for cases diagnosed 20 years previously. NS was estimated using the unbiased and consistent Pohar Perme estimator rather than indirectly through relative survival. Traditional methods of estimation using relative survival have recently been shown to result in biased estimates of net survival under certain circumstances.^{Note 2}

Methods

Data sources

The CCR is a dynamic, person-oriented, population-based database comprised 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 for each case they submit.^{Note 3} An analytic file was created by linking the November 2017 CCR tabulation file to mortality information complete through December 31, 2014. This information was obtained from the Canadian Vital Statistics Death database^{Note 4} and T1 Personal Master Files (as reported on tax returns). The analytic file follows the multiple primary coding rules of the International Agency for Research on Cancer.^{Note 5} Survival time is measured in days, while variables for age at diagnosis and diagnosis year are available to three decimal places. More information on the linkage process and on the resulting death-linked analytic file is available upon request.^{Note 6}

Expected survival probabilities, necessary for the calculation of NS, were mostly obtained from sex-specific annual provincial life tables.^{Note 7} As complete life tables were not available for Prince Edward Island or the territories, expected survival for these jurisdictions were derived, up to age 99, from abridged life tables for Canada and the affected jurisdictions^{Note 8} and from complete Canadian life tables^{Note 7} using a method suggested by Dickman et al.^{Note 9} For ages 100 to 109, where this was not possible for these jurisdictions, complete Canadian life table values were directly used.

Analytical techniques

Cases were defined based on the International Classification of Diseases for Oncology, Third Edition^{Note 10} and classified using Surveillance, Epidemiology, and End Results (SEER) Program grouping definitions.^{Note 11} Analyses were based on all primary cancers.^{Note 12Note 13Note 14} with the exception of second- and higher-order primary cancers of the same site.^{Note 15} The relative number of such cases was extremely small but, as expected, increased with the aging of the CCR. Records were also excluded if age at diagnosis was younger than 15 or older than 99; diagnosis had been established through autopsy only or death certificate only; or the year of birth or death was unknown (both extremely rare). Incident cases from the province of Quebec were excluded because cancer incidence data from this province have not been submitted to the CCR since the 2010 data year.

Unstandardized (crude) survival analysis estimates were derived using an algorithm^{Note 16} that has been augmented by Ron Dewar of the Nova Scotia Cancer Care Program (Dewar R 2018, email communication, 19th April) to include the Pohar Perme estimator of NS^{Note 2} using the hazard transformation approach. The following assumptions apply to the estimation of net survival. First, it is assumed that the non-cancer mortality rate in the group being examined is adequately estimated by the mortality rates in the general population life tables. Secondly, conditional independence between the probability of death due to the cancer of interest and the probability of death due to other causes is assumed.

To better satisfy the first assumption, expected survival data used in the calculation of NS for prostate and female breast cancer were adjusted for cancer-specific mortality rates in the general population.^{Note 17Note 18Note 19} In each case, the proportion of deaths among Canadian residents due to the specific cancer, by sex, five-year age group and year of death, was used for the adjustment. Provincial-specific mortality estimates were used for those aged 55 to 59 and older age groups. Otherwise, national estimates were used. The second assumption, concerning the conditional independence of probabilities of death, noted above, may be considered tenuous when analyzing cases diagnosed among the most elderly. For this reason and several others, including the increased variability inherent in the estimates and the limited potential for useful interpretation, both the age-specific description of results and the related trend analyses excluded the 85 to 99 age group.

Age-standardized NS estimates were calculated using the direct method by weighting age-specific estimates for a given cancer to the age distribution of people aged 15 to 99 diagnosed with that cancer. The incident age-distributions were based on cases diagnosed from 2010 to 2014 included in the November 2017 CCR tabulation file (except Quebec). Standard errors for age-standardized NS estimates were estimated by taking the square root of the sum of the squared, weighted, age-specific NS standard errors. See Appendix Table A for a comparison of five-year NS estimates age-standardized using the Canadian cancer survival standard weights described above and, alternatively, weights developed from data collected for the EUROCARE-2 study.^{Note 20} Canadian survival standard weights for individual cancers are provided in Appendix Table B. Corresponding weights for specific groups of cancers (e.g., colorectal cancer, leukemias) are available upon request at STATCAN.infostats-infostats.STATCAN@canada.ca.

NS estimates for the 2012-to-2014 period were calculated using the period method;^{Note 21} for 1992 to 1994 and 2002 to 2004, the cohort method was used. NS estimates were suppressed if the corresponding standard error was greater than 0.1; caution was indicated if the standard error was greater than 0.05 but less than or equal to 0.1. The percentage-point increase in five-year NS was used as the measure of improvement in survival. Differences in NS 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 point estimates. The P values were provided to aid the general interpretation of the differences—in particular to avoid the over-interpretation of potentially large but fairly imprecise differences—rather than as a means to formal hypothesis testing.

Wide variation in survival by cancer type

NS varied considerably across cancer types (Table 1). Long-term NS was highest for thyroid and testicular cancer and lowest for mesothelioma—a cancer of the thin layer of tissue that covers most of the internal organs—and pancreatic cancer. Five-year NS ranged from 98% for thyroid cancer to 7% for mesothelioma. A similar range (from 97% to 5%) was observed for 10-year survival for these cancers. However, when only the first year after diagnosis was considered the range was considerably smaller (from 99% for thyroid to 28% for pancreas).

In addition to the cancers mentioned previously, five-year NS was also particularly high for cancers of the testis (97%) and prostate (93%). On the other end of the spectrum, five-year survival was also especially low for cancers of the esophagus (15%), lung and bronchus (19%), liver (19%), and brain (21%), as well as for acute myeloid leukemia (21%). With the exception of stomach cancer (28%), five-year NS was 44% or greater for all other cancers studied. Cancer survival is generally higher among females than among males (Appendix, Table A). A more extensive evaluation of sex differences in cancer survival in Canada is available elsewhere.^{Note 22}

The largest declines in NS for years two through five post-diagnosis were observed for mesothelioma (34 percentage points), multiple myeloma (32), and ovarian cancer (31). From 5 to 10 years, the largest decline (16 percentage points) was seen for chronic lymphocytic leukemia (CLL) followed by laryngeal cancer (13) and multiple myeloma (12). The relatively poor outlook for CLL over the subsequent five years among those surviving the first five years has previously been noted.^{Note 23} For thyroid and testicular cancer, there was little difference between estimates of NS at 1 year and at 10 years.

Greatest progress for chronic myeloid leukemia

Among the cancers studied herein, the median change in age-standardized five-year NS over the 20 years between the 1992-to-1994 and the 2012-to-2014 periods was 6.1 percentage points (Table 2). At 23.9 percentage points, progress was greatest among those diagnosed with chronic myeloid leukemia (CML). Substantial progress in CML survival in Canada has previously been observed and has been attributed to improvements in treatment.^{Note 24} A large majority of the improvement for CML over this time period, and seemingly, the entire increase for prostate cancer, occurred in the first decade. In contrast, more progress was observed in the most recent decade for both Hodgkin lymphoma and cancer of the lung and bronchus.

Increases of over 15 percentage points were also observed for non-Hodgkin lymphoma (19.6), cancer of the small intestine (17.4) and multiple myeloma (16.8). Progress was similar across the two decades for these cancers. For cancers of the anus, larynx, soft tissue, uterus, and for mesothelioma, predicted five-year NS estimates for 2012 to 2014 were quite similar to corresponding estimates from 20 years earlier.

Decreasing survival with advancing age

NS generally decreases with advancing age. Among the cancers considered in this study, the median decline in five-year NS between the 15-to-44 and 75-to-84 age groups was 24 percentage points (Table 3). The largest declines were observed for both acute myeloid leukemia and brain cancer (59 percentage points), followed by CML (56) and ovarian cancer (54). In general, decreases from one age group to the next tended to be larger between age groups comprised of older individuals.

The pattern in which five-year NS declined from the youngest to the oldest age groups varied considerably between cancers. For the cancers with the largest declines, as well as others like Hodgkin lymphoma, multiple myeloma and cervical cancer, relatively large decreases were observed with almost all progressions from one age group to the next. Another group of cancers had little to modest changes up to the 65-to-74 age group followed by large declines thereafter. This pattern was particularly exemplified by anal and prostate cancer, and to a lesser extent by breast cancer. For cancers of the pancreas and liver, declines were largest between the first and second age groups. Subsequent declines for these cancers were more tempered, in part because NS was already very poor.

Increases in survival across all age groups

Generally speaking, increases in cancer survival were similar across all age groups studied, except for the 75-to-84 age group, where it was somewhat diminished (Table 4). The median increase in age-specific five-year NS between the 1992-to-1994 and the 2012-to-2014 periods ranged from 6.9 (65 to 74) to 8.6 (55 to 64) percentage points among the first four age groups; for the 75-to-84 age group, it was 5.1.

With the exception of those aged 75 to 84, increases in five-year NS over the 20-year period were greatest for CML ranging from 25.6 percentage points (55 to 64) to 44.3 percentage points (45 to 54). In addition to CML, relatively large increases were also noted in all age groups studied for non-Hodgkin lymphoma, multiple myeloma, cancer of the small intestine, and to a lesser degree colon, rectum, liver and prostate cancer. CML, acute myeloid leukemia, non-Hodgkin lymphoma, as well as cancers of the liver, pancreas, and prostate were all associated with increases of over 20 percentage points in the 15-to-44 age group. Among those 75 to 84, thyroid cancer was associated with the largest increase (26.1) followed by non-Hodgkin lymphoma (20.3).

Decreases in five-year NS were noted in the 75-to-84 age group for both anal and brain cancer. Trends in these particular strata were examined in more detail by deriving estimates for three-year overlapping periods (data not shown). For anal cancer, NS estimates declined sharply between the 1993-to-1995 and the 1996-to-1998 periods, rose back to levels seen in the early 1990s by the mid-2000s, then experienced another period of sharp decline. For brain cancer, the decrease appears restricted to a drop between 1996 to 1998 and 1997 to 1999, which itself seems to be mainly driven by data from the populous province of Ontario.

Conclusion

This study highlights the wide variation in NS across many types of cancer while also revealing important areas of progress in cancer outcomes in Canada since the early 1990s. At the same time, it sheds light on cancers for which there has been seemingly no improvement in five-year net survival over a 20-year period.

NS estimates for the 2012-to-2014 period were based on predictions. This is because survival estimates based on the actual follow-up experience of people diagnosed with cancer during these years, especially long-term estimates, will not be known for some time. Predicted estimates were derived from a commonly used methodology that has been shown empirically to provide more up-to-date estimates of survival than would otherwise be available.^{Note 25Note 26Note 27} Still, to the extent that survival appears to have been increasing in the most recent years for a particular cancer or cancers, these predictions may underestimate the survival which will eventually be revealed.^{Note 25Note 28}

Proper interpretation of cancer survival trends can be challenging, and it is best approached on an individual cancer basis, taking many potential factors into consideration.^{Note 29Note 30} While such a task is beyond this study’s scope, this article does provide a general overview of changes in net survival over time in Canada and serves as a starting point for such investigations. Factors that may influence trends in survival include, but are not limited to: improvements in the completeness of case registration; more effective treatments, new screening programs and whether they effect real improvement or simply result in a lead-time bias; changes in the proportion of histologic subtypes where survival differs between the subtypes; completeness of follow-up in addition to methodological factors relating to the length of operation of the cancer registry.^{Note 29Note 30Note 31} Finally, a complete interpretation is best undertaken in conjunction with corresponding cancer incidence and mortality trends.^{Note 29}

Appendix

The use of Canadian weights, rather than ones from an external source, results in a higher degree of interpretability of age-standardized NS estimates—particularly for estimates corresponding to more recently diagnosed cases—and differences in such estimates over time. However, for international comparison purposes Table A provides estimates derived using both Canadian and international standards. The largest differences were observed for cancers of the cervix uteri and brain—six percentage points lower and higher, respectively, when the international standard weights were used.