Conditional survival analyses across cancer sites

by Larry F. Ellison, Heather Bryant, Gina Lockwood and Lorraine Shack

Survival statistics are an indicator of the effectiveness of cancer detection and treatment.1  These statistics are used to compare cancer control over time2 and across jurisdictions.3,4  They are also of interest to clinicians providing direct care and to patients, who usually want an estimate of their prognosis.5

Survival estimates are typically presented as the probability—or the ratio of observed and expected probabilities in the case of relative survival—of surviving a given length of time (for example, five years) after diagnosis.  However, these estimates are less informative for people who have survived one or more years, as the risk of death due to cancer is often greatest in the first few years.  After this initial period, the prognosis can improve substantially, so the earlier estimates no longer apply.6 The outlook for such people can be estimated more appropriately using conditional survival.

For the first time in Canada, predicted conditional relative survival estimates are presented for a large number of cancers.  Cancers with the greatest relative improvement in prognosis since diagnosis are highlighted.  Cancers showing less improvement are also identified. (See The data)

Prognosis improves over time

For almost all the individual cancers studied, the relative probability of living an additional five years improved when measured at increasingly longer periods after diagnosis, the effect being strongest in the first one to two years (Table 1).  A notable exception was chronic lymphocytic leukemia (CLL), for which five-year relative survival was just under 80% at diagnosis and did not appear to improve over the subsequent five years.

Table 1 Predicted five-year relative survival ratios (RSR), by type of cancer and conditional on having survived up to five years, Canada excluding Quebec, 2004 to 2006Table 1 Predicted five-year relative survival ratios (RSR), by type of cancer and conditional on having survived up to five years, Canada excluding Quebec, 2004 to 2006

After five years' survival, the conditional five-year relative survival ratio (RSR) had risen to at least 95% for the cancers with an initial five-year RSR of at least 80%.  The exception was breast cancer (five-year RSR of 93% after five years).  Thyroid, prostate and testicular cancers had five-year prognoses of 95% or more at diagnosis; for skin melanoma and cancer of the corpus uteri, this level was achieved after three years, and for Hodgkin lymphoma, it took five years.  A five-year conditional RSR of 90% or higher was achieved for breast cancer after two years, and by one year for all other cancers in this group.

Among cancers with five-year RSRs of 50% to 79% at diagnosis, a five-year conditional RSR of 95% or higher was observed for cancers of the cervix uteri and colon after five years.  A five-year conditional RSR of 90% or higher was achieved for cancers of the bladder, kidney and renal pelvis, and soft tissue after three years, and for rectal cancer, after four years.  It was not achieved for the other cancers with initial five-year RSRs of 50% to 79%, although improvements from the mid-sixty percents at diagnosis to the mid-eighty percents among those surviving the first five years were noted for oral and laryngeal cancers, and for non-Hodgkin lymphoma.  The apparent lack of improvement in prognosis over time for people diagnosed with CLL was unique among the cancers associated with at least a little excess mortality at diagnosis.  This is consistent with findings reported in a recent study using Surveillance, Epidemiology and End Results data from the United States.7

A much better expectation of continued survival than at diagnosis was achieved for all cancers for which the initial five-year relative prognosis was less than 50% (Figure 1).  Nonetheless, the five-year conditional RSR remained below 90% five years after diagnosis for all but two cancers in this group.  Stomach cancer and leukemia (excluding CLL) reached a 90% five-year conditional RSR after five years' survival—despite original prognoses of 24% and 34%, respectively.  The lowest five-year relative survival estimates at diagnosis were for cancers of the esophagus (13%) and pancreas (6%); however, by the fifth year of survival, both had five-year conditional RSRs in the mid- to upper-eighty percent range.  The corresponding improvement for multiple myeloma (from 37% to 60%) was more modest.

Figure 1 Five-year predicted conditional relative survival ratios, selected cancers, Canada excluding Quebec, 2004 to 2006Figure 1 Five-year predicted conditional relative survival ratios, selected cancers, Canada excluding Quebec, 2004 to 2006

Ranking of five-year RSRs

Cancers of the pancreas and colon showed the most striking improvements in the ranking of five-year RSRs from diagnosis to five years later (Table 2).  Among the cancers studied here, pancreatic cancer ranked 26th at diagnosis, but given five years' survival, it ranked 16th.  Colon cancer rose from 15th to 6th position.  Cancers of the stomach and esophagus, and leukemia (excluding CLL) improved seven, six and five positions, respectively.

Table 2 Relative rank of cancer type based on predicted five-year relative survival ratio (RSR) at diagnosis and conditional on having survived five years, Canada excluding Quebec, 2004 to 2006Table 2 Relative rank of cancer type based on predicted five-year relative survival ratio (RSR) at diagnosis and conditional on having survived five years, Canada excluding Quebec, 2004 to 2006

A substantial drop in ranking―from 8th to 23rd position―was observed for CLL, due largely to the lack of a predicted increase in relative survival among those surviving the first five years after diagnosis.  The rankings of multiple myeloma, breast and larynx each fell six to seven positions, indicating that survival had not improved to the same degree as for many other cancers.

Patterns similar for males and females

Cancer-specific patterns in relative survival conditional on surviving one to five years after diagnosis were generally similar by sex (data not shown).  For example, at diagnosis, the five-year RSRs for rectal cancer were 63% for males and 65% for females; at five years' survival, the figures were 92% and 94%, respectively. 

In some instances, an apparent sex-specific survival advantage at diagnosis disappeared during the first five years of follow-up.  Five-year RSRs at diagnosis were at least four percentage points higher for females than males for skin melanoma, lung and bronchial cancer, Hodgkin lymphoma and brain cancer, but differed by no more than one percentage point at or before the fifth anniversary of diagnosis (data not shown).  Similarly, an apparent survival advantage for men diagnosed with bladder cancer disappeared relatively soon after diagnosis (data not shown). 

Conclusion

The conditional survival estimates presented here are population-based, and therefore, reflect the average survival time of large groups of people rather than an individual's prognosis.  Even so, the figures are a useful update of the initial prognosis for a number of cancers, and are generally a cause for optimism.  For most cancers, the outlook for people who have survived one or more years after diagnosis is better than that at diagnosis, sometimes substantially so.  For some cancers for which survival was already very promising, such as thyroid cancer, little additional improvement could be expected. 

Conditional survival data provide more accurate prognostic information about how the risk of death changes over time.  These results could assist people who have survived one or more years after a cancer diagnosis in adjusting their view of the future, and help cancer care providers in planning follow-up.  Calculations of conditional survival for individual cancers by subsite or histological subtype and by age group would further inform clinical understanding.

Acknowledgements

The Canadian Cancer Registry is maintained by Statistics Canada. It is comprised of data supplied by the provincial and territorial cancer registries whose cooperation is gratefully acknowledged.  The authors thank Kathryn Wilkins of Statistics Canada for her review of the manuscript.