|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Originally published as JCO Early Release 10.1200/JCO.2004.11.972 on December 22 2003 © 2004 American Society of Clinical Oncology.
Do Estimates of Long-Term Survival Tell Us Whether Patients Diagnosed With Breast Cancer Before Age 50 Years Are Ever Cured?Fred Hutchinson Cancer Research Center, Seattle, WA In this issue of the Journal of Clinical Oncology, Brenner and Hakulinen examine absolute and relative survival in cohorts of Finnish breast cancer patients diagnosed before age 50 years [1]. Survival improved over the decades, with relative survival estimates at 10 years increasing from 49% in the 1953 through 1959 cohort to 68% in the 1983 through 1989 cohort, which was associated with similar increases in absolute survival. Long-term follow-up is not available for the most recent cohort, 1993 through 1999. Forty-year estimates for this group have been calculated using a method known as period analysis, which uses the most recent years with available complete information to estimate conditional survival (conditional on having survived the previous interval) for each successive interval without follow-up. The authors cite a previous article [2] to support their contention that this method provides accurate estimates. Based on the observation that relative survival estimates (relative to age-adjusted population estimates) remain less than 1.0 for each decade through 40 years, the authors conclude that breast cancer patients remain at higher risk of death over their entire lifetime. Although the observations are quite interesting, interpretation of them is problematic. Even if the period analysis method of estimation is an improvement over cohort analysis, it cannot be expected to adequately reflect expected survival of current breast cancer patients over 40 years. As noted in the article, conditional survival still must be estimated from patients diagnosed decades ago, at times when disease detection and management may have been quite different. The authors' previous article addresses only 10-year survival estimates. Although superior to cohort analysis, even for this relatively short interval, period analysis did not accurately predict breast cancer survival for that population. Estimates might reasonably be expected to become successively worse over the decades as older and older data are used. To illustrate how inaccurate period analysis can be, consider the following scenarios. For simplicity, exponential survival is assumed, with changes occurring at the beginning of decades. The hazard used for the initial decade is 0.1, which corresponds to a 7-year median. The four scenarios describe different patterns over the last four decades and four future decades: (1) no change over the decades, (2) no change until the fifth decade when a cure is discovered (hazard, 0.025; median, 28 years), (3) 10% improvement at the beginning of each decade, and (4) 5% worsening at the beginning of each decade (eg, owing to disease or definition drift). Table 1 shows the true value for patients diagnosed in the past year (the first year of the decade) compared with future estimates using period analysis. Also, for simplicity, the most recent year with complete data is used for the period estimates. As the table illustrates, the estimates can either be better or worse than reality, the estimates get proportionally worse over the decades, and the discrepancies between estimates and reality can be large if there is a major change in death rate after the periods on which estimates are based. For example, period analysis predicts 2% survival at 40 years for the cure scenario, when the true 40-year survival rate is 37%.
Even if we assume that the estimates are accurate, the title of the article is misleading. The analysis does not address cure, as higher death rates are not necessarily attributable to cancer. For instance, long-term cardiac damage, immune compromise, or secondary cancers could all contribute to elevated risk of death, even if patients are cured of breast cancer. Another possibility is that a subset of patients is cured, but higher rates are due to a subset that is not. Consider the following survival curves comparing population survival with cancer patient survival. For simplicity, exponential survival is again used, with a population hazard of 0.025 and an unimproved breast cancer hazard of 0.1 (ie, death rate four times that of the general population). Figure 1A represents the simplest relationship, with death rate for breast cancer patients improved to 1.8 times that of the general population throughout 40 years. As shown in Figure 1B, 30% of patients die at the breast cancer rate, whereas 70% are cured of breast cancer but die at 1.4 times the rate of the general population. In Figure 1C, 50% of patients are cured, with death rate the same as that of the general population. Figure 1D shows the relative survivals for the three models. There are small visual differences among the three, but the differences are not striking and are certainly not sufficient to distinguish cure from no-cure by inspection, when variability introduced from real data is added.
Use of relative conditional survivals (as in Table 2 of the article by Brenner and Hakulinen) is helpful but still does not necessarily allow identification of a cure. Table 2 provides some examples with more complex structure than assumed for the previous examples. Population conditional probabilities for this Table are assumed to decrease over time from 0.8 for the conditional probability of surviving to 20 years given survival to 10 years, to 0.7 for the conditional probability of 30 years given survival for 20 years and 0.5 for the conditional probability of 40 years given survival for 30 years. Table 2 shows relative survivals for three cases of cure, and three of no cure. As can be seen from the Table, relative survival remains less than 1.0 throughout the decades, even in the presence of cure. Both cure and no-cure cases can be accompanied by increasing, decreasing, or stable relative conditional survivals depending on the patterns of death rates in the cured and not cured patients, and similar patterns occur for each.
The observation of decreased relative conditional survival in the last decade reported in this article seems particularly difficult to interpret, because rates of relapse generally decrease over time. Rate estimates from a Southwest Oncology Group study [3] with more than 20 years of follow-up are 0.16 per person years of follow-up for 0 to 2 years, 0.10 for 2 to 5 years, 0.04 for 5 to 10 years, and 0.02 for 10 to 20 years. Although there are late relapses reported on this study, after a median follow-up of 22 years, the majority of events are deaths without relapse. An increase in relapses in the fourth decade would seem surprising; therefore, the decreased relative survival rate would seem likely to reflect an increase in death owing to other causes. As suggested by the authors, it might indeed be of interest to study late causes of death, although changes in treatment might make conclusions more of academic than practical interest. As much as we would like to accurately estimate future survival for patients today, outdated information can take us only so far. If our assumptions are incorrect, our estimates will be wrong. Assuming that conditional survival estimates are more stable than absolute estimates may be reasonable, but more stable still does not mean accurate. Other information, such as average improvement over time, could possibly be used to formulate other assumptions to generate better estimates. However, any estimate of survival 40 years from now is necessarily speculative and dependent on the quality of our guesses regarding the future. Author's Disclosures of Potential Conflicts of Interest The author indicated no potential conflicts of interest.
REFERENCES
1. Brenner H, Hakulinen T: Are patients diagnosed with breast cancer before age 50 ever cured? J Clin Oncol 22:432-438, 2004
2. Brenner H, Hakulinen T: Up-to-date long term survival estimates of patients with cancer by period analysis. J Clin Oncol 20:826-832, 2002 3. Rivkin S, Green S, Lew D, et al: Adjuvant chemotherapy with cyclophosphamide, methotrexate, and 5-flourouracil, vincristine, and prednisone compared with single-agent L-phenylalanine mustard for patients with operable breast carcinoma and positive axillary lymph nodes. Cancer 97:21-29, 2003[Medline]
Related Article
Related Correspondence
This article has been cited by other articles:
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2004 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
