Estimating breast cancer-specific and other-cause mortality in clinical trial and population-based cancer registry cohorts

Abstract

BACKGROUND:

To compute net cancer-specific survival rates using population data sources (eg, the National Cancer Institute's Surveillance, Epidemiology, and End Results [SEER] Program), 2 approaches primarily are used: relative survival (observed survival adjusted for life expectancy) and cause-specific survival based on death certificates. The authors of this report evaluated the performance of these estimates relative to a third approach based on detailed clinical follow-up history.

METHODS:

By using data from Cancer Cooperative Group clinical trials in breast cancer, the authors estimated 1) relative survival, 2) breast cancer-specific survival (BCSS) determined from death certificates, and 3) BCSS obtained by attributing cause according to clinical events after diagnosis, which, for this analysis was considered the benchmark “true” estimate. Noncancer life expectancy also was compared between trial participants, SEER registry patients, and the general population.

RESULTS:

Among trial patients, relative survival overestimated true BCSS in patients with lymph node-negative breast cancer; whereas, in patients with lymph node-positive breast cancer, the 2 estimates were similar. For higher risk patients (younger age, larger tumors), relative survival accurately estimated true BCSS. In lower risk patients, death certificate BCSS was more accurate than relative survival. Noncancer life expectancy was more favorable among trial participants than in the general population and among SEER patients. Tumor size at diagnosis, which is a potential surrogate for screening use, partially accounted for this difference.

CONCLUSIONS:

In the clinical trials, relative survival accurately estimated BCSS in patients who had higher risk disease despite more favorable other-cause mortality than the population at large. In patients with lower risk disease, the estimate using death certificate information was more accurate. For SEER data and other data sources where detailed postdiagnosis clinical history was unavailable, death certificate-based estimates of cause-specific survival may be a superior choice. Cancer 2009. © 2009 American Cancer Society.

For individuals with a diagnosis of cancer, it would be ideal to have reliable disease-specific mortality information, for example, an estimate of the median survival from diagnosis to death from cancer or the proportion of individuals who will likely die of the cancer by a certain time landmark, such as 5 years. It is believed that these measures reflect the impact of the disease better than all-cause survival, which varies widely by demographic factors, such as age and race, and may exhibit geographic and temporal trends. Thus, methods have been developed to estimate quantities representing the net influence of the cancer on the lifetime, and these frequently are used in the reporting of cancer survival trends. For example, the Surveillance, Epidemiology and End Results (SEER) Program, a geographically based cancer registry cohort of the US National Cancer Institute, reports net cancer survival obtained by relative survival estimates, an established method (described below) that does not depend on patient-specific cause of death (COD) information.1, 2 This method, as well as alternatives that rely on attributed cause of death, suffers from shortcomings that may limit the degree to which true cancer-specific survival metrics are obtained.

Once the specific COD is determined, then (under certain assumptions) calculating cause-specific mortality is relatively straightforward.3 However, COD can be surprisingly difficult to ascertain accurately in population data. For example, in the SEER Program and similar registries, COD may be unreported, erroneous because of a tendency toward over attribution to the cancer once it is diagnosed, or simply too ambiguous with respect to attributing a single cause, requiring some subjective rule for resolution.4-6 In multicenter clinical trials, similar problems arise in establishing a COD.7, 8 Specifically, COD information can be incomplete or inaccurate and often is not reviewed rigorously or coded by nosologists. For this reason and others, clinical trials commonly report all-cause survival as the primary endpoint.

The relative survival estimate is a frequently used alternative that does not require COD information and accounts for demographic variations in expected mortality.1, 2 Relative survival estimates equal observed all-cause survival divided by expected survival estimated from the general population in a comparable geographic area with comparable age, sex, race, and calendar year. Relative survival typically is reported by SEER for cancer survival estimates9, 10 and has been used widely to examine temporal and geographic trends in cancer survival rates.11, 12 However, this approach relies on population-expected survival data, which may lack sufficient representativeness of the cancer patients at hand. For example, it is well known that patients who are diagnosed with early stage colorectal or prostate cancer at a screening examination may tend to have lower other-cause mortality than the general population. The consequence of then using a population adjustment for expected survival in the cancer patient cohort is an overestimate of cancer-specific survival. For clinical trials, relative survival is not used typically; because, although it aims to be representative, participant selection is not population-based in any strict sense, so the appropriate expected survival adjustment is unknown.

In the current study, we addressed several questions related to the estimation of cancer-specific survival using data from multicenter cancer clinical trials that were conducted over the past 20 years as well as data from the SEER Program for comparable year periods. First, in the clinical trial cohort, we compared relative survival with cancer-specific survival as determined from clinical history (ie, disease progression events after initial diagnosis), which, for the purposes of this study, we considered the true cancer-specific survival. We also compared survival estimates based on reported COD from death certificates with relative survival estimates. Second, we examined relative survival and cancer-specific survival as determined by death certificate information in comparable SEER cohorts. Finally, to evaluate the extent to which both clinical trial cohorts and patients included in SEER may reflect the population at large with respect to noncancer mortality, we estimated noncancer life expectancy in these respective groups.

MATERIALS AND METHODS

Patient Cohorts

The National Surgical Adjuvant Breast and Bowel Project (NSABP) is a National Cancer Institute-sponsored Cancer Cooperative Group with over 500 participating clinical centers throughout North America.13 Institutions that enroll patients include university hospitals, cancer research centers, and community-based hospitals and practices. For this study, we included randomized trials for early invasive breast cancer (ie, no metastatic disease) that were open to patient accrual between 1982 and 1998. These trials investigated adjuvant systemic therapy regimens (tamoxifen, multidrug cytotoxic chemotherapy) that feasibly could be delivered in a variety of care settings. Trial eligibility criteria included operable unilateral breast cancer (with or without axillary lymph node involvement, depending on the specific trial), no prior history cancer, health status consistent with the ability to undergo adjuvant therapy, and specific hormone receptor criteria for some trials. Follow-up continued for at least 15 years on earlier trials, and more recent studies remain in active follow-up.

From the SEER 9 registries, we selected patients from diagnosis years 1989 through 2003 who had follow-up through 2004, excluding patients who had intraductal disease (ductal carcinoma in situ [DCIS]), locally advanced (stage III) disease, and metastatic (stage IV) disease. Patients were grouped by axillary lymph node status (denoted lymph node-negative or lymph node-positive) for comparability with the NSABP trials, which used lymph node status as the principal determinant of eligibility for individual trials. In the SEER Program, lymph node status was available beginning in 1989; and, at the time of this analysis, 2004 was the most recent follow-up year available. Consequently, the maximum follow-up length for the cohort was 15 years.

The clinical trial cohort consisted of 20,737 patients from 11 clinical trials (Table 1). Among these patients, there have been 6676 deaths; among the lymph node-negative patients, 47% of deaths were attributed to breast cancer (according to death certificates), and 22% had missing cause compared with 71% breast cancer COD and 18% missing cause among lymph node-positive patients. Patients from SEER tended to be older at diagnosis; had smaller tumors; and, among lymph node-positive patients, had a smaller number of positive lymph nodes (Table 1). The proportion of deaths attributable to breast cancer (based on death certificates) was lower for the SEER cohort, particularly for lymph node-negativepatients.

Table 1. Characteristics of Patients From National Surgical Adjuvant Breast and Bowel Clinical Trials and the Surveillance, Epidemiology, and End Results Cohort

Characteristic	Trials: Diagnosis 1982-1998				SEER: Diagnosis 1989-2003
	LN Negative		LN Positive		LN Negative		LN Positive
	No.	%	No.	%	No.	%	No.	%
SEER indicates Surveillance, Epidemiology, and End Results; LN, lymph node; AA, African American; COD, cause of death. * The percentage of patients who died.
Age at diagnosis, y
<50	4466	43.7	5729	54.5	27,657	23.6	16,199	32.3
50-64	4289	41.9	4013	38.2	41,187	35.2	17,633	35.2
≥65	1473	14.4	768	7.3	48,193	41.2	16,257	32.5
Race/ethnicity
White/Caucasian	8847	86.5	8971	85.4	100,785	86.1	41,984	83.8
Black/AA	802	7.8	930	8.8	7777	6.6	4534	9.1
Other/unknown	579	5.7	609	5.8	8475	7.2	3571	7.1
Tumor size, cm
≤2	6282	61.4	4370	41.6	89,047	76.1	25,333	50.6
2.1-4.9	3501	34.2	4910	46.7	24,886	21.3	23,115	46.2
≥5	444	4.3	1230	11.7	3104	2.7	1641	3.3
Lymph node status
Negative	10,228	100	0	0	117,037	100	0	0
Positive, no.
1-3	0	0	6284	59.8	0		34,491	69.3
4-9	0	0	3064	29.2	0		10,648	21.2
≥10	0	0	1147	11	0		4626	9.3
Treatment
Tamoxifen	5343	52.2	5374	51.1	—	—	—	—
Chemotherapy	5272	51.5	10,089	96	—	—	—	—
Vital status
Alive	7820	76.5	6242	59.4	93,878	80.2	34,105	68.1
Dead	2408	24.5	4268	40.6	23,159	19.8	15,984	31.9
Death certificate COD*
Breast cancer	1150	47.8	3015	70.6	6629	28.6	9656	60.4
Other cause	718	29.8	468	11	15,623	67.4	5795	36.3
Unknown	540	22.4	785	18.4	907	3.9	533	3.3

Methods for Estimating Breast Cancer-Specific Survival

Three methods used to estimate breast cancer-specific survival are described here and are summarized in Table 2.

Table 2. Methods Used to Estimate Breast Cancer-Specific Survival

Method	Description	Comments
COD indicates cause of death.
Relative survival	Observed all-cause survival in the cancer patient cohort was divided by survival for a cohort with comparable age, race, sex, and year from an appropriate geographic region	Does not use or require COD information; external reference population must be suitably representative of cancer cohort
Death certificate-based survival	Survival was computed using death with cause reported as breast cancer (or unknown) as the event; other-cause deaths were treated as censored	Depends on the accuracy of reported COD; how missing CODs are treated will influence estimate (in this study, they were treated as breast cancer deaths)
Recurrence algorithm-based survival	Survival was computed by assigning deaths preceded by breast cancer events (recurrence, new primary) as breast cancer deaths; deaths without breast cancer events after initial diagnosis were treated as censored	Requires clinical event history—whether cancer recurred after initial diagnosis; information about time and type of recurrence also will be informative

Estimates based on clinical history

Patients in NSABP trials have regular protocol-specified follow-up, and breast cancer recurrence events at these visits or in intervening intervals are documented and reported. The protocol calls for reporting of the first recurrence at any anatomic site (local, regional, or distant) and the first distant metastatic recurrence; although, typically, all events that occur (ie, multiple distant site failures) are reported. Second primary cancers also are reported. By examining this clinical history, deaths can be classified as either 1) subsequent to a breast cancer event (recurrence or second primary [contralateral] breast cancer) or 2) without documentation of any preceding breast cancer events. For the current analysis, we considered all deaths of the former type as breast cancer-related and considered the latter as deaths from causes other than breast cancer. This estimate (denoted recurrence algorithm-based survival) was used as the benchmark estimate for breast cancer-specific survival to which relative survival and death certificate-based estimates were compared in the trial cohort.

Estimates based on death certificates

By using International Classification of Disease (ICD) codes provided on death certificates, COD was assigned as either breast cancer (ICD code 174.0) or other causes. In the NSABP trials, COD codes were missing in 18% to 22% of patients (Table 1). In the trials, whether the COD was missing was unrelated to the clinical history (recurrence events, second cancers), because the proportion of patients who had recurrent disease was similar to the proportion that reported COD (not shown). For an estimate based on COD, we adopted the conservative approach of treating deaths with unknown COD as breast cancer-related, providing a lower bound estimate of breast cancer-specific survival (assuming COD assignment is otherwise correct).

For the SEER cohort, COD codes were unknown for approximately 3% of patients (Table 1). In an analysis of earlier SEER data in which the proportion with missing codes was greater, Gamel and Vogel evaluated different ways to allocate unknown cause deaths and recommended a proportional allocation between causes.14 Because the proportions missing in the current SEER data are small, estimates do not differ by greater than 2 percentage points through 15 years regardless of whether the missing COD is from breast cancer death or from other causes. Thus, like in the trials, we attributed deaths with missing COD to breast cancer for the death certificate-based estimates.

For both the recurrence algorithm and the death certificate method of assigning COD, breast cancer-specific survival was computed by using life-table (actuarial) methods.15 Specifically, we treated other-cause deaths as censored observations, and, assuming independent competing risks, we were able to interpret this estimate as a net survival function.3

Estimates using relative survival

Relative survival estimates were computed for both the NSABP patients and the SEER patients. For relative survival analysis, observed all-cause survival estimates were divided by age, race, sex, and calendar year-specific life expectancy (from US life tables16) to produce estimates that are interpreted commonly as (approximately) breast cancer-specific survival, accounting for strata-specific differences in overall life expectancy. Interval-specific estimates were restricted to ≤1.0 (ie, 100%); thus, all relative survival curves are nonincreasing. The SEER*STAT software package (available at: http://seer.cancer.gov/seerstat/ accessed on July 28, 2009) was used to produce these estimates.

Estimation of Noncancer Life Expectancy

To investigate the representation of the noncancer (other cause) life expectancy of the cancer patients in the NSABP trials and the SEER data compared with the general population, other-cause life expectancy for cancer patients was estimated by treating patient age as the time axis and estimating a left-truncated Kaplan-Meier survival function with other-cause deaths treated as the event, breast cancer deaths treated as censored, and age at diagnosis treated as the left truncation point. The resulting survival curve represents the probability of noncancer survival conditional on survival to the earliest age at diagnosis (eg, the youngest patient). For comparison, overall life expectancy curves for the US population in the years 1990 and 2000 obtained from US life tables (National Center for Health Statistics) were averaged at each age and plotted. Differences in life expectancy curves between the cancer patient cohorts and the general population indicate whether the former tend to have more or less favorable other-cause mortality relative to the general population. Note that, because the actual proportion of cancer deaths in the population is small, the all-cause life expectancy for the general population provides a suitable estimate of noncancer life expectancy.1

RESULTS

Comparison of Breast Cancer Survival Estimates

Clinical trial cohort

We computed recurrence algorithm-based breast cancer survival estimates (breast cancer deaths based on prior clinical history), relative survival estimates, and estimates based on reported COD from death certificates for the clinical trial cohort data separately by lymph node status. Among women with lymph node-negative breast cancer, relative survival estimates were greater than the algorithm-based estimate, with absolute differences of 2 percentage points at 5 years and 2.9 percentage points at 10 years (Fig. 1A). In women with lymph node-positive disease, in which a greater proportion of deaths are likely to be caused by breast cancer, relative survival and the algorithm-based survival estimates were more similar, with an absolute difference of 1 percentage point at 5 years and 1.4 percentage points at 10 years (Fig. 1B). For estimates based on death certificates, the resulting survival estimate slightly underestimated the algorithm-based breast cancer survival in lymph node negative patients (Fig. 1A) and more closely resembled the relative survival and algorithm-based estimates in lymph node-positive patients (Fig. 1B).

Figure 1. Recurrence algorithm-based survival, relative survival, and breast cancer-specific survival estimates are illustrated based on death certificate (Death cert.) information on patients with breast cancer from (A,B) clinical trials and (C,D) the Surveillance, Epidemiology, and End Results Program. COD indicates cause of death.

Surveillance, Epidemiology, and End Results cohort

In the SEER cohort, only death certificate-based survival and relative survival are estimable, because recurrence history is not available. In lymph node-negative patients, the difference between death certificate-based and relative survival rates resemble those of the trial data, in that the relative survival estimate is higher (Fig. 1C). For lymph node-positive patients, these estimates are nearly identical, similar to the results for the trial cohort (Fig. 1D).

Estimates Stratified by Patient and Disease Characteristics

Discrepancies between relative survival and true disease-specific survival often are attributable to dissimilarities between the “adjustment” population and the actual noncancer survival experience of the cohort. Specifically, greater health resource access can result in both earlier stage cancer diagnosis with more favorable features (ie, small tumors) and more favorable other-cause mortality. We explored this possible relation between breast cancer characteristics and overall mortality risk by examining estimates within patient and tumor characteristic strata for the clinical trial cohort. In addition to lymph node status, the strata are defined based on commonly used breast cancer prognostic groupings for age at diagnosis (ages <50 years, 50-64 years, and ≥65 years) and tumor size (≤2 cm, 2.1-4.9 cm, ≥5 cm).

Figure 2 shows algorithm-based and relative survival estimates within age and tumor size groups. For younger patients (aged <50 years), in whom any deaths that occur are more likely to be caused by breast cancer than in older patients (both because young patients are expected to have lower other-cause mortality and because their disease tends to be more aggressive), relative survival closely estimates the algorithm-based estimate. In patients with large tumors (≥5 cm), in whom screening is less likely the sole detection method, the 2 estimates also are very similar. For patients with disease characteristics indicative of high mortality risk, such as >10 positive lymph nodes, the relative and algorithm-based survival estimates were nearly identical (not shown). In contrast, for older patients and for those with smaller tumors, the 2 estimates deviate more widely, with relative survival overestimating algorithm-based survival. For some subsets, such as lymph node-negative patients aged ≥65 years, the relative survival estimate equals 100%, indicating that overall survival in the trial cohort (including cancer deaths) is greater than the expected survival based on the age-comparable general population. This problematic behavior is sometimes observed in very early screen-detected cancers, such as DCIS.

Figure 2. Survival estimates are illustrated by age at diagnosis (A-C) and tumor size (D-F) for patients from clinical trials. SEER indicates Surveillance, Epidemiology, and End Results.

For both lymph node status groups, the death certificate-based estimate that treated a missing cause as breast cancer deviated more widely from the algorithm-based estimate as prognosis improved and/or as the patient's likelihood of other-cause death increased (ie, older age at diagnosis, smaller tumors) (Fig. 2C,D), because a greater proportion of these missing cause cases did not have prior recurrence, which the algorithm-based approach takes into account.

Because overall mortality differs by race/ethnicity (as does cancer-specific mortality in population-based studies), we examined differences between relative survival and breast cancer-specific survival separately for blacks and whites. The approaches did not appear to perform differently by race/ethnicity (results not shown).

Noncancer Life Expectancy

A critical issue in the use of relative survival and the interpretation of clinical trials results in general is the health status of participants aside from their cancer diagnosis. For example, if other-cause life expectancy for trial participants is greater than that for the reference population, for example, because they have better healthcare use, then relative survival estimates that incorporate expected survival from this reference population are likely to be inflated.

We computed noncancer life expectancy for the NSABP trials and SEER data and compared these estimates with the life expectancy of the United States female population (separately by race) obtained from life tables. In patients with lymph node-negative breast cancer, noncancer life expectancy among trial participants was greater than that for the population (Fig. 3A,C). The SEER patients had greater but somewhat more similar noncancer life expectancy relative to the population that coincided approximately with the population until about age 70 years for whites and age 75 years for blacks; then, it was higher but was not as favorable as the life expectancy of the trial participants. In patients with lymph node-positive disease, the noncancer life expectancy of trial participants again was greater than that of the population, but the noncancer life expectancy of the SEER patients was lower (Fig. 3B,D). Next, we examined estimates separately by tumor size at diagnosis, which may serve as a surrogate for breast screening activity and, consequently, greater health resource use. This analysis was restricted to whites only, because race and tumor size are associated. Among lymph node-negative patients, trial participants with small tumors had superior life expectancy to the population; whereas, for patients who had large tumors, patients in the trial and SEER cohorts had similar life expectancy similar to that of the population (Fig. 4A-C). When lymph node-positive patients were stratified by tumor size, the SEER patients had a trend toward less favorable life expectancy relative to the population with increasing tumor size (Fig. 4D-F), which may have led to an underestimation of cause-specific survival. Among trial participants, life expectancy remained greater than in the population for all tumor size groups.

Figure 3. Noncancer life expectancy estimates are illustrated for (A,C) patients with lymph node-negative disease and (B,D) patients with lymph node-positive disease (B,D). SEER indicates Surveillance, Epidemiology, and End Results.

Figure 4. Noncancer life expectancy estimates in whites stratified by tumor size at diagnosis are illustrated for (A-C) patients with lymph node-negative disease and (D-F) patients with lymph node-positive disease. SEER indicates Surveillance, Epidemiology, and End Results.

DISCUSSION

Relative survival estimates frequently are used to report net cancer survival statistics, despite some anticipated shortcomings for certain disease sites and types of patients. In the current study, we observed that relative survival overestimates breast cancer-specific survival in clinical trial participants with lower risk (lymph node-negative) breast cancer, particularly among those who are older and those with small tumors (Fig. 2C,D). In contrast, among trial participants with a less favorable prognosis (lymph node-positive or lymph node-negative with large tumors) or when breast cancer is the predominant COD, such as in patients aged <50 years, relative survival and breast cancer-specific survival were remarkably comparable (Fig. 2A,F). Similar results were observed in the SEER patient cohort when we compared breast cancer-specific survival (based on death certificate information) with relative survival. Thus, as the putative risk of breast cancer death increased, the 2 approaches yielded similar estimates; whereas, for lower risk patients, relative survival produced overestimates. These observations support the use of relative survival in trials and in data sources like SEER in instances with high cancer-specific mortality, but we suggest that caution is warranted in when the cancer mortality risk may be lower.

To extrapolate observations from either observational cohorts or clinical trials to the cancer patient population, the cohorts must be suitably representative. In the case of clinical trials, participants often differ in important ways from the population at large; and, although within-trial treatment comparisons nonetheless are valid, treatment efficacy realized in the trial may be diminished in the population if trial participants are highly nonrepresentative. In the current study, trials participants had more favorable noncancer life expectancy than the general population (Fig. 3A,B), reflecting a likely “healthy participant” effect with respect to morbidities. It is noteworthy that, when comparisons were stratified by tumor size at diagnosis, the greatest discrepancy was observed for the women with small tumors (Fig. 4A,D); whereas, for large tumors, the estimates were similar (Fig. 4C,F). For lymph node-positive patients with large tumors from the SEER Program, life expectancy was less favorable than that in the general population, possibly reflecting the finding that women with large tumors may have postponed responding to symptoms or using screening and concurrently also may have other untreated health problems (Fig. 4F). This situation may be less likely to occur in clinical trials because of eligibility requirements with respect to health status. In general, the greater degree of similarity between the SEER cohort and the general population (Fig. 3C,D) supports using relative survival as an estimate of cancer-specific survival from SEER.

Even under ideal conditions, the assignment of a single, definitive COD can be difficult and necessarily is subjective to some degree. In large, geographically dispersed registries like SEER and in multicenter clinical trials, additional limitations result from incomplete records or inconsistent rules with respect to cause attribution on death certificates. Although an independent review panel could apply well defined rules and could attribute COD in an internally consistent manner, the process is labor intensive and is limited by the information centrally available. Relative survival rates provide an alternative that does not require COD assignment but may be inaccurate when the cancer patient cohort is not comparable to the reference population. For cancers that can be detected by screening before any symptomatic evidence, such as breast, prostate, and early stage colon cancers, those patients who are diagnosed at a very early stage tend to have better health status than the general population (because they have access to and make use of preventative medical services with greater frequency); thus, their noncancer morbidity and mortality are lower than those in the general population. Then, when a standard reference population is used to adjust observed all-cause survival, the resulting cancer-specific survival curve is overly high. This phenomenon most likely explains the overestimates observed among individuals with small tumors and those aged ≥65 years at diagnosis, and it also resembles the association between DCIS diagnosis and more favorable cardiovascular disease risk.17 To address the effects of heterogeneity in life expectancy when computing relative survival, numerous stratification and modeling approaches have been proposed.18-21

There are limitations to the approach used to determine true breast cancer survival, which we considered the reference standard to which relative survival was compared. Our tautological definition of breast cancer death (death after a recurrence event) will result in some misclassification, and the approach has not been validated against a more comprehensive COD determination based on, for example, a panel review of all relevant information. Also, we included all recurrence types, including local (eg, breast) and regional (eg, axilla) sites, which alone may not increase the risk of death. However, among patients with local/regional failure from these trials, there was a substantially increased risk of distant recurrence (and, consequently, breast cancer death).22, 23 In addition, to the extent that the reported COD is accurate, we note that, among patients who reportedly died of breast cancer according to their death certificate ICD code, >90% had prior breast cancer recurrence events. In general, recurrence has not proven to be an acceptable surrogate for survival in early stage breast cancer, although it is an important and widely used endpoint.24 This is unlike the case of colon cancer, in which disease-free survival (time without recurrence, second primary cancer, or death) recently was advocated as a primary endpoint in clinical trials.25, 26 In the current study, we used recurrence not as a surrogate endpoint in the usual sense but, rather, to conclude whether the eventual death was because of breast cancer. Incorporating anatomic site of recurrence (ie, distant metastatic vs local recurrence), time of recurrence in relation to death, and information on second primary cancer incidence after recurrence could improve COD classification.

A broader concern is ambiguity in the meaning of “cause of death” determined by any method. In therapy development, all-cause survival has been appropriately adopted as an endpoint, because it implicitly accounts for both beneficial and adverse treatment effects; and, indeed, many cancer treatments are associated with high risk for morbidity and mortality. Assigning COD solely according to evidence of disease progression ignores increased risk of other chronic diseases that may be attributable to cancer treatment. Thus, although identifying cancer-specific survival is useful for biologic considerations and for studying trends in populations, all-cause survival will remain primary as an endpoint in developing cancer treatment.

With respect to limitations in extending findings on performance of relative survival from the trials to the population-based (SEER) data, SEER patients tended to be older and to have more favorable disease characteristics. The clinical trial cohort was more favorable than the SEER cohort with respect to other-cause life expectancy. The frequency of missing COD differed markedly between the data sources, and it is unknown how accuracy of deaths certificate COD compares between clinical trials and SEER. It is noteworthy that the relative survival and death certificate-based estimates were remarkably similar between SEER and trial participants, particularly among lymph node-positive patients (Fig. 1).

In summary, our current evaluation of the performance of relative survival confirmed the concern that, when the cancer patient cohort has more favorable noncancer survival than the population at large, the method overestimates cancer-specific survival. In these instances, and when information on intervening clinical events is not available (eg, in SEER), the estimation of cancer-specific survival using COD information and treating deaths from unknown causes as caused by cancer provides a reasonable alternative, provided that missing COD is infrequent and effectively random. When the cohort has population-typical other-cause mortality or when the dominant mortality cause is expected to be cancer, as we observed among lymph node-positive patients and, specifically, in younger patients and those with larger tumors, then relative survival accurately estimates breast cancer-specific survival. In clinical trials, although overall survival will remain an endpoint of primary interest, cancer-specific survival derived from clinical event histories is an attractive option that requires further validation. Relative survival appears to be a reasonable alternative in clinical trials when the aforementioned conditions hold. With progress toward targeted therapies and more complex multiagent regimens, accurate estimation of disease-specific survival will be critical to assessing both the benefits and the risks of new treatments.

Conflict of Interest Disclosures

Supported by contract 263-MQ-611334 to J.J.D. from the Division of Cancer Control and Population Sciences/Surveillance Research Program/Statistical Research and Applications Branch of the US. National Cancer Institute and by a research grant from the Susan G. Komen for the Cure Foundation. The clinical trials were supported by Public Health Service grants NCI-U10-CA-69,651 and NCI-U10-CA-12,027 from the US. National Cancer Institute.