Presented in part at the 40th Annual Meeting of the American Society for Clinical Oncology, New Orleans, Louisiana, June 5–8, 2004.
Temporal comparisons of case survival are commonly used to assess improvement in cancer treatment at the population level. However, such comparisons may be confounded by secular trends in disease prognosis, even within conventional stage categories. The objective of the current study was to characterize within-stage migration of tumor size in breast carcinoma, and to estimate the effect of this shift on reported breast carcinoma survival.
Population-based Surveillance, Epidemiology, and End Results (SEER) cancer registry data were used to evaluate secular trends in tumor size at the time of diagnosis and relative survival among localized and regional invasive breast carcinoma patients diagnosed between 1975–1999. Outcomes were stage-specific tumor size distribution, 5-year relative survival, relative survival standardized to the tumor size distribution of the cohort diagnosed between 1975–1979, and the percentage of improvement in relative survival attributable to shifts in tumor size distribution.
Within each stage category, the proportion of smaller tumors increased significantly over time. Comparing patients diagnosed between 1995–1999 with those diagnosed between 1975–1979, within-stage migration of tumor size accounted for 61% and 28%, respectively, of the relative survival increases noted in localized and regional breast carcinoma.
Breast carcinoma mortality has declined substantially over the past 30 years, from 31.4 deaths per 100,000 women per year in 1975 to 25.9 deaths per 100,000 women per year in 2001.1 Because the mortality rate depends on both the incidence of newly diagnosed tumors and the death rate of women diagnosed with the disease, and the incidence of breast carcinoma has increased considerably since 1975,1 the decline in breast carcinoma mortality necessarily implies an improved survival rate among diagnosed patients. Indeed, population-based cancer registry data document improvements in breast carcinoma survival over time, both across and within disease stage categories.1, 2
The secular trend toward better breast carcinoma survival in the U.S. has coincided with important advances in both screening and treatment. The first national guidelines recommending population-wide mammography screening were issued in the late 1970s,3 and the use of screening increased dramatically in subsequent decades.4 During this same time period, the effectiveness of cytotoxic and hormonal systemic agents was established in both the adjuvant and metastatic settings,5–7 and their use has increased considerably.8
Ultimately the goal of both screening and treatment is the same—to reduce mortality from the disease—but distinguishing the relative contributions of these two modalities is difficult. The impact of screening can be assessed only on the basis of mortality rates in the screened population.9 Case survival cannot be used for this purpose because the date of diagnosis, the time origin for calculating survival, is advanced by an unknown time interval due to a screening test even if the test has no beneficial impact on mortality. Independent of the true benefits of screening, screen-detected tumors will always have longer survival times than symptom-detected tumors because the diagnosis is made earlier in the natural history of the disease, a phenomenon known as lead time bias.9–12
Although the impact of screening must be evaluated using mortality rates in the population at risk for breast carcinoma, the impact of treatment is assessed more directly by studying changes in case survival because cases represent the population to which treatment is applied. For this purpose, changes in survival, or relative survival, have been used to monitor secular trends in the U.S. and elsewhere.2, 13–17 However, temporal comparisons of this nature rely on the assumption that the inherent composition of diagnosed cases with respect to prognosis does not change over time, or that such changes can be suitably controlled by stratification on the basis of stage of disease.
Even within conventional stage categories, the distribution of important prognostic characteristics may change over time. Although a number of factors affect disease outcome, tumor size is an obvious marker of natural history and a strong predictor of breast carcinoma survival.18–20 Tumor size also is a straightforward, reliably evaluated, consistently available metric. The Surveillance, Epidemiology, and End Results (SEER) Program, which tracks trends in cancer incidence and survival in the U.S., has collected detailed information regarding the size of newly diagnosed breast tumors since 1975.21
The objective of the current study was to assess the impact of within-stage migration of tumor size on changes over time in breast carcinoma survival. We characterized the tumor size distribution of localized and regional breast carcinomas diagnosed between 1975–1999 to document the extent of within-stage migration of tumor size. We then used this information to determine the percentage of improvement in relative survival that could be attributable to shifts in the tumor size distribution. Our goal was to provide a more accurate estimate of the impact of the evolution of breast carcinoma treatment on survival.
MATERIALS AND METHODS
The current study sample was comprised of nonmetastatic, invasive breast carcinoma cases reported to the SEER cancer registries between 1975–1999. We used data from the nine registries that have been included in SEER since its inception in 1973.21 Together, these registries cover geographic regions representing approximately 10% of the U.S. population. The registries collect clinical and sociodemographic data concerning all incident cancers in their coverage area, with active follow-up for date and cause of death. The SEER registries maintain high standards of data quality, and the program's overall completeness of case ascertainment is reported to be 98%.22
We included first primary invasive breast tumors designated as localized or regional. Stage at diagnosis was based on the SEER historic staging system in which localized tumors are those tumors entirely confined to the breast, and regional tumors are those tumors with extension beyond the breast into surrounding tissues or axillary lymph nodes.23 Cases for which stage information was missing (< 3%) were excluded. We also excluded localized tumors with axillary lymph node involvement recorded as positive (< 1% of cases), assuming that either their stage or lymph node status was misclassified. Although the staging system developed by the American Joint Committee on Cancer (AJCC) stratifies tumors more precisely according to specific categories of tumor size, lymph node involvement, and distant metastases, AJCC stage was not recorded by SEER until 1988, and data limitations for tumors diagnosed between 1975–1987 prevented us from assigning AJCC stage designations to these cases.
Because our analysis focused on shifts in the tumor size distribution within stage categories, we evaluated localized and regional tumors separately. These two stages constitute approximately 92% of the invasive breast carcinomas with known stage in SEER. We did not study women who presented with distant metastases at the time of diagnosis because the extent of local disease is largely unrelated to prognosis in those cases. Tumors diagnosed only at the time of death also were excluded.
In the SEER public-use files, detailed extent-of-disease information (i.e., tumor size, tumor extension, and lymph node involvement) is available only for cases diagnosed since 1988. We obtained additional SEER data from the National Cancer Institute and reviewed the relevant coding manuals to identify tumor size for cases diagnosed prior to 1988. Specific variables used in the analysis are listed in Table 1.
Table 1. SEER Extent-of-Disease Variables
Year of diagnosis
EOD: extent of disease.
Expanded EOD-CP53 Expanded EOD-CP54
Expanded EOD-CP59 Expanded EOD-CP60
Tumor Size Classification
Since 1975, three different coding schemes have been used in SEER to classify tumor size. In all years, size was recorded as the maximum dimension of the primary tumor. Between 1975–1982, two measurements of tumor size were recorded: the size given in the pathology or surgical report and the size noted at the time of clinical examination. For both measurements, SEER used 11 different size categories. For the current analysis, we used the size recorded in SEER from the pathology or surgical report if available, and the size noted on the clinical examination if not. Since 1983, only one tumor size measurement has been recorded, but the SEER registrars prioritize the size listed in the pathology report, the surgical report, and the physical examination in that order. Size is recorded in millimeters, with tumors measuring less than 3 mm combined into 1 category. Between 1983–1987, tumors measuring 10.0 cm and larger were combined. Since 1988, exact tumor size has been specified up to 99.0 cm. In all years, diffuse or inflammatory disease is identified as a distinct category.
We integrated the three coding schemes to group tumor size into clinically meaningful intervals while maintaining a consistent definition across time periods. The resulting tumor size categories for the current analysis were: < 1.0 cm, 1.0–1.9 cm, 2.0–2.9 cm, 3.0–3.9 cm, 4.0–4.9 cm, ≥ 5.0 cm, and size not stated. Tumors with microscopic focus or foci and those described as mammographically detected, nonpalpable tumors were classified as measuring < 1.0 cm. Tumors described as diffuse or inflammatory were grouped with those tumors measuring ≥ 5.0 cm. Cases of Paget disease of the nipple and carcinomas in which the tumor size was reported as “no tumor” were excluded from the analysis; these tumors constituted fewer than 1% of all cases in each time period.
Year of diagnosis was grouped into 5-year intervals: 1975–1979, 1980–1984, 1985–1989, 1990–1994, and 1995–1999. Follow-up for vital status was available through 2001, the study cutoff point for survival analysis. For the cohort of cases in each time period, crude 5-year relative survival was calculated as the ratio between the overall survival probability observed in the breast carcinoma cohort and the survival probability expected in the absence of breast carcinoma.24 Relative survival probabilities and standard errors were estimated using SEER*Stat (SEER*Stat, version 5.2.2; Information Management Services Inc., Silver Spring, MD), a software package developed specifically for the analysis of SEER data. SEER*Stat generates expected survival probabilities from U.S. population statistics, matched to SEER cases by race (white, black, American Indian, Chinese, Japanese, Filipino, Hawaiian, Puerto Rican, or Hispanic), gender, age (by year), and the date at which age was coded.25 Cases with race described as “other” or “unknown” were excluded from relative survival calculations.
Tumor Size–Standardized Relative Survival
To standardize relative survival to the tumor size distribution of the 1975–1979 cohort, we first calculated tumor size-specific relative survival probabilities within each cohort. The tumor size-standardized relative survival for each cohort was then estimated by calculating a weighted average of these relative survival probabilities across tumor sizes, in which the weights for each tumor size category were based on the tumor size distribution of the 1975–1979 cohort. For each of the later cohorts, this yielded a size-standardized relative survival estimate, with the 1975–1979 cohort serving as the standard population. This method of direct standardization is commonly used in epidemiologic research to age-adjust rates of disease incidence and mortality.26
Survival analysis and tumor size standardization were performed separately for localized and regional breast carcinoma. We also repeated the analysis, within each disease stage, stratified by age group (ages 25–49 yrs, ages 50–64 yrs, and age ≥ 65 yrs).
Changes in the within-stage tumor size distributions were assessed using the Cochran–Armitage test for trend.27, 28 Time trends were considered significant if the test statistic had a two-sided P value of less than 0.05.
To estimate increases in relative survival over time, we compared each cohort with the 1975–1979 cohort. We calculated the change in crude relative survival as well as the change in size-standardized relative survival. The percent of change in relative survival explained by a shift in the tumor size distribution is shown by the following formula:
in which Ri is crude relative survival in the ith cohort, RSt is size-standardized relative survival, and R75 is relative survival in the 1975–1979 cohort.
Other than the estimation of relative survival, calculations and statistical analyses were performed using Excel (Microsoft Corporation, Redmond, WA) and SAS (SAS version 9.0; SAS Institute Inc., Cary, NC) software.
The sizes of incident invasive breast tumors in SEER have decreased substantially since 1975. The magnitude of this change is shown in Figure 1 for localized breast carcinoma and Figure 2 for regional breast carcinoma. Among localized breast carcinomas (n = 166,317), fewer than 10% of tumors diagnosed between 1975–1979 measured < 1cm, whereas 25% of tumors diagnosed between 1995–1999 measured < 1cm. Among regional breast carcinomas (n = 99,522), the proportion of tumors measuring < 2 cm increased from approximately 20% in the 1975–1979 cohort to 33% in the 1995–1999 cohort. In both localized and regional breast carcinoma cases, the secular trend toward an increasing proportion of smaller tumors was found to be statistically significant (P < 0.0001).
Crude Relative Survival
In localized breast carcinoma, crude relative survival increased from 90.7% in the 1975–1979 cohort to 97.4% in the 1995–1999 cohort (Table 2). Despite this substantial increase, tumor size-specific 5-year relative survival changed by much smaller amounts during the 25-year period. The greatest absolute change occurred among tumors measuring 1.0–1.9 cm, with relative survival increasing from 94.5% in the 1975–1979 cohort to 99.9% in the 1995–1999 cohort. For other size categories, the changes were small to nonexistent. Absolute survival improvements were somewhat greater among regional breast carcinomas (Table 3), with crude relative survival increasing from 67.8% to 79.6% over the 25-year period. The greatest absolute relative survival increases were noted in tumors measuring < 1 cm and tumors measuring ≥ 5 cm.
Table 2. 5 Year Relative Survival (%) in Patients with Localized Breast Carcinoma
The absolute change reflects the comparison between the cohort diagnosed between 1975–1979 and the cohort diagnosed between 1995–1999.
Crude RS (SE)
Tumor size-specific RS
Size not stated
Size standardized RS
Size-Standardized Relative Survival
Both crude 5-year relative survival and tumor size-standardized relative survival increased between 1975 and 1999. However, as shown in Tables 2 and 3, standardization on the basis of tumor size attenuated the magnitude of this increase. Comparing the earliest (1975–1979) and most recent (1995–1999) cohorts, the size-standardized relative survival for localized breast tumors increased from 90.7% to only 93.3%, versus 97.4% without standardization; therefore size standardization accounted for 61% of the apparent improvement in 5-year relative survival among localized breast tumors. Size-standardization explained 28% of the apparent improvement in relative survival observed among regional breast tumors.
Stratification by Age
Among both localized and regional breast carcinomas, the proportion of survival improvement that was attributable to within-stage migration of tumor size was found to be greater for older patients (Table 4). In localized breast carcinoma cases, tumor size standardization explained 38%, 52%, and 96% of the observed improvement in relative survival among women ages 25–49 years, 50–64 years, and 65 years and older, respectively. In regional breast carcinoma cases, tumor size standardization explained twice the proportion of survival benefit in women age 65 and older (44%) compared with women ages 25–49 years (23%) and 50–64 years (22%).
Table 4. Crude and Size-Standardized 5-Year Relative Survival (%) by Patient Age at Diagnosis
Localized breast carcinoma
Regional breast carcinoma
RS: relative survival.
% explained by size shift
In this analysis, we used a population-based data source to describe changes over time in the distribution of tumor size among incident breast carcinomas. The literature regarding screening mammography has demonstrated that screening causes a shift toward earlier stage disease, primarily through the detection of smaller tumors.29–34 The results of the current study characterize the extent to which the tumor size distribution has shifted, even within stage categories. These results also demonstrate that failure to adequately control for this shift leads to inflated estimates of the impact of secular changes in treatment on stage-specific survival. A substantial proportion of the observed improvement in stage-specific breast carcinoma survival over the past 30 years can be attributed to this within-stage migration of tumor size. Standardization of survival rates in contemporary cohorts to the tumor size distribution of women diagnosed between 1975–1979 was found to explain 61% of the observed survival improvement in localized breast carcinoma and 28% of the improvement in regional breast carcinoma over the past 25 years. We believe that size standardization is a refinement of relative survival that greatly improves our ability to interpret secular trends in breast carcinoma survival.
We used relative survival as an indicator of disease outcome because it is a preferred measure of cancer survival.35 However, changes in relative survival over time may be influenced by secular changes in important causes of death other than cancer. This is controlled to some degree by the use of expected survival rates that correspond with each cancer patient's date of diagnosis. Relative survival estimates also may be problematic if expected survival in the cohort of interest is not representative of the general population, even after matching based on age, gender, race, and date. This could be the case if women with screen-diagnosed early-stage breast carcinoma are healthier on average than their unscreened peers. Therefore, changes in relative survival, even after tumor size adjustment, cannot be considered a definitive measure of the impact of treatment.
In addition, adjustment of relative survival estimates using the traditional method of direct standardization has been criticized as conceptually inconsistent because crude and adjusted relative survival probabilities may not be equal in the standard population.36, 37 In patients with localized and regional breast carcinoma who were diagnosed between 1975–1979 (the standard population in the current analysis) crude and tumor size-adjusted relative survival were nearly identical (data not shown), suggesting that the potential limitations of direct standardization do not substantially threaten the validity of the current findings. Moreover, we repeated all analyses using disease-specific survival as the primary outcome, and the results were similar to those based on relative survival. However, disease-specific survival is also not a definitive metric because cause-of-death attribution is subject to a variety of biases and errors.38–41
Another trend that could potentially confound the results of the current study is an increase in the intensity of diagnostic evaluation.42 For example, if the rate of detection of lymph node metastases increased over time (either because of advances in technology or changes in clinical practice), then some cases of truly regional disease may have been mistakenly classified as localized in the earlier time periods. Relative survival would appear to increase over time as a result of this misclassification. We found that among both localized and regional breast carcinoma patients, the percentage of women who had any lymph nodes examined decreased slightly over time. Among those who had any lymph nodes examined, the average number of lymph nodes examined either decreased or remained stable. Because sentinel lymph node biopsy was not standard practice in the community by 1999,43 it is an unlikely explanation for the decrease in lymph node sampling. Despite this slight decrease in axillary lymph node evaluation, we found that the proportion of regional tumors with lymph node involvement actually increased over time, from 84% in the 1975–1979 cohort to 92% in the 1995–1999 cohort. This suggests that the sensitivity of lymph node evaluation improved during this time period, and it is possible that this trend could have an impact on size-specific survival. However, this concern applies to all analyses of survival trends, not just the innovation that we are proposing.
A final caveat concerns the possible impact of length-biased sampling on the current results. Breast carcinoma cases that are screen-detected tend to involve slower-growing tumors, and therefore have longer survival, than cases that are not screen-detected.12 For population-based analyses, length-biased sampling should have no effect on overall temporal comparisons of survival. However, because mammography rates have changed over time, length-biased sampling may affect the proportional allocation of cases to size categories, and therefore could have an unforeseen impact on size-standardization, although we suspect that this impact, if it exists at all, is likely to be minimal.
Although we cannot isolate the effect of treatment advances on cancer survival, the results of the current analysis reveal differences in the magnitude of the tumor size-standardized increases in relative survival across age groups. Comparing the earliest and latest cohorts, the absolute difference in tumor size-standardized relative survival was found to be much greater in women younger than 65 years than in women age 65 years and older, in both stage categories. This suggests patient subgroups in which treatment has exerted its greatest impact.
The results of the current study also reveal differences in size-standardized relative improvement across stage categories. Comparing the earliest and latest cohorts, the absolute difference in tumor size-standardized relative survival was approximately 4 times as great in patients with regional breast carcinoma as in those with localized breast carcinoma (8.5% vs. 2.7%). However, the proportional reduction in the percentage of breast carcinoma patients who did not survive for 5 years, absent other causes of death, (i.e., 1 minus size-standardized relative survival) was similar for localized and regional breast carcinoma patients.
Understanding the impact of different cancer control strategies on disease outcomes in the population is a high priority for health policy decision makers. To this end, the National Cancer Institute established the Cancer Intervention and Surveillance Modeling Network (CISNET), a consortium of investigators developing complex, sophisticated computer models to simulate the effects of screening and treatment in the U.S. population, and thereby inform cancer control efforts and facilitate health care planning and resource allocation.44 The current analysis shows that when evaluating temporal trends in case survival, stratification by tumor size can have a significant bearing on the interpretation of these trends. By comparing size-standardized survival statistics across cohorts, this method also provides a context for a more nuanced understanding of the reasons for age disparities in breast carcinoma outcomes.
The authors gratefully acknowledge Lynn Ries, Kathleen Cronin, Eric Feuer, and the Surveillance, Epidemiology, and End Results (SEER) Program for providing extent-of-disease data for breast carcinoma cases in SEER diagnosed prior to 1988, for supplying the historic coding manuals necessary to interpret these data, and for offering helpful comments on a draft of this article.