Survival of Lymph Node-Negative Breast Cancer Patients in Relation to Number of Lymph Nodes Examined

Abstract

Objective

To assess the relationship between number of lymph nodes examined and survival of patients diagnosed with node-negative localized breast cancer using a large sample of patients from population-based cancer registries in the United States.

Summary Background Data

Conflicting results have been reported from studies on the relationship between number of lymph nodes examined and survival of patients diagnosed with node-negative localized breast cancer.

Methods

The study included 69,543 patients diagnosed in 1988–97 with localized invasive node-negative breast cancer reported to nine population-based registries in the U.S. National Cancer Institute’s Surveillance, Epidemiology and End Results (SEER) Program of population-based cancer registries. Hazard ratios for groups defined by number of nodes examined were analyzed in Cox proportional hazards regression models that included age, tumor size and grade, race/ethnicity, and other variables.

Results

A significantly higher risk of death from breast cancer was found among patients with 0, 1 to 3, or 4 to 10 nodes examined than with 20-plus nodes examined, even among patients with tumors 2 cm or smaller.

Conclusions

Future studies of survival of node-negative patients, by number of nodes examined, should include information on comorbidity and treatment.

Conflicting results have been reported from studies ^1–3 examining the relationship between number of lymph nodes examined and survival of patients diagnosed with “node-negative” breast cancer localized to the breast. Sosa et al. ¹ studied 464 node-negative stage I surgical patients at the Johns Hopkins Hospital with invasive breast cancer 2 cm or smaller in diameter and found better survival among those with 10 or more (up to 33) versus less than 10 (1–9) nodes removed; numbers of deaths were too small for separate analysis of mortality from breast cancer versus other causes. Camp et al., ² however, reported that among 290 women who underwent surgical resection with axillary lymph node dissection (1 or more nodes) and had node-negative invasive ductal breast cancer at Yale-New Haven Hospital (1983–93), survival was poorer among those with 20 or more nodes than with 1 to 9 nodes examined, even after adjustment for prognostic factors. However, the small sample and small number of deaths from breast cancer (i.e., 28) suggested a need for additional studies. ³

Moorman et al. ³ reported no relationship between number of lymph nodes examined (1 to 19 vs. 20 or more, with maximum of 43) and risk of dying from breast cancer among 911 patients diagnosed in 1985–93 with localized invasive node-negative breast cancer (5 cm or smaller) at Duke University Medical Center. The hazard ratio for 1 to 19 versus 20-plus nodes examined was 0.99, but 95% confidence limits were wide (0.58–1.64). ³ The present study used larger samples of patients from a group of high-quality population-based U.S. cancer registries.

METHODS

The nine original registries in the National Cancer Institute’s Surveillance, Epidemiology and End Results (SEER) Program cover about 10% of the U.S. population, and data are believed to be generally representative of the entire United States (in the absence of a national system of cancer registration). ⁴ A SEER public-use data file ⁵ was used to identify all patients diagnosed in 1988–97 with breast cancer while living in a SEER area (i.e., San Francisco-Oakland, Connecticut, Detroit, Atlanta, Seattle, Hawaii, Iowa, New Mexico or Utah). Cancers diagnosed in 1998 were available but were excluded because of limited potential follow-up (i.e., the cutoff date for follow-up was Dec. 31, 1998). ⁵

Tumor size has been coded since 1988, and 70,720 women were diagnosed in 1988–97 with invasive cancer 5.0 cm or smaller of localized stage (i.e., confined to the breast) as their only or first reportable cancer, after excluding cancers ascertained from death certificate only or autopsy. After excluding 307 patients with no (or unknown) cancer-directed surgery, and 870 with unknown number of nodes examined, the sample size was 69,543.

Categories of number lymph nodes examined included 1 to 3 nodes and 20-plus nodes (a category used in previous studies). ³ Other categories (4–10, 11–14, and 15–19) were selected to provide nearly equal sample sizes. In addition, removal of four or more nodes is the SEER definition of lymph node dissection. ⁶

This study, in contrast to previous reports, ^1–3 also included patients with no nodes examined (i.e., clinical staging only) because of probable high rates of misclassification of nodal status, ⁷ which could affect survival.

Analyses included tumor grade, an independent predictor of survival, ⁸ which is coded in SEER as 1 (well differentiated), 2 (moderately differentiated), 3 (poorly differentiated), or 4 (anaplastic or undifferentiated). Tumor size was recoded as 1 cm or less (AJCC stage T1a and b), 1.1 to 2.0 cm (stage T1c), and 2.1 to 5.0 cm (stage T2).

Codes for both the “race” and Hispanic ethnicity items were used to define non-Hispanic white, black, Hispanic white, and other/unknown categories. ⁶ Only county of residence was included on the data file ⁵ for use in creating ecologic or surrogate indicators of socioeconomic status (SES) by linking with data from the 1990 Census. Poverty rate of all persons in the county was selected because, unlike median household income, household size is taken into account.

Cox proportional hazards regression ⁹ was used to examine the independent association between number of lymph nodes examined and risk of death from breast cancer (underlying cause, ICD-9 code 174) or from cancer of unspecified site (ICD-9 codes 199.0, 199.1), which could include breast cancer. Risk of death from other (underlying) causes of death also was examined. Models included tumor size and grade, age (<45, 45–54, 55–64, 65–74, and 75-plus years), race/ethnicity (as defined above), marital status (not married, married. or unknown), and poverty rate of county (<8.0%, 8.0–14.4%, and 14.5%-plus). Separate models were obtained for patients with tumors 2 cm or smaller. ¹

RESULTS

Table 1 shows that higher proportions of patients with none or few (vs. 20-plus) nodes examined were diagnosed in 1996–97 (vs. 1988–95). The proportions of patients with zero or one to three nodes examined were low in 1988–89 (1.6% and 6.0%) but had increased by 1996–97 (2.4% and 12.4%, respectively). The proportion with 4 to 10 nodes examined increased from 19.6% in 1988–89 to 22.8% in 1996–97 (data not shown). Noteworthy was the increase in the proportion of patients diagnosed at age 75 years or older with decreasing number of nodes examined, reaching 46% for the group with no nodes examined (Table 1). The proportion of patients known dead, especially from an underlying cause other than breast cancer, also showed a similar trend. Life-table analyses (data not shown) indicated differences in survival curves among the groups. However, the differences in distribution of age and other characteristics among the nodal groups must be taken into account in analyses of risk of death.

Table 1. DEMOGRAPHIC DATA BY NUMBER OF LYMPH NODES EXAMINED

* Only selected subgroups are shown. Chi-square tests compared not married vs. all others, non-Hispanic white vs. all other race/ethnicity groups, tumor size 2.1–5.0 vs. <2.1 cm, tumor grade 3 or 4 vs. all other, poverty rate of county 14.5%+ vs. <14.5%, and dead (vs. alive).

In the multivariate Cox proportional hazards models with all patients (i.e., all tumor sizes up to 5.0 cm), adjusted hazard ratios (HRs) for risk of death from breast cancer were statistically significant for tumor grade and size (Table 2). They also were elevated for younger age at diagnosis and for black versus non-Hispanic white patients but were reduced for married versus unmarried patients. The adjusted HRs for high and unknown, versus low, tumor grade were significantly elevated for death from breast cancer but not for other causes (HR 1.06, P = .097, for grade 3, 4; HR 1.02, P = .421, for unknown grade; data not shown). Ungraded cases are probably disproportionately high grade in view of lower survival rates compared to graded cases. ⁶ Risk of death from causes other than breast cancer increased sharply with age (data not shown), in contrast to the findings in Table 2.

Table 2. COX PROPORTIONAL HAZARDS MODELS FOR RISK OF DEATH FROM BREAST CANCER

All models included age at diagnosis, marital status, race/ethnicity, and poverty rate of county of residence (data not shown for model 2).

*P < .05; CI, confidence interval.

With regard to number of nodes examined, in the first Cox proportional hazards regression model, the adjusted HRs were significantly elevated for 0, 1 to 3, and 4 to 10 nodes examined, using 20-plus nodes examined as the reference category. The group 4 to 10 (but not 1 to 3) nodes examined would be defined as having lymph node dissection according to the SEER rules. The adjusted HR was highest for no nodes examined. Adjusted HRs for number of nodes examined for risk of death from causes other than breast cancer were similar to those shown in Table 2: HRs were significantly elevated for 0 nodes (HR 2.30, P < .001), 1 to 3 nodes, 4 to 10 nodes, and 11 to 14 nodes, but not for 15 to 19 nodes (vs. the reference category of 20-plus nodes) (data not shown). The association between number of nodes examined and risk of death from breast cancer was not significant in a model (not shown) with the same variables as shown in Table 2 but comparing 20-plus nodes to 1 to 19 nodes (reference category); the HR was 0.93 (P = .106). This comparison of patients with “node dissection” has been made in previous studies, with conflicting results. ^2,3

Within the subgroup with tumors 2 cm or smaller, the adjusted HRs for death from breast cancer were statistically significant for 4 to 10 and 0 nodes and approached statistical significance for 1 to 3 nodes versus 20-plus nodes examined (model 2, Table 2). As in model 1, tumor grade was associated with risk of death from breast cancer but not from other causes (data not shown).

DISCUSSION

Study limitations include some misclassification, as indicated by SEER reabstracting and recoding audits of stage and extent of disease, number of nodes examined, and tumor size. ¹⁰ Also, identifying deaths due to breast cancer (vs. other causes) is problematic, ¹¹ although the lack of association between tumor grade and risk of death from causes other than breast cancer could suggest limited misclassification.

While this study focused on number of lymph nodes examined, other associations with risk of death from breast cancer were consistent with those reported in the literature:^12,13 diagnosis at a young age, black versus non-Hispanic white race/ethnicity, and unmarried versus married marital status were associated with increased risk of death from breast cancer (see Table 2). The associations with marital status and race/ethnicity may be due to residual confounding with SES or to other (e.g., psychosocial) factors. ¹³ Adjusting for stage and SES in a managed care population (i.e., with presumably uniform medical care access), however, eliminated the black versus white difference in survival in Detroit. ¹⁴ The weak association with the (county-level) SES indicator (see Table 2) could reflect the crudeness of the indicator used and also the inclusion of other variables in the model (e.g., tumor size) that are more directly associated with SES.

A recent study reporting no association with risk of death focused on 20-plus versus 1 to 19 nodes examined, ² and a comparison of 10-plus to 1 to 19 nodes in the present study also showed no association with risk of death from breast cancer. However, the large samples in the present study allowed analyses of several subgroups defined by number of nodes examined (i.e., 1–3, 4–10, 11–14, and 15–19 nodes). A slight but significant elevated risk of death from breast cancer was found for patients with 4 to 10 versus 20-plus nodes examined, and there was also a higher (also significant) risk ratio for 1 to 3 versus 20-plus nodes (see Table 2). One previous study did report an association for 10-plus versus 1 to 9 nodes among patients with cancers 2 cm or less. ¹

In a 5-year follow-up study of clinically node-negative patients treated without axillary surgery, ¹⁵ the risk of nodal relapse was 9% for cancers 2 cm or smaller versus 34% for cancers larger than 2 cm. Despite the expectation that avoidance of axillary dissection would have a negligible effect on outcome for patients with small tumors, ¹⁵ the association between number of nodes examined and risk of death from breast cancer was similar among patients with tumors 2 cm or smaller and among all patients (i.e., with tumors 5 cm or smaller) (see Table 2).

As noted by Sosa et al., ¹ an association between more extensive versus less extensive axillary dissection and reduced risk of death from breast cancer could be explained by either a direct therapeutic effect or understaging of patients with few nodes removed (i.e., misclassification of node-positive patients as node-negative). Under the latter hypothesis, removal of more nodes permits better classification of nodal status, which is a marker for (but does not directly influence) prognosis. ¹⁶

The highest risk of death from breast cancer in this study was for patients with no nodes examined, a group not included in previous studies. This finding could be consistent with the unreliability of clinical assessment of the axilla. ⁷ However, the finding of an association between number of nodes examined (vs. 20-plus nodes, as the reference category) and risk of death (i.e., underlying cause on death certificates) from causes other than breast cancer suggests possible confounding with comorbidity. In other words, surgeons may not want to perform axillary dissection, or may perform less extensive dissection, on patients with comorbid conditions. ³ Another possible explanation is misclassification of cause of death on death records (i.e., underestimation of the role of breast cancer in affecting death). ¹¹

With regard to confounding with comorbidity, administrative (e.g., hospital discharge) databases are widely used to crudely assess comorbidity but may provide only a slight improvement over age adjustment. ¹⁷ Age was included in the analyses performed in this study (see Table 2). In view of the apparent association between number of nodes examined and risk of death, future studies should include careful review of medical records to ascertain comorbid conditions and determine cause of death.

Other (clinical) implications concern the use of adjuvant therapy. Even if removal of more nodes did not in itself improve survival, ¹⁶ which is a controversial issue, ¹ adjuvant therapy may be withheld from some node-negative patients, ¹⁸ including a subgroup misclassified as node-negative. A decision analysis has suggested some potential benefit of axillary dissection for low-risk cancers 1 cm or smaller, ¹⁹ where lymph node status may strongly influence the use of chemotherapy. ²⁰ Adjuvant therapy (including chemotherapy) is included for most or all patients in two clinical trials on sentinel node biopsy versus axillary dissection, ²¹ which may lead to new clinical guidelines. However, actual compliance with guidelines (including use of adjuvant therapy) may vary by hospital and geographic area. ¹⁸ Future studies should examine comorbidity and treatment in relation to survival among node-negative patients by number of nodes examined (including sentinel node alone).

About half of the patients with no nodes examined were age 75 years or older at diagnosis (see Table 1), and consensus guidelines for adjuvant therapy are uncertain for women over the age of about 70 years, due to limited evidence and concerns that “comorbid medical conditions and mortality from non-cancer causes will influence the overall benefits.”²⁰ However, a significantly elevated adjusted HR for risk of death from breast cancer was evident for patients with 4 to 10 (vs. 20-plus) nodes examined, as well as those with only 1 to 3 nodes examined. The latter group was uncommon (although the proportion increased over time), but the former group represented 16.7% of all patients (14,812/69,543; see Table 2), and only 21.0% were age 75 years or older (see Table 1).

Although the present study was limited to localized stage “node-negative” patients, the introduction of more sensitive methods of diagnosis or staging results in “stage migration” (toward earlier or more favorable stages), and the survival of patients in each stage ostensibly improves over time (the “Will Rogers phenomenon”). ²² This has been suggested for prostate cancer after the introduction of prostate-specific antigen testing. ²³ This phenomenon affects survival statistics in clinical research and in databases used by clinicians and other investigations. In the present study, however, the temporal trend was toward lesser (not greater) use of axillary dissection. Thus, over time, the stage distribution could shift or “migrate” (toward node-negative), while increases would occur in both the proportion of node-positive cases with extensive nodal involvement (easily detected clinically or by removal of only a few nodes) and the proportion of false-negatives among node-negative cases. The survival of both node-positive and node-negative patients would ostensibly decrease over time, but overall survival (of all breast cancer patients) would not decline unless removal of nodes had a direct therapeutic effect or changes occurred in the use of adjuvant therapies that improve survival. Examining this hypothesis would require further follow-up of patients diagnosed in recent years, and interpretation would be complicated by the increasing use of mammography (which detects smaller, often node-negative cancers).

Another trend may be toward the use of a combination of pathologic and biologic tumor characteristics as predictors of axillary node status, at least for small tumors and especially for cases in which adjuvant therapy decisions do not rest fully on confirmation of nodal status. ^24–26 However, axillary dissection may still be needed for some patients, at least until sentinel node mapping can be shown to be the standard of care. ²⁶

Acknowledgment

The author thanks two reviewers for comments on the first draft of this manuscript.