Open Access
Open Peer Review

This article has Open Peer Review reports available.

How does Open Peer Review work?

Modeling the effect of age in T1-2 breast cancer using the SEER database

  • Patricia Tai1Email author,
  • Gábor Cserni2,
  • Jan Van De Steene3,
  • Georges Vlastos4,
  • Mia Voordeckers3,
  • Melanie Royce5,
  • Sang-Joon Lee6,
  • Vincent Vinh-Hung3 and
  • Guy Storme3
BMC Cancer20055:130

DOI: 10.1186/1471-2407-5-130

Received: 25 February 2005

Accepted: 08 October 2005

Published: 08 October 2005

Abstract

Background

Modeling the relationship between age and mortality for breast cancer patients may have important prognostic and therapeutic implications.

Methods

Data from 9 registries of the Surveillance, Epidemiology, and End Results Program (SEER) of the United States were used. This study employed proportional hazards to model mortality in women with T1-2 breast cancers. The residuals of the model were used to examine the effect of age on mortality. This procedure was applied to node-negative (N0) and node-positive (N+) patients. All causes mortality and breast cancer specific mortality were evaluated.

Results

The relationship between age and mortality is biphasic. For both N0 and N+ patients among the T1-2 group, the analysis suggested two age components. One component is linear and corresponds to a natural increase of mortality with each year of age. The other component is quasi-quadratic and is centered around age 50. This component contributes to an increased risk of mortality as age increases beyond 50. It suggests a hormonally related process: the farther from menopause in either direction, the more prognosis is adversely influenced by the quasi-quadratic component. There is a complex relationship between hormone receptor status and other prognostic factors, like age.

Conclusion

The present analysis confirms the findings of many epidemiological and clinical trials that the relationship between age and mortality is biphasic. Compared with older patients, young women experience an abnormally high risk of death. Among elderly patients, the risk of death from breast cancer does not decrease with increasing age. These facts are important in the discussion of options for adjuvant treatment with breast cancer patients.

Background

In many clinical situations, age is an important determinant of treatment decision in breast cancer. For example, after mastectomy, patients with T2 tumors and one to three positive nodes are at high risk of isolated loco-regional recurrences. Authors have advocated the routine use of postmastectomy radiotherapy in those patients who have T2 tumors and who are younger than 45 years [1]. In another study about close margins at mastectomy, the subgroup of patients aged 50 or younger with clinical T1-2 tumors and 0–3 positive nodes who have close (5 mm or less) or positive margins were at high risk (28% at 8 years) for chest wall recurrence regardless of adjuvant systemic therapy. Therefore, such patients should be considered for postmastectomy radiation [2]. Young women aged less than 45 should be regarded as high-risk patients, on the basis of age alone, and should be given adjuvant cytotoxic treatment [3]. The latter study showed a non-linear relationship between age and relative risk of dying.

At the other end of the age spectrum, breast cancers in elderly patients have been considered by some authors to exhibit a less aggressive behavior than in younger patients [4, 5]. Other authors have argued that breast cancer does not become more indolent as age increases [6].

There are still controversial issues about the relationship between age and prognosis in breast cancer. Detailed analysis would be useful in order to provide more insight into this relationship. In the present study, we used proportional hazards to model the survival of T1-2, node-negative (N0) and node-positive (N+) breast cancer patients. Outcomes which we considered included all-cause mortality and cancer specific mortality from breast cancer. The primary aim of the study is to present how age relates with the risk of death. The secondary objective is to search for a simple algebraic representation of this relationship.

Methods

The Surveillance, Epidemiology, and End Results Program (SEER) of the United States collected data about the incidence of cancer and related matters from 11 population-based registries [7]. The data extracted in this study was from 9 registries: San Francisco-Oakland, Connecticut, Metropolitan Detroit, Hawaii, Iowa, New Mexico, Seattle (Puget Sound), Utah, and Metropolitan Atlanta. Selected patients were women who were without previous history of cancer and presented with non-inflammatory invasive breast carcinoma, diagnosed and histologically confirmed pT1-2 pM0 between 1988 and 1997, and for whom curative surgery and axillary lymph node dissections were performed. In 1987, the American Joint Committee on Cancer (AJCC) staging defined pT1 tumors as 2 cm or less in greatest dimension, and pT2 tumor as more than 2 cm but not more than 5 cm in greatest dimension. These definitions did not change until 1997. Some records were rejected because of concerns about the quality of data: non-hospital based data records, uncertain sequence of treatment, unknown month of diagnosis and unknown race. Records with missing histological grade and receptor status were not excluded. Examination of statistical outliers excluded one case with 75 nodes involved. Events for the study were death from all causes and death from breast cancer. Follow-up cutoff date was December 31, 1999 as provided by the database.

In order to verify the linearity of the continuous variables, the martingale residuals (differences between observed and expected numbers of events) were used. The martingale residuals were examined by a non-parametric smoothing (fitting the scatter-plots of residuals) against the quantitative covariates of interest. The smoothing used a Poisson regression implementation of generalized additive model (GAM) [8]. The GAM procedure provided two outputs. One was the non-parametric smoothed curves approximating the residuals. The other was a significance test of the non-linearity of the curves. For the covariates that significantly departed from linearity, an iterative search was performed to identify parametric families of functions that approximated the curves. The criteria used to end the search were: [a] simple parametric expression, [b] the corresponding function introduced as a transform in the Cox model satisfying the GAM linearity test, and [c] without deteriorating the model fit as assessed by the sum of squares of "deviance residuals" [8]. If the transforms were valid, the graphical displays should be linear shapes, and the non-linearity test results should be non-significant. Finally, scaled Schoenfeld residuals were used to verify that the relative hazards were constant over time [9]. The hypothesis underlying this dual modeling approach was as follows. If the algebraic functions are valid, their use as plug-in transforms should appropriately linearize the functional forms of the covariates of interest. Other information about the implementation of these procedures have been described earlier [1012].

The analysis was applied first to node-negative cases ("training set") in order to find a simple expression of the functional form which relates age to mortality. The functional form obtained from node-negative cases was then applied to node-positive cases ("validation set"). In addition to the validation with the same transformation which was obtained for node-negative patients, a further iterative search was performed in order to improve the fit for node-positive patients.

This analysis was applied also to a European dataset, the German Breast Cancer Study Group (GBSG-2), in which the outcome studied was disease-free survival [14]. From a data analysis perspective, this GBSG-2 dataset is a very different database of 686 patients containing some extreme observations. One case had 51 involved nodes (range for other patients 1–38), and another case had a tumor size of 120 mm (range for other patients 3–100). There were 299 events (either recurrence of disease or death) in this German database.

The statistical analyses were performed with Splus (Insightful Corporation, Seattle, WA, USA) statistical software. Parametric fitting of curves used TableCurve 2D (Systat Software Inc, Richmond, CA, USA).

Results

There are 83,804 T1-2 cases (58,139 N0 and 25,665 N+, mean: 4 nodes involved, range: 1–48) available for analysis from the SEER database. Table 1 shows the characteristics of the patients. This table has been presented elsewhere [13]. Except for 28 additional cases (because of updated registration), there are no noticeable differences in the distribution of the characteristics. Table 2 shows the results of proportional hazards models in N0 and N+ groups, without using transforms for covariates. The supplemental Table 2b (Additional file 1) shows results of the check for Cox proportional hazards for all covariates. Note that some P-values are very small because of the very large size of the data. The rho-values (slope) indicate very small departures from the assumption of proportional hazards.
Table 1

Characteristics of the patients.

Characteristics of patients

Total No.

% of all cases

Total N(0)

% of all cases

Total N(+)

% of all cases

Patient number

83804

100.0%

58139

69.4%

25665

30.6%

SEER area:

      

East (Connecticut, Detroit, Atlanta)

31795

37.9%

21683

25.9%

10112

12.1%

Central (Iowa, New Mexico, Utah)

20459

24.4%

14202

16.9%

6257

7.5%

West (San Francisco-Oakland, Hawaii, Seattle)

31550

37.6%

22254

26.6%

9296

11.1%

Year of diagnosis:

      

1988–89

15390

18.4%

10424

12.4%

4966

5.9%

1990–91

16660

19.9%

11589

13.8%

5071

6.1%

1992–93

17042

20.3%

11894

14.2%

5148

6.1%

1994–95

17072

20.4%

11897

14.2%

5175

6.2%

1996–97

17640

21.0%

12335

14.7%

5305

6.3%

Age at diagnosis (years):

      

20–39

5869

7.0%

3396

4.1%

2473

3.0%

40–49

16231

19.4%

10262

12.2%

5969

7.1%

50–59

18051

21.5%

12362

14.8%

5689

6.8%

60–69

20703

24.7%

15110

18.0%

5593

6.7%

70–79

17161

20.5%

12802

15.3%

4359

5.2%

80+

5789

6.9%

4207

5.0%

1582

1.9%

Race:

      

White and Other

78025

93.1%

54572

65.1%

23453

28.0%

Black

5779

6.9%

3567

4.3%

2212

2.6%

Marital status at diagnosis:

      

Single, widowed, other

33224

39.6%

23171

27.6%

10053

12.0%

Married

50580

60.4%

34968

41.7%

15612

18.6%

Topography:

      

Inner quadrant

12537

15.0%

9726

11.6%

2811

3.4%

Others

71267

85.0%

48413

57.8%

22854

27.3%

Histology:

      

Ductal

64370

76.8%

43924

52.4%

20446

24.4%

Others

19434

23.2%

14215

17.0%

5219

6.2%

Grade:

      

Poor/undifferentiated

25161

30.0%

15265

18.2%

9896

11.8%

Others

58643

70.0%

42874

51.2%

15769

18.8%

ER/PR status (+ includes unspecified):

      

ER+ and PR+

63306

75.5%

44332

52.9%

18974

22.6%

ER+ and PR-

7001

8.4%

4852

5.8%

2149

2.6%

ER- and PR+

2554

3.0%

1696

2.0%

858

1.0%

ER- and PR-

10943

13.1%

7259

8.7%

3684

4.4%

Tumor size (mm) (T stage):

      

< = 5 (T1a)

3349

4.0%

3069

3.7%

280

0.3%

> 5 and < = 10 (T1b)

16997

20.3%

14691

17.5%

2306

2.8%

> 10 and < = 20 (T1c)

36754

43.9%

26443

31.6%

10311

12.3%

> 20 and < = 50 (T2)

26704

31.9%

13936

16.6%

12768

15.2%

Number of nodes examined:

      

1–9

14225

17.0%

10525

12.6%

3700

4.4%

10–14

27004

32.2%

19210

22.9%

7794

9.3%

15–19

23102

27.6%

15820

18.9%

7282

8.7%

20+

19473

23.2%

12584

15.0%

6889

8.2%

Number of nodes involved:

      

0

58139

69.4%

58139

69.4%

0

0.0%

1–3

16778

20.0%

0

0.0%

16778

20.0%

4+

8887

10.6%

0

0.0%

8887

10.6%

Breast conserving surgery/Radiotherapy

      

no/no

46862

55.9%

31752

37.9%

15110

18.0%

no/yes

3972

4.7%

1024

1.2%

2948

3.5%

yes/no

4438

5.3%

2946

3.5%

1492

1.8%

yes/yes

28532

34.0%

22417

26.7%

6115

7.3%

ER = estrogen receptor, N(0) = node negative, N(+) = node positive, PR = progesterone receptor, SEER = Surveillance, Epidemiology, and End Results Program.

Table 2

Results of Cox proportional hazards models without transform. Hazard ratios (95% confidence intervals), values > 1 indicate increased risk of death.

 

Overall mortality

Breast cancer specific mortality

 

N0

N+

N0

N+

SEER central area

1.00 (0.95–1.05)

0.93 (0.87–0.98)

1.03 (0.94–1.13)

0.92 (0.86–0.99)

SEER western area

0.92 (0.88–0.97)

0.87 (0.82–0.92)

0.82 (0.75–0.90)

0.86 (0.80–0.91)

Race black

1.38 (1.27–1.49)

1.42 (1.31–1.52)

1.37 (1.21–1.55)

1.44 (1.31–1.57)

Married

0.75 (0.72–0.78)

0.82 (0.78–0.86)

0.88 (0.81–0.95)

0.91 (0.86–0.97)

Inner quadrant

1.09 (1.04–1.15)

1.16 (1.08–1.24)

1.32 (1.21–1.44)

1.3 (1.19–1.41)

Ductal histology

1.14 (1.09–1.2)

1.11 (1.04–1.17)

1.44 (1.31–1.58)

1.12 (1.05–1.21)

ER negative

1.39 (1.28–1.5)

1.52 (1.41–1.64)

1.58 (1.4–1.78)

1.58 (1.44–1.73)

PR negative

1.11 (1.03–1.18)

1.27 (1.18–1.36)

1.36 (1.21–1.52)

1.39 (1.28–1.52)

Grade 3–4

1.23 (1.17–1.29)

1.34 (1.28–1.41)

1.6 (1.48–1.73)

1.48 (1.39–1.57)

BCS

1.12 (1.02–1.24)

0.96 (0.86–1.07)

1.01 (0.84–1.20)

0.98 (0.86–1.12)

Radiotherapy

1.12 (0.96–1.29)

0.89 (0.83–0.95)

1.31 (1.08–1.6)

0.91 (0.84–0.99)

BCSxRT

0.61 (0.51–0.73)

0.92 (0.8–1.06)

0.63 (0.48–0.82)

0.88 (0.74–1.04)

Year Diagnosis (continuous, year)

0.98 (0.97–0.99)

0.96 (0.95–0.97)

0.92 (0.91–0.94)

0.94 (0.93–0.95)

Age at Diagnosis (continuous, year)

1.05 (1.05–1.05)

1.02 (1.02–1.03)

1.00 (1.00–1.01)

1.00 (1.00–1.01)

Tumor size (continuous, mm)

1.03 (1.03–1.03)

1.02 (1.02–1.02)

1.05 (1.04–1.05)

1.02 (1.02–1.03)

Number positive nodes (continuous, n)

(-)

1.07 (1.07–1.08)

(-)

1.09 (1.08–1.09)

Number nodes examined (continuous, n)

0.99 (0.99–0.99)

0.98 (0.97–0.98)

0.99 (0.99–1.00)

0.97 (0.96–0.97)

BCS = breast conserving surgery, BCSxRT = breast conserving surgery and radiotherapy, ER = estrogen receptor, N0 = node negative, N+ = node positive, PR = progesterone receptor, SEER = Surveillance, Epidemiology, and End Results Program.

Figures 1 and 2 show graphically the effect of age on the log hazard ratio for death from all causes, for N0 and N+ patients, respectively. Both curves have similar U-shapes. The mortality is lowest for patients about 50 years of age at diagnosis. The mortality increases the farther away from 50 years of age at diagnosis, for both younger and older patients.
https://static-content.springer.com/image/art%3A10.1186%2F1471-2407-5-130/MediaObjects/12885_2005_Article_318_Fig1_HTML.jpg
Figure 1

Mortality for all causes as a function of age for N0 patients. Dotted lines: twice-standard-error.

https://static-content.springer.com/image/art%3A10.1186%2F1471-2407-5-130/MediaObjects/12885_2005_Article_318_Fig2_HTML.jpg
Figure 2

Mortality for all causes as a function of age for N+ patients. Dotted lines: twice-standard-error.

The shape of the smoothed curve for age suggests the use of a quadratic function. A fractional polynomial analogous to Sauerbrei and Royston [14], but with different exponents, combining a linear term (age) and a quasi-quadratic term |age-50| 1.5, i.e. age+ |age-50| 1.5, provides a good fit and passes the test of linearity (Chi-square = 6.530, P = 0.089) in N0 patients (Table 3).
Table 3

Results of GAM and proportional hazards tests for SEER data

Overall mortality

Node-negative

Node-positive

 

no transform

transform age+|age-50|1.5

transform age+|age-50|1.8

no transform

transform age+|age-50|1.5

transform age+|age-50|1.8

GAM Chisq for Age (smaller value better)

582.4

6.53

52.4

439.2

18.4

1.83

GAM p-value for Age (larger better)

<.0001

0.089

<.0001

<.0001

0.0003

0.557

Test PH for Age: rho (smaller absolute value better)

0.069

0.071

0.075

0.048

0.044

0.045

Test PH for Age: chisq (smaller value better)

53.6

46.3

48.2

21.1

15.8

16.0

Full PH model Rsquare (larger value better)

0.088

0.094

0.093

0.126

0.135

0.135

Full PH model Likelihood ratio test (larger value better)

5364

5745

5656

3443

3713

3709

Breast cancer specific mortality

Node-negative

Node-positive

 

no transform

transform age+|age-50|1.5

transform age+|age-50|1.8

no transform

transform age+|age-50|1.5

transform age+|age-50|1.8

GAM Chisq for Age (smaller value better)

52.7

11.1

8.10

63.7

9.20

4.92

GAM p-value for Age (larger better)

<.0001

0.010

0.040

<.0001

0.027

0.179

Test PH for Age: rho (smaller absolute value better)

-0.027

-0.038

-0.042

-0.007

-0.023

-0.027

Test PH for Age: chisq (smaller value better)

2.5

4.5

5.5

0.3

2.8

3.6

Full PH model Rsquare (larger value better)

0.03

0.03

0.031

0.1

0.1

0.1

Full PH model Likelihood ratio test (larger value better)

1779

1799

1803

2698

2715

2716

Chisq = chi-square value, GAM = generalized additive model, PH = proportional hazards. The additional file 3 "output4rev.doc" lists the output of the full PH models.

We note that the age transform derived from node-negative cases does not provide a perfect linearization in N+ patients (Table 3). A better linearization in N+ patients was obtained by replacing the 1.5 exponent with 1.8, though without improving global model fit (Table 3).

The proportional hazard check for age shows a deviation from the assumption of constant hazard (Table 3). The "rho" values are positive when considering overall mortality, i.e. an increasing risk of death with longer follow-up. The values are negative when considering breast cancer specific mortality, i.e. a decreasing risk of breast cancer death with longer follow-up.

The age transforms suggest two components in the effect of age. One component is linear (linear for the log hazard ratio, i.e. exponential for the hazard ratio) and corresponds to a natural increase in mortality with each year of age. The other component is quasi-quadratic and is centered around age 50. It contributes to an increased risk of mortality as age increases beyond 50. It suggests a hormonally related process, not pre- versus post-menopausal, but perimenopausal versus non-perimenopausal (premenopausal + postmenopausal). The further age at diagnosis is from the age at menopause, the more prognosis is influenced by the quasi-quadratic component.

The results display a complex functional form of the effect of age on mortality. The curves clearly highlight the biological anomaly that younger patients experience the same relative mortality risk from all causes as do older patients. Figures 1 and 2 show that a 30-year old patient has a risk of death almost equal to a 60-year old patient.

The marked increase in mortality risk at older ages is attributable to the increased risk of death from causes other than breast cancer (co-morbidity). It should be noted that breast cancer does not become less virulent in older patients. An increase in the risk of death from breast cancer associated with older age was observed both in N0 and in N+ patients (Figures 3 and 4).
https://static-content.springer.com/image/art%3A10.1186%2F1471-2407-5-130/MediaObjects/12885_2005_Article_318_Fig3_HTML.jpg
Figure 3

Breast cancer mortality as a function of age for N0 patients. Dotted lines: twice-standard-error.

https://static-content.springer.com/image/art%3A10.1186%2F1471-2407-5-130/MediaObjects/12885_2005_Article_318_Fig4_HTML.jpg
Figure 4

Breast cancer mortality as a function of age for N+ patients. Dotted lines: twice-standard-error.

The German Breast Cancer Study Group GBSG-2 dataset [14] is a separate database of 686 patients. Using the GAM procedure on the GBSG-2 data, age was significantly non-linear (Chi2 = 31.744, 3 degrees of freedom, P < 0.000001). The age transforms improved the linearity for the age variable, and also improved the proportional hazards model (Table 4).
Table 4

Results of GAM and proportional hazards tests for GBSG-2 data

GBSG-2 data set

no transform

transform age+|age-50|1.5

transform age+|age-50|1.8

Hazard ratio for Age

1.001

1.004

1.002

p-value of Hazard ratio for Age

0.92

0.019

0.010

GAM Chisq for Age

31.74

9.29

8.70

GAM p-value for Age

0.0000006

0.0256

0.0336

Test PH for Age: rho

0.131

0.0459

0.0240

Test PH for Age: chisq

6.118

0.581

0.169

Full PH model Rsquare

0.135

0.142

0.143

Full PH model Likelihood ratio test

99.7

105

106

Chisq = chi-square value, GAM = generalized additive model, GBSG-2 = German Breast Cancer Study Group, PH = proportional hazards. The Additional file 2 "outputgbsg2.doc" lists the other covariates included in the PH models.

Discussion

In studies addressing the effect of age on breast cancer, several authors have reported a biphasic mortality [1519]. This large study concurs with others in the literature. As in any modeling, the validity and the utility of the model may be questioned. Data from the GBSG-2 study were considered for verification of the model. The GBSG-2 study differs from the present SEER study in several respects. This German study was a prospective controlled clinical trial about the adjuvant treatment of node-positive breast cancer patients. Inclusion of patients was not restricted by tumor size. Histopathological classification and grading were performed centrally by one reference pathologist. The GBSG-2 data have been extensively investigated for the effect of age on the prognosis of breast cancer [20]. The GBSG-2 data thus provide an indication of the capability of our results to be extrapolated to a different population. It is also complementary, since the SEER has no data on recurrence and can provide no information on disease-free survival.

Applying different methods to estimate the effect of age on event-free survival of breast cancer (linear, categorization based on cutpoints, classification and regression trees, quadratic, fractional polynomial, cubic splines), Hollaender found that all methods showed a decrease in risk with increasing age up to 45–50 years [20]. A slight increase in risk was observed for older patients in the GBSG-2 data. Taking into account the wide confidence intervals for ages older than 80 years, our Figure 4 for node-positive breast cancer specific survival shows a good concordance with the node-positive GBSG-2 event-free survival.

Regarding the proportional hazards assumption, Hollaender noted that assuming a linear risk function, a small correlation value rho of 0.147 was obtained [20]. Our result for the GBSG-2 data shows the value of rho to be 0.131 (Table 4). The small difference is attributable to the incorporation of different covariates to our proportional hazards model (additional file 2 "outputgbsg2.doc"). For the SEER data, the rho values are smaller (Table 3).

Our results are also in keeping with a closely related investigation of the SEER data in which a group of 4,616 patients 35 years old or younger was compared to a group of 20,319 patients aged 50–55 years [21]. The authors observed that younger breast cancer patients had poorer survival explained in part by presentation with later stage disease and more aggressive tumors, in terms of grade and receptor status. But the known factors could not account for the remaining unexplained difference in survival. In contradiction, recently Rapiti et al have argued that age is not an independent prognostic factor when accounting for breast tumor characteristics and treatment [22]. However, this latter study included only 82 patients who were 35 years old or younger.

In order to try to understand the biphasic mortality, we looked at hormonal status and treatments of the patients. The age of 50 corresponds to the menopause. A large proportion of younger women were estrogen receptor (ER) negative (Figure 5). The proportion of ER-negative patients decreases with increasing age without any inflection. On the other hand, the proportion of progesterone receptor (PR) negative patients increases at age 50 then slowly decreases again. The reporting of hormonal receptor status is incomplete in SEER (~33–35% missing data).
https://static-content.springer.com/image/art%3A10.1186%2F1471-2407-5-130/MediaObjects/12885_2005_Article_318_Fig5_HTML.jpg
Figure 5

Distribution of hormone receptor status by age.

Data on systemic treatment were not available from the SEER database, but the types of surgery and radiotherapy were provided. Mastectomy was performed less frequently on younger patients, but increased markedly among older patients. Post-operative radiotherapy was given less frequently at both ends of the age spectrum; somewhat less frequently in the young and considerably less frequently in the elderly patients (Figure 6). Researchers have reported under-treatment of elderly patients and this fact may account in part for the poor prognosis in the elderly [2325]. Whether hormonal status or type of treatment or other factors may explain the biphasic mortality will need to be researched.
https://static-content.springer.com/image/art%3A10.1186%2F1471-2407-5-130/MediaObjects/12885_2005_Article_318_Fig6_HTML.jpg
Figure 6

Distribution of treatment by age.

There are several limitations in the present analysis. The data are retrospective. Several orders of statistically significant interactions have not been incorporated in the models. Receiving systemic treatment is a particularly important prognostic factor in younger patients [3], but data on systemic treatment were not available for analysis.

Despite the limitations and regardless of the modeling, our major finding is that the relationship of age and mortality is biphasic. Such a finding has been described by many other authors [16, 17, 20, 26]. It is important to remember this biphasic relationship when analyzing the effect of age on patients with breast cancer. Otherwise, there is a substantial risk of misinterpreting results when age is inappropriately categorized [26] or inappropriately modeled. (Table 2 would suggest erroneously almost no effect of age on mortality). Taking into account the full shape of the relationship between age and breast cancer specific mortality, we conclude that: 1) young women experience a much higher risk of death than do older patients; 2) among elderly patients, the risk of death from breast cancer does not decrease with increasing age. These are two facts that should be remembered by those when discussing adjuvant treatment with breast cancer patients.

Conclusion

The present analysis confirms that the relationship between age and mortality is biphasic. It is important that clinical research takes this relationship into account.

Declarations

Acknowledgements

GCs: János Bolyai Research Fellowship from the Hungarian Academy of Sciences.

Authors’ Affiliations

(1)
Department of Radiation Oncology, University of Saskatchewan, Faculty of Medicine
(2)
Bács-Kiskun County Teaching Hospital, Surgical Pathology
(3)
AZ-VUB, Oncologisch Centrum, and BISI-VUB, Computer Science and Medical Informatics
(4)
Department of Gynecology and Obstetrics, Senology and Gynecologic Oncology Unit, Geneva University Hospitals
(5)
Cancer Research and Treatment Center, University of New Mexico
(6)
Department of Internal Medicine, Division of Epidemiology and Biostatistics, University of New Mexico

References

  1. Fodor J, Polgar C, Major T, Nemeth G: Locoregional failure 15 years after mastectomy in women with one to three positive axillary nodes with or without irradiation the significance of tumor size. Strahlenther Onkol. 2003, 179: 197-202. 10.1007/s00066-003-1010-7.View ArticlePubMedGoogle Scholar
  2. Freedman GM, Fowble BL, Hanlon AL, Myint MA, Hoffman JP, Sigurdson ER, Eisenberg BL, Goldstein LJ, Fein DA: A close or positive margin after mastectomy is not an indication for chest wall irradiation except in women aged fifty or younger. Int J Radiat Oncol Biol Phys. 1998, 41: 599-605. 10.1016/S0360-3016(98)00103-5.View ArticlePubMedGoogle Scholar
  3. Kroman N, Jensen MB, Wohlfahrt J, Mouridsen HT, Andersen PK, Melbye M: Factors influencing the effect of age on prognosis in breast cancer: population based study. BMJ. 2000, 320: 474-478. 10.1136/bmj.320.7233.474.View ArticlePubMedPubMed CentralGoogle Scholar
  4. Hughes KS, Schnaper LA, Berry D, Cirrincione C, McCormick B, Shank B, Wheeler J, Champion LA, Smith TJ, Smith BL, Shapiro C, Muss HB, Winer E, Hudis C, Wood W, Sugarbaker D, Henderson IC, Norton L, Cancer and Leukemia Group B; Radiation Therapy Oncology Group; Eastern Cooperative Oncology Group: Lumpectomy plus tamoxifen with or without irradiation in women 70 years of age or older with early breast cancer. N Engl J Med. 2004, 351: 971-977. 10.1056/NEJMoa040587.View ArticlePubMedGoogle Scholar
  5. Kunkler I, Williams L, Prescott R, King C: Re: Breast-conserving surgery with or without radiotherapy: pooled-analysis for risks of ipsilateral breast tumor recurrence and mortality. J Natl Cancer Inst. 2004, 96: 1255-View ArticlePubMedGoogle Scholar
  6. Vinh-Hung V, Verschraegen C, Royce M, Van de Steene J, Tai P, Cserni G, Storme G, Vlastos G: Response: Breast-conserving surgery with or without radiotherapy: pooled-analysis for risks of ipsilateral breast tumor recurrence and mortality. J Natl Cancer Inst. 2004, 96: 1255-1257.View ArticleGoogle Scholar
  7. National Cancer Institute: Surveillance, Epidemiology, and End Results (SEER) Program Public-Use Data (1973–1999), National Cancer Institute, DCCPS, Surveillance Research Program, Cancer Statistics Branch, released April based on the November 2001 submission. 2002, National Cancer Institute, Bethesda, MD, [http://seer.cancer.gov]Google Scholar
  8. Therneau TM, Grambsch PM: Modeling survival data: extending the Cox model. 2000, New York, NY, Springer-Verlag, 87-152.Google Scholar
  9. Grambsch PM, Therneau TM: Proportional hazards tests and diagnostics based on weighted residuals. Biometrika. 1994, 81: 515-526.View ArticleGoogle Scholar
  10. Vinh-Hung V, Burzykowski T, Cserni G, Voordeckers M, Van De Steene J, Storme G: Functional form of the effect of the numbers of axillary nodes on survival in early breast cancer. Int J Oncol. 2003, 22: 697-704.PubMedGoogle Scholar
  11. Verschraegen C, Vinh-Hung V, Cserni G, Gordon R, Royce ME, Vlastos G, Tai P, Storme G: Modeling the effect of tumor size in early breast cancer. Ann Surg. 2005, 241: 309-318. 10.1097/01.sla.0000150245.45558.a9.View ArticlePubMedPubMed CentralGoogle Scholar
  12. Vinh-Hung V, Gordon R: Quantitative target sizes for breast tumor detection prior to metastasis: a prerequisite to rational design of 4D scanners for breast screening. Technol Cancer Res Treat. 2005, 4: 11-21.View ArticlePubMedGoogle Scholar
  13. Vinh-Hung V, Burzykowski T, Van de Steene J, Storme G, Soete G: Post-surgery radiation in early breast cancer: survival analysis of registry data. Radiother Oncol. 2002, 64: 281-290. 10.1016/S0167-8140(02)00105-6.View ArticlePubMedGoogle Scholar
  14. Sauerbrei W, Royston P: Building multivariable prognostic and diagnostic models: transformation of the predictors by using fractional polynomials. J Roy Stat Soc A Sta. 1999, 162 (Part 1): 71-94. 10.1111/1467-985X.00122.View ArticleGoogle Scholar
  15. Bryant J, Fisher B, Gunduz N, Costantino JP, Emir B: S-phase fraction combined with other patient and tumor characteristics for the prognosis of node-negative, estrogen-receptor-positive breast cancer. Breast Cancer Res Treat. 1998, 51: 239-253. 10.1023/A:1006184428857.View ArticlePubMedGoogle Scholar
  16. Sauerbrei W, Royston P, Bojar H, Schmoor C, Schumacher M: Modelling the effects of standard prognostic factors in node-positive breast cancer. German Breast Cancer Study Group (GBSG). Br J Cancer. 1999, 79: 1752-1760. 10.1038/sj.bjc.6690279.View ArticlePubMedPubMed CentralGoogle Scholar
  17. Adami HO, Malker B, Holmberg L, Persson I, Stone B: The relation between survival and age at diagnosis in breast cancer. N Engl J Med. 1986, 315: 559-563.View ArticlePubMedGoogle Scholar
  18. Fisher ER, Anderson S, Tan-Chiu E, Fisher B, Eaton L, Wolmark N: Fifteen-year prognostic discriminants for invasive breast carcinoma: National Surgical Adjuvant Breast and Bowel Project Protocol-06. Cancer. 2001, 91: 1679-1687. 10.1002/1097-0142(20010415)91:8+<1679::AID-CNCR1183>3.0.CO;2-8.View ArticlePubMedGoogle Scholar
  19. Aebi S, Gelber S, Castiglione-Gertsch M, Gelber RD, Collins J, Thurlimann B, Rudenstam CM, Lindtner J, Crivellari D, Cortes-Funes H, Simoncini E, Werner ID, Coates AS, Goldhirsch A: Is chemotherapy alone adequate for young women with oestrogen-receptor-positive breast cancer?. Lancet. 2000, 355: 1869-1874. 10.1016/S0140-6736(00)02292-3.View ArticlePubMedGoogle Scholar
  20. Holländer N: Estimating the functional form of the effect of a continuous covariate on survival time. Dissertation. 2002, Dortmund University, [http://http/eldorado.uni-dortmund.de:8080/FB5/ls9/forschung/2002/Hollaender/hollaenderunt.pdf]Google Scholar
  21. Maggard MA, O'Connell JB, Lane KE, Liu JH, Etzioni DA, Ko CY: Do young breast cancer patients have worse outcomes?. J Surg Res. 2003, 113: 109-113. 10.1016/S0022-4804(03)00179-3.View ArticlePubMedGoogle Scholar
  22. Rapiti E, Fioretta G, Verkooijen HM, Vlastos G, Schafer P, Sappino AP, Kurtz J, Neyroud-Caspar I, Bouchardy C: Survival of young and older breast cancer patients in Geneva from 1990 to 2001. Eur J Cancer. 2005, 41: 1446-1452. 10.1016/j.ejca.2005.02.029.View ArticlePubMedGoogle Scholar
  23. Bouchardy C, Rapiti E, Fioretta G, Laissue P, Neyroud-Caspar I, Schafer P, Kurtz J, Sappino AP, Vlastos G: Undertreatment strongly decreases prognosis of breast cancer in elderly women. J Clin Oncol. 2003, 21: 3580-3587. 10.1200/JCO.2003.02.046.View ArticlePubMedGoogle Scholar
  24. Lash TL, Silliman RA, Guadagnoli E, Mor V: The effect of less than definitive care on breast carcinoma recurrence and mortality. Cancer. 2000, 89: 1739-1747. 10.1002/1097-0142(20001015)89:8<1739::AID-CNCR14>3.0.CO;2-F.View ArticlePubMedGoogle Scholar
  25. Truong PT, Lee J, Kader HA, Speers CH, Olivotto IA: Locoregional recurrence risks in elderly breast cancer patients treated with mastectomy without adjuvant radiotherapy. Eur J Cancer. 2005, 41: 1267-1277. 10.1016/j.ejca.2005.02.027.View ArticlePubMedGoogle Scholar
  26. Vinh-Hung V, Verschraegen C, Royce M, Van de Steene J, Tai P, Cserni G, Storme G, Vlastos G: Re: Breast-conserving surgery with or without radiotherapy: pooled-analysis for risks of ipsilateral breast tumor recurrence and mortality. J Natl Cancer Inst. 2004, 96: 1255-1257.View ArticleGoogle Scholar
  27. Pre-publication history

    1. The pre-publication history for this paper can be accessed here:http://www.biomedcentral.com/1471-2407/5/130/prepub

Copyright

© Tai et al; licensee BioMed Central Ltd. 2005

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Advertisement