• breast carcinoma;
  • prognostic factors;
  • long-term survival;
  • multivariate analysis


  1. Top of page
  2. Abstract


Some investigators have suggested a decreased prognostic value for conventional prognostic factors over time in patients with breast carcinoma. The objective of this study was to assess the effect of prognostic factors on the risk of death in patients with breast carcinoma over a long follow-up.


The authors assessed clinicopathologic prognostic factors in patients with early-stage breast carcinoma over a follow-up > 25 years and analyzed the variation of their effect on death in consecutive 5-year follow-up intervals. The study included 2410 women who primarily underwent complete surgical resection. Time-dependent variables were analyzed by using different multivariate models.


Four factors were related strongly to the risk of death in the first 5 years: tumor size, histologic grade, the number of involved axillary lymph nodes, and age at diagnosis. After 10–15 years of follow-up, only age at diagnosis was related to the risk of death. The effect of powerful prognostic factors, except age at diagnosis, on the risk of death was time limited, and no effects or very small effects were detectable after 10 years of follow-up.


Conventional and widely accepted prognostic factors may explain a significant portion of early deaths among patients with early-stage breast carcinoma, but they were of limited value to explain late mortality, that also may be influenced by late events, such as new primary malignancies and treatment complications. Cancer 2006. © 2006 American Cancer Society.

One-third of patients who present with operable breast carcinoma die in the first 10 years of follow-up, and 1 in 5 survivors dies in the next 10 years.1 Most patients die of disease dissemination. Adjuvant treatments to surgery have demonstrated no effect or a moderate effect on long-term survival not greater than a 10% gain in 10-year survival.2–5 The long-term risk of death relates mostly to the natural history of each breast tumor, although early and adequate local and systemic treatments have a moderate impact.2–5, 6 The prediction of lethal events is of paramount importance for 1) the knowledge of the natural history of the disease, 2) the medical decision concerning indications for adjuvant treatments, and 3) the stratification of randomized trials for the evaluation of these treatments. Many large studies have been conducted to isolate prognostic factors,7–12 and the most widely recognized are tumor size13 and histologic axillary involvement. Other factors (such as histologic grade8; age14–18; hormone and hormone receptor status19, 20; DNA tumor content21, 22; proliferative kinetics of tumor cells23; tumor proteins, oncogenes, or suppressor genes24–27; and, more recently, genomics profiling28) have variable importance, depending on whether other concomitant factors are analyzed, the selected endpoints, and the total length of follow-up. Univariate analyses should be avoided for these data; in fact, most prognostic factors are interdependent, and multivariate analyses should be used in an attempt to define the individual, independent value of each factor studied.29–33

The objectives of the current study were to assess conventional prognostic factors over a long follow-up and to analyze the variation of their effects on the risk of death in consecutive, 5-year follow-up intervals. It seemed important to determine whether conventional factors of prognosis maintain their impact on survival after 10–15 years of follow-up.

Langlands et al.34 suggested that there was a decreased effect over time for prognostic factors among patients with breast carcinoma. Those authors showed that factors with known prognostic significance at 5 years of follow-up had less influence than may have been expected when the ratio of observed-to-expected deaths was considered for over a longer follow-up. Similar results have been reported by other authors for patients with breast carcinoma or other malignancies.35–39

In the experience of the Institut Gustave-Roussy (IGR), along with tumor size,13 two factors have shown consistent prognostic value in consecutive studies: axillary lymph node involvement and histologic grade, as defined by Bloom and Richarson40 and modified by Contesso et al.,8 who previously reported their results over 10 years of follow-up.8 The current report deals with a wider clinical series including patients with breast carcinoma recruited since 1954, who were treated primarily with surgery and who were followed for up to 30 years.


  1. Top of page
  2. Abstract

Population Description

The study included 2410 women younger than age 76 years with resectable breast carcinoma who were treated between 1954 and 1978 at the IGR. These patients presented with unilateral, primary, infiltrating carcinoma of the breast that measured ≤ 7 cm in greatest dimension, and they were without fixed axillary lymph nodes (N0 or N1), without inflammatory signs, and without distant metastases (M0). Patients who previously had undergone an excisional biopsy in another hospital and foreigners whose follow-up was unavailable were not included in the study. The main patient characteristics according to axillary lymph node status are shown in Table 1. Patients were followed every 4 months during the first 2 years after treatment, every 6 months from 2 years to 5 years, once annually from 5 years to 10 years, and every 2 years thereafter. If patients were followed by another hospital or by their own physician, then regular follow-up data were obtained from those hospitals or physicians. If no follow-up data could be obtained, then the patient's status (alive or dead) was obtained regularly from their birthplace town hall. Five percent of patients were lost to follow-up at 10 years (because of mistakes in the birthplace registration or emigration). The mean follow-up among survivors was 19 years (standard deviation, 7 yrs).

Table 1. Patient Characteristics According to Axillary Lymph Node Status
Patient characteristicsLymph node statusTotal n = 2410
Negative n = 985Positive n = 1425
  1. 2 SD: two standard deviations.

Mean age in yrs (2 SD)55 (21)53 (21)54 (21)
Clinical tumor size in cm (2 SD)2.8 (2.8)3.5 (2.7)3.2 (2.8)
Medial tumor, %544951
Skin fixation. %243
Muscle fixation, %475
Histologic grade, %   
 Grade 1281520
 Grade 2455048
 Grade 3273532
No. of positive axillary lymph nodes, %   
 > 10138

Primary Treatment

The primary treatment was surgery, which included mastectomy (Halsted or Patey operation) or conservative treatment (lumpectomy and breast irradiation) and axillary dissection. Details on the indications for internal mammary-chain dissection and adjuvant radiotherapy have been published previously.41, 42 Briefly, the standard protocol depended on three factors: histologic axillary lymph node examination, tumor size, and tumor site. Patients who had positive axillary lymph nodes or medial tumors (located in the central or inner quadrants) underwent internal mammary-chain dissection and/or received postoperative lymph node radiotherapy. Other patients did not receive these adjuvant treatments. Since 1959, most premenopausal patients with involved axilla underwent ovarian ablation by radiotherapy or surgery, except for 1) patients who were enrolled on a randomized trial, 2) patients who had negative hormone receptor status after 1974, and 3) patients who presented with a tumor that measured < 2 cm in greatest dimension. A summary of applied treatments according to axillary lymph node status is provided in Table 2. Overall, 34% of patients underwent internal mammary-chain dissection, and 50% of patients received postoperative radiotherapy (99.5% of those patients had positive lymph node status). Twenty-one percent of patients underwent ovarian ablation (99.8% of those patients had positive lymph node status). No patients received adjuvant chemotherapy or additional hormone therapy.

Table 2. Primary and Adjuvant Treatments According to Axillary Lymph Node Status
TreatmentLymph node statusTotal
  • RT: radiotherapy; IMC: internal mammary chain.

  • a

    These patients presented with histologically involved IMC lymph nodes.

Breast-conserving surgery, %1158
Postoperative RT on lymph nodes, %4a8150
Complete axillary dissection, %979998
IMC dissection, %234134
Ovarian ablation, %2a3421

Statistical Methods

Preliminary univariate analyses using log-rank tests were performed to identify the factors that played a role in overall survival. All potential prognostic factors that were considered significant according to this method were included in a Cox multivariate analysis,43 and a Peduzzi–Hardi–Holford stepwise procedure44 was used to identify the major independent prognostic factors. The effects of prognostic factors over time were analyzed by using time-dependent covariates and testing the related proportional hazard. Computations were made by using BMDP software.44 The endpoint was overall survival. Because the analyzed population was treated over a span of 25 years, all analyses were stratified by 5-year calendar periods to avoid a potential cohort effect in the evaluation of prognostic factors. The reference category was the factor that corresponded in general to a lower risk of death. Regarding age, the reference category was the group of patients ages 35–44 years, because the group of patients age 35 years or younger had a higher mortality.

In a first step, the entire period of follow-up was taken into account; and, in a second step, the effects of prognostic factors were analyzed in each 5-year interval up to 25 years. Two types of tests were used to evaluate the effect of each prognostic factor as a function of the length of follow-up. First, we used the Cox model to determine whether the effect of a covariate (e.g., histologic grade) was different before and after 10 years. Second, we used the test of proportional hazard proposed by Kalbfleisch and Prentice45 by adding a time-dependent variable (variable × [ln (time)]) to the Cox model for each covariate. A different time dependency was assumed before and after 5 years. Finally, for each covariate with nonproportional hazards, we plotted its associated estimated relative risk as a function of time.


  1. Top of page
  2. Abstract

Number of Deaths

Approximately 50% of patients (1245 women) died during the entire follow-up, and 1211 women died in the first 25 years. The distribution of deaths by 5-year follow-up interval is shown in Table 3.

Table 3. Patient Distribution and Tumor Characteristics According to Five Follow-Up Periods
CharacteristicFollow-up period
0–5 yrs5–10 yrs10–15 yrs15–20 yrs20–25 yrs
  • a

    These values were the number of patients at risk at the beginning of each period.

No. of patientsa241018861319679363
No. of deaths5243561938949
Patient and tumor characteristics (%)     
 Tumor size     
  ≤ 3 cm6065655853
  3.1–5.0 cm3430303641
  5.1–7.0 cm65566
 Axillary lymph node involvement     
   1–3 lymph nodes3334353638
   4–10 lymph nodes1815121312
   > 10 lymph nodes84322
Histologic grade     
 Grade 12025282929
 Grade 24849474644
 Grade 33226252527
 < 35 yrs33223
 35–44 yrs1819212426
 45–54 yrs3233343537
 55–64 yrs2827272727
 65–75 yrs191816127

Univariate Analysis

The univariate results indicated that only the following six variables had a significant effect on survival: clinical tumor size, skin fixation, muscle fixation, histologic grade, the number of histologically involved axillary lymph nodes, and age at diagnosis.

Multivariate Analyses

Among the six variables mentioned above, the following four variables were selected for their independent effect on survival when the entire period of follow-up was taken into account: tumor size, histologic grade, the number of involved axillary lymph nodes, and age at diagnosis. The distribution of these variables in each 5-year interval up to 25 years is shown in Table 3. A moderate decrease was observed in the frequency of patients presenting with a bad prognostic factor, e.g., > 10 involved lymph nodes. However, a sufficient number of patients and events still were present in each subset to analyze each 5-year interval.

The effects of prognostic factors on survival over the entire follow-up period and in each 5-year follow-up interval are shown in Table 4. The 4 analyzed factors were related strongly to the risk of death in the first 5 years, and all of them, except tumor size, were related strongly to the risk of death between 5 years and 10 years. Between 10 years and 15 years of follow-up, only 2 factors still were predictors: age at diagnosis an axilla with more than 10 histologically involved lymph nodes. After 15 years, only age at diagnosis was related to the risk of death.

Table 4. The Relative Risk of Death According to Patient and Tumor Characteristics for Five Follow-Up Periods
Patient and tumor characteristicsLength of follow-up
Total follow-up n = 24100–5 Yrs n = 24105–10 Yrs n = 188610–15 Yrs n = 131915–20 Yrs n = 67920–25 Yrs n = 363
RRP valueRRP valueRRP valueRRP valueRRP valueRRP value
  • RR: relative risk; NS: nonsignificant.

  • a

    Reference category.

Tumor size            
 ≤ 3 cm1.0a 1.0a 1.0a 1.0a1.0a1.0a   
 3.1–5.0 cm1.210− 21.510− 31.2NS1.2NS0.9NS0.9NS
 5.1–7.0 cm1.410− 21.610− 21.2NS1.6NS1.7NS0.9NS
Axillary lymph node status            
 Negative1.0a 1.0a 1.0a 1.0a 1.0a 1.0a 
 Positive: No. of lymph nodes            
  1–31.410− 51.710− 41.610− 31.2NS1.2NS0.7NS
  4–102.310− 93.110− 92.810− 91.4NS1.3NS0.8NS
  > 104.510− 95.610− 93.810− 93.010−
Histologic grade            
 Grade 11.0a 1.0a 1.0a 1.0a 1.0a 1.0a 
 Grade 21.610− 63.110− 71.710− 31.4NS1.2NS0.9NS
 Grade 32.010−45.610− 91.810− 31.0NS0.9NS0.9NS
 < 35 yrs1.40.071.4NS1.90.050.5NS0.8NS1.0NS
 35–44 yrs1.0a 1.0a 1.0a 1.0a 1.0a 1.0a 
 45–54 yrs1.1NS0.9NS1.3NS1.6NS1.4NS2.6NS
 55–64 yrs2.010− 91.410− 21.910− 33.110−− 4
 65–75 yrs2.510−− 64.910− 85.910− 612.010− 4

Table 5 shows the result of the analysis on the effect of partitioning each variable before and after 10 years of follow-up. In the first step, models that included only one partitioned variable were compared with the conventional model without time partitions. Each partitioned variable improved the likelihood for survival of the model, and the improvement was highly significant for histologic grade, lymph node status, and age at diagnosis (P < 10−9). When all variables were included and time partitioned in the last model, the comparison showed a very significant likelihood of improvement. In addition, the test of proportional hazards using time-dependent variables was rejected for each variable after 5 years. The resulting variation in the relative risk of death associated with each variable is shown in Figure 1. The prognostic strength of each variable, except age, decayed along the time axis. In summary, the performed analyses showed consistently that the effect of powerful prognostic factors, except age at diagnosis, on the risk of death is time limited, and no effects or very small effects were detectable after 10 years of follow-up.

Table 5. Effect on the Risk of Death Before and After 10 Years of Follow-Up
ModelLog likelihoodChi-square statisticaDegrees of freedomP value
  • a

    The chi-square statistic is equal to twice the log-likelihood difference of each partitioned model and the conventional model.

  • b

    The conventional model is the Cox model that includes all four variables without time partitioning.

Conventionalb− 6877.46   
With time partitioning of variables    
Histologic grade− 6849.0756.783< 10−9
Axillary involvement− 6858.2238.484< 10−9
Tumor size− 6874.446.033< 0.025
Age at diagnosis− 6853.8547.225< 10−9
All variables− 6814.81125.2915< 10−9
thumbnail image

Figure 1. This graph illustrates the variation with time in the relative risk (RR) of overall mortality associated with each covariate. RR estimates were derived from the coefficients of the fixed and time variables in the Cox model described in the text (see Materials and Methods). The RR values indicate the prognostic strength of a variable category compared with the strength of the immediately lower variable (e.g., histologic Grade 3 vs. Grade 2). The prognostic strength of the 4 factors does not change significantly with time in the first 5 years. After 5–10 years, the prognostic strength of each variable, except age, decays significantly along the time axis.

Download figure to PowerPoint


  1. Top of page
  2. Abstract

The current results show that the effects on the risk of death of most well known and strong prognostic factors are time limited in women with operable breast carcinoma. The effects of major prognostic factors are observed in the first 5 years of follow-up. After 10 years, the effects tend to disappear, as shown in Table 4, in which each 5-year follow-up interval is analyzed separately.

This study was based on a breast carcinoma data base that included patients who were treated over 25 years. To avoid bias due to a cohort effect because of changes in recruitment and medical policies, all analyses were stratified for each 5-year calendar period. Another possible bias in the evaluation of prognostic effects may have been the early death of patients with poor prognostic factors, although this was not the case, as shown in Table 3 according to the distribution of prognostic variables in each 5-year follow-up interval up to 25 years. Even if there was a moderate decrease in the frequency of patients presenting with poor prognostic factors, there were sufficient numbers of patients and events present in each subset to analyze adequately each 5-year interval.

The first step in the multivariate analysis showed that the major prognostic effect of the covariates studied was observed in the first 5–10 years of follow-up (Table 4). The second step was a comparison of the effects of variables when the time axis was partitioned into 2 categories: before 10 years and after 10 years. The results shown in Table 5 demonstrate clearly that the model was improved in its likelihood when each partitioned-time variable was added and when the overall partitioned-time model was compared with the nonpartitioned or conventional model.

Finally, the described time-dependency phenomenon was modeled according to the proposition of Kalbfleisch and Prentice,45 and the model calculated in this mater fit quite well with the observed values. The results depicted in Figure 1 provide a concise image of the prognostic effect of each covariate along the time axis.

The ominous effect on survival of younger age has been described in other series.14–18 It is worth noting that, in the current analysis, this effect disappeared after 10-years of follow-up. This finding may be explained by the hypothesis that a subgroup of younger patients who present with tumors of poor prognosis influence the high and early mortality rate.

To interpret the observations of the current study, we should take into account the natural history of breast carcinoma, which is dominated by two main factors: 1) the growth rate of the primary tumor and of distant metastases, both of which are related strongly; and 2) the size of the tumor when distant dissemination occurs. This critical tumor size is correlated with several factors, such as histologic grade and the number of involved axillary lymph nodes. The involvement of lymph nodes certainly is not the cause of distant dissemination, but it is a good index of the propensity of tumor cells to disseminate. Patients with initially poor prognostic factors are those who have early distant dissemination and, thus, relatively large occult metastases at the time they under surgery. Moreover, although histologic grade, axillary involvement, and tumor growth rate23, 46 each have independent prognostic significance, they are correlated closely. Hence, the time intervals between treatment and the clinical emergence of distant metastases and between the detection of metastases and death are likely to be shorter. However, this scheme is blurred by the occurrence of local recurrence, which, in a large proportion of patients, may originate distant metastases.47

The current study confirmed on a larger basis the vanishing long-term effect of prognostic variables that has been suggested by others.34–39 Most prognostic studies analyzed series with a mean follow-up < 10 years, and the effects described here should not alter their results or interpretation. However, with follow-up > 10 years, careful attention should be paid to time-dependent covariates. Our results allow us to hypothesize that a patient with breast carcinoma who remains alive after 10 years of follow-up mostly is free of the ominous significance of a poor initial prognostic factor, and her risk of death is mostly from older age. Of course, most of these older patients will die from intercurrent diseases, but we do not have enough high-quality information on the late causes of death to analyze this issue.

The flattening of curves observed in Figure 1 suggests that patients who had a previously treated breast carcinoma may reach curability after 10–15 years of follow-up. However, this delay may be much longer in patients with slow tumor growth rates and late local recurrences.47 The curability of breast carcinoma has been a matter of debate in the literature.48–51 Some authors reported long-term mortality after 20 years of follow-up.52

Indeed, the results of our analysis do not indicate that patients are free of the risk of dying from primary breast carcinoma after a long follow-up, because other unknown prognostic factors may explain late recurrences. Although our current knowledge of conventional and widely accepted prognostic factors may explain a significant proportion of early deaths among women with breast carcinoma, it remains very limited with regard to an explanation for late mortality,53 which also may be influenced by new primary malignancies,54–56 including contralateral breast carcinoma, and late treatment effects.4, 5, 56

Regarding limitations of the current study, it is noteworthy that the results were obtained in the absence of systemic adjuvant treatments, except ovarian suppression, that partially have changed the natural history of breast carcinoma at least in the first 15 years of follow-up.57 Another issue is that, taking into account the long period investigated, a proportion of these patients would have been assessed with metastatic disease at the time of their diagnosis with some of the diagnostic examinations practiced today, such as bone scans and liver and lung explorations, which may account for an increased rate of early mortality.

In conclusion, the time-limited effects of conventional prognostic factors on overall survival obligate investigators to take this phenomenon into account when long-term results are reported. In addition, further research is warranted on new prognostic factors that have effective, long-term influence on late events that may affect long-term mortality.


  1. Top of page
  2. Abstract
  • 1
    Early Breast Cancer Trialists' Collaborative Group (EBCTCG). Treatment of early breast cancer: worldwide evidence 1985–1990, vol 1. Oxford: Oxford University Press, 1990.
  • 2
    Early Breast Cancer Trialists' Collaborative Group (EBCTCG). Ovarian ablation in early breast cancer: an overview of the randomised trials. Lancet. 1996; 348: 11891196.
  • 3
    Early Breast Cancer Trialists' Collaborative Group (EBCTCG). Polychemotherapy for early breast cancer: an overview of the randomised trials. Lancet. 1998; 352: 930942.
  • 4
    Early Breast Cancer Trialists' Collaborative Group (EBCTCG). Tamoxifen for early breast cancer: an overview of the randomised trials. Lancet. 1998; 351: 14511467.
  • 5
    Early Breast Cancer Trialists' Collaborative Group (EBCTCG). Favourable and unfavourable effects on long-term survival of radiotherapy for early breast cancer: an overview of the randomised trials. Lancet. 2000; 355: 17571770.
  • 6
    Nystrom L, Andersson I, Bjurstam N, Frisell J, Nordenskjold B, Rutqvist LE. Long term effects of mammography screening: updated overview of the Swedish randomised trials. Lancet. 2002; 359: 909919.
  • 7
    Hibberd AD, Horwood LJ, Wells JE. Long-term prognosis of women with breast cancer in New Zealand: study of survival to 30 years. Br Med J. 1983; 286: 17771779.
  • 8
    Contesso G, Mouriesse H, Friedman S, Genin G, Sarrazin D, Rouesse J. The importance of histologic grade in long-term prognosis of breast cancer. A study of 1010 patients uniformly treated at the Institut Gustave-Roussy. J Clin Oncol. 1987; 5: 13781386.
  • 9
    McGuire WL. Prognostic factors for recurrence and survival in human breast cancer. Breast Cancer Res Treat. 1987; 10: 510.
  • 10
    Arriagada R, Rutqvist LE, Kramar A, Johansson H. Competing risks determining event-free survival in early breast cancer. Br J Cancer. 1992; 66: 951957.
  • 11
    Fisher ER, Redmond C, Fisher B, Bass G, and Contributing NSABP Investigators. Pathologic findings from the National Surgical Adjuvant Breast and Bowel Projects (NSABP): prognostic discriminants for 8-year survival for node-negative invasive breast cancer patients. Cancer. 1990; 65: 21212128.
  • 12
    Fisher ER, Anderson S, Redmond C, Fisher B, for the NSABP Collaborating Investigators. Pathologic findings from the National Surgical Adjuvant Breast Project Protocol B-06: 10-year pathologic and clinical prognostic discriminants. Cancer. 1993; 71: 25072514.
  • 13
    Koscielny S, Tubiana M, Le MG, et al. Breast cancer, relationship between the size of the primary tumour and the probability of metastatic dissemination. Br J Cancer. 1994; 49: 709715.
  • 14
    Rutqvist LE, Wallgren A. Influence of age on outcome in breast carcinoma. Acta Radiol Oncol. 1983; 22: 289294.
  • 15
    Høst H, Lund E. Age as a prognostic factor in breast cancer. Cancer. 1986; 57: 22172221.
  • 16
    De la Rochefordiere A, Asselain B, Campana F, et al. Age as a prognostic factor in premenopausal breast carcinoma. Lancet. 1993; 341: 10391043.
  • 17
    Bonnier P, Romain S, Charpin C, et al. Age as a prognostic factor in breast cancer: relationship to pathologic and biologic features. Int J Cancer. 1995; 62: 138144.
  • 18
    Kroman N, Jensen MB, Wolfahrt J, Andersen KW, Melbye M. Factors influencing the effect of age on the prognosis of breast cancer: population based study. BMJ. 2000; 320: 474478.
  • 19
    Spyratos F, Hacene K, Tubiana-Hulin M, Pallud C, Brunet M. Prognostic value of estrogen and progesterone receptors in primary infiltrating ductal breast cancer. A sequential multivariate analysis of 1262 patients. Eur J Cancer. 1989; 25: 12331240.
  • 20
    Arriagada R, Rutqvist LE, Skoog L, Johansson H, Kramar A. Prognostic factors and natural history in lymph nodes-negative breast cancer patients. Breast Cancer Res Treat. 1992b: 21: 101109.
  • 21
    Spyratos F, Martin PM, Hacene K, et al. Multiparametric prognostic evaluation of biological factors in primary breast cancer. J Natl Cancer Inst. 1992; 84: 12661272.
  • 22
    Lonn U, Lonn S, Nilsson B, Stenkvist B. Breast cancer: prognostic significance of c-erb-B2 and Int-2 amplification compared with DNA ploidy, S-phase fraction, and conventional clinicopathological features. Breast Cancer Res Treat. 1994; 29: 237245.
  • 23
    Tubiana M, Pejovic MH, Chavaudra J, Contesso G, Malaise EP. The long-term prognostic significance of the thymidine labelling index in breast cancer. Int J Cancer. 1984; 33: 441445.
  • 24
    Thor AD, Yandel DW. Prognostic significance of p53 overexpression in node-negative breast carcinoma: preliminary studies support cautious optimism. J Natl Cancer Inst. 1993; 85: 176177.
  • 25
    Barnes DM, Dublin EA, Fisher CJ, Levison DA, Millis RR. Immunohistochemical detection of p53 protein in mammary carcinoma. An important new independent indicator of prognosis? Hum Pathol. 1993; 24: 469476.
  • 26
    Foekens JA, Van Putten WLJ, Portengen H, et al. Prognostic value of p53 and cathepsin D in 710 human primary breast tumors: multivariate analysis. J Clin Oncol. 1993; 11: 899908.
  • 27
    Harris CC, Hollstein M. Clinical implications of the p53 tumor-suppressor gene. N Engl J Med. 1993; 329: 13181327.
  • 28
    Paik S, Shak S, Tang G, et al. A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer. N Engl J Med. 2004; 351: 28172826.
  • 29
    Harrell FE, Lee KL, Califf RM, Pryor DB, Rosati RA. Regression modelling strategies for improved prognostic prediction. Stat Med. 1984; 3: 143152.
  • 30
    Simon R, Altman DG. Statistical aspects of prognostic factor studies in oncology. Br J Cancer. 1994; 69: 979985.
  • 31
    Clark GM, Hilsenbeck SG, Radvin PM, De Laurentiis M, Osborne CK. Prognostic factors: rationale and methods of analysis and integration. Breast Cancer Res Treat. 1994; 32: 105112.
  • 32
    Mazumdar M, Glassman JR. Categorizing a prognostic variable: review of methods, code for easy implementation and applications to decision-making about cancer treatments. Stat Med. 2000; 19: 113132.
  • 33
    Klinger A, Dannegger F, Ulm K. Identifying and modelling prognostic factors with censored data. Stat Med. 2000; 19: 601615.
  • 34
    Langlands AO, Pocock SJ, Kerr GR, Gore SM. Long-term survival of patients with breast cancer: a study of the curability of the disease. Br Med J. 1979; 2: 12471251.
  • 35
    Gamel JW, McLean IW, Greenberg RA. Interval-by-interval Cox model analysis of 3680 cases of intraocular melanoma shows a decline in the prognostic value of size and cell type over time after tumor excision. Cancer. 1988; 61: 574579.
  • 36
    Gray RJ. Flexible methods for analyzing survival data using splines, with applications to breast cancer prognosis. J Am Stat Assoc. 1992; 87: 942951.
  • 37
    Lipponen P, Aaltomaa S, Eskelinen M, Kosma VM, Marin S, Syrjanen K. The changing importance of prognostic factors in breast cancer during long-term follow-up. Int J Cancer. 1992; 51: 698702.
  • 38
    Yoshimoto M, Sakamoto G, Ohashi Y. Time dependency of the influence of prognostic factors on relapse in breast cancer. Cancer. 1993; 72: 29933001.
  • 39
    Nab HW, Hop WC, Crommelin MA, Kluck HM, van der Heidgen L, Coebergh JW. Changes in long term prognosis for breast cancer in a Dutch cancer registry. Br Med J. 1994; 309: 8386.
  • 40
    Bloom HJG, Richardson W. Histological grading and prognosis in breast cancer. Br J Cancer. 1957; 11: 359377.
  • 41
    Lacour J, Le MG, Hill C, Kramar A, Contesso G, Sarrazin D. Is it useful to remove internal mammary nodes in operable breast cancer? Eur J Surg Oncol. 1987; 13: 309314.
  • 42
    Arriagada R, Le MG, Mouriesse H, et al. Long-term effect of internal mammary chain treatment. Results of a multivariate analysis of 1204 patients with operable breast cancer and positive axillary nodes. Radiother Oncol. 1988; 11: 213322.
  • 43
    Cox JD. Regression models and life tables. J R Stat Soc. 1972; 34: 187220.
  • 44
    Dixon WJ, Brown MB, Engelman L, Frane JW, Hill MA. BMDP statistical software manual. Berkeley: University of California Press, 1985.
  • 45
    Kalbfleisch JD, Prentice RL. The statistical analysis of failure time data. New York: John Wiley & Sons, 1980: 8998.
  • 46
    Silvestrini R, Daidone MG, Valagussa P, et al. 3H-thymidine-labeling index as a prognostic indicator in node-positive breast cancer. J Clin Oncol. 1990; 8: 13201326.
  • 47
    Koscielny S, Tubiana M. The link between local recurrences and distant metastases in human breast cancer. Int J Radiat Oncol Biol Phys. 1999; 43: 1124.
  • 48
    Duncan W, Kerr GR. The curability of breast cancer. Br Med J. 1976; 2: 781783.
  • 49
    McBride CM, Brown BW, Thompson JR, Westbrook KC, Milne CA. Can patients with breast cancer be cured of their disease? A sample of the M. D. Anderson Hospital experience. Cancer. 1983; 51: 938945.
  • 50
    Fentiman IS, Cuzick J, Millis RR, Hayward JL. Which patients are cured of breast cancer? Br Med J. 1984; 289: 11081111.
  • 51
    Rutqvist LE, Wallgren A, Nilsson B. Is breast cancer a curable disease? A study of 14,731 women with breast cancer from the Cancer Registry of Norway. Cancer. 1984; 53: 17931800.
  • 52
    Le MG, Hill C, Rezvani A, Sarrazin D, Contesso G. Long-term survival of women with breast cancer [letter]. Lancet. 1984; ii: 922.
  • 53
    Schemper M, Henderson R. Predictive accuracy and explained variation in Cox regression. Biometrics. 2000; 56: 249255.
  • 54
    Rubino C, de Vathaire F, Shamsaldin A, Labbe M, Le MG. Radiation dose, chemotherapy, hormonal treatment and risk of second cancer after breast cancer treatment. Br J Cancer. 2003; 89: 840846.
  • 55
    Roychoudhuri R, Evans H, Robinson D, Moller H. Radiation-induced malignancies following radiotherapy for breast cancer. Br J Cancer. 2004; 91: 868872.
  • 56
    Darby SC, McGale P, Taylor CW, Peto R. Long-term mortality from heart disease and lung cancer after radiotherapy for early breast cancer: prospective cohort study of about 300,000 women in US SEER cancer registries. Lancet Oncol. 2005; 6: 557565.
  • 57
    Early Breast Cancer Trialists' Collaborative Group (EBCTCG). Effects of chemotherapy and hormonal therapy for early breast cancer on recurrence and 15-year survival: an overview of the randomised trials. Lancet. 2005; 365: 16871717.