|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
© 2000 American Society for Clinical Oncology Time-Dependent Relevance of Steroid Receptors in Breast CancerFrom the Unità Operativa Determinanti Biomolecolari nella Prognosi e Terapia; Statistica Medica e Biometria; Unità di Chirurgia Ginecologica; Unità di Anatomia Patologica, Istituto Nazionale per lo Studio e la Cura dei Tumori; Istituto di Statistica Medica e Biometria, Università degli Studi; and Centro per lo Studio della Patologia Cellulare, Consiglio Nazionale delle Ricerche, Milan, Italy. Address reprint requests to Danila Coradini, PhD, Unità Operativa Determinanti Biomolecolari nella Prognosi e Terapia, Istituto Nazionale Tumori, Via Venezian 1, 20133 Milan, Italy; email coradini @istitutotumori.mi.it.
PURPOSE: To analyze the time-dependent prognostic role of the investigated variables, considered, when appropriate, on a continuous scale, for the purpose of evaluating and describing the interrelationships between clinically relevant patient and tumor characteristics (age, size and histology, and estrogen receptor [ER] and progesterone receptor content) and the risk of new disease manifestation. PATIENTS AND METHODS: We applied a flexible statistical model to a case series of 1,793 patients with axillary lymph node–negative breast cancer with a minimal potential follow-up of 10 years. To avoid a potential confounding effect of adjuvant treatment, only patients given local-regional therapy until relapse were considered. RESULTS: ER content and tumor size (adjusted for all the other covariates) showed a time-dependent relationship with the risk of new disease manifestations. In particular, ER content failed to show a prognostic effect within the first years of follow-up; thereafter, a positive association with risk of relapse was observed. For tumor size, within the first years of follow-up, the risk of relapse was directly related to size for only tumors up to 2.5 cm in diameter; thereafter, the impact on prognosis progressively decreased. CONCLUSION: The availability of a long follow-up on a large breast cancer series, as well as the use of innovative statistical approaches, allowed us to explore the functional relation between steroid receptors and clinical outcome and to generate a hypothesis on the involvement of ER in favoring long-term metastasis development.
ESTROGEN RECEPTOR (ER) and progesterone receptor (PgR) content has been used since 1974 in the clinical management of human breast cancer, initially as an indicator of endocrine responsiveness,1,2 then as a prognostic factor for early recurrence.3 During the two successive decades, hundreds of articles have been published on these two related, although different, aspects with clinical implications. Whereas the role of ER and PgR content as a predictor of response to treatment has been consistently recognized,4,5 its prognostic relevance has been often contradicted6-11 and is still debated. Disagreement over the prognostic role of steroid receptors may be ascribed to heterogeneity in analytic techniques and clinical characteristics of case series and to an improper use of statistical approaches to analyze their prognostic impact. In particular, small or large size of data sets, length of follow-up, differences in clinicopathologic features of patients (ie, status and degree of axillary lymph node involvement and different types of local-regional and/or systemic adjuvant treatments), and exclusion in the statistical analysis of some pathobiologic or clinical variables relevant for breast cancer natural history have led to different conclusions concerning the prognostic impact of steroid receptor content. In addition, the use of an arbitrary cut point to discriminate between receptor-rich and receptor-poor tumors (with the consequent concept of receptor status)—although it simplifies statistical analysis, results interpretation, and their direct translation to clinical practice—is not free from several risks.12 In fact, the practice oversimplifies prognostic relationships, which leads to the implication of a threshold prognostic relationship, an assumption which is generally not evaluated, and of an equal clinical outcome for patients who belong to the same categories determined by cutoff values. All of these assumptions lead to the loss of quantitative information in an early phase of the decision process and to the introduction of possible bias effects.13 The use of a continuous scale to analyze the association between steroid receptors content and other relevant clinical variables and to evaluate their joint prognostic contribution is expected to improve clinical practice by providing helpful details on tumor biology. The aim of the present study was to elucidate and describe the interrelationships between clinically relevant patient and tumor characteristics (age, size and histology, and ER and PgR content) and the risk of new disease manifestation, considering pathologic and biologic variables on a continuous scale and considering time-dependent effects. To avoid any potential confounding effect of systemic therapy, we analyzed a case series of 1,793 previously untreated patients who were homogeneous for conventional clinicopathologic factors, ie, axillary lymph node–negative (N-) tumors and receiving no postsurgical adjuvant treatment until relapse, who were treated in a single institution and had a minimal potential follow-up of 10 years. The availability of a long-term follow-up and the application of a suitable statistical methodology allowed us to thoroughly investigate the time-dependent role of hormone receptors on prognosis, which has been reported in previous studies,11,14,15 and to emphasize the complexity of ER-related mechanisms.16,17
Patients The case series consisted of 1,793 women with primary resectable invasive breast cancer, histologically N-, and with no radiologic or clinical evidence of distant metastasis, a synchronous bilateral tumor, or a concomitant second primary tumor. Cases with these clinicopathologic features and with a minimum potential of 10 years of follow-up (ie, the time elapsed from the date of surgery to the date of the last updating of the patient records) were selected from the approximately 7,000 women with operable tumors, consecutive with respect to ER and PgR determinations at the time of diagnosis, who underwent surgery at the Istituto Nazionale Tumori of Milan during the period from January 1981 to December 1986. Patients were treated with mastectomy (864 cases; 48%) or quadrantectomy plus radiotherapy (929 cases; 52%),18 and all of them underwent axillary node dissection (median number of examined nodes, 18). None of the women received systemic postoperative therapy until new disease manifestation was documented. A total of 804 patients (45%) were  50 years, and 806 (45%) were in premenopause (ie, actively menstruating or having the last menses less than 2 years before breast cancer detection). The median age for the entire series was 52 years (25th percentile, 45 years; 75th percentile, 62 years). The greatest pathologic tumor diameter was 2 cm (T1) in 1,257 patients (70%). The median tumor size for the entire series was 1.7 cm (25th percentile, 1.2 cm; 75th percentile, 2.3 cm). The most frequent histologic type was invasive ductal carcinoma, pure (1,217 cases; 68%) or associated with lobular (122 cases; 7%) or with other histologic types (85 cases; 5%). Lobular invasive carcinoma included 213 pure cases (12%) and 22 cases associated with histologic type other than ductal (1%). Other histologic types (tubular, medullary, papillary, and mucinous) accounted for a total of 134 cases (7%). Women underwent follow-up examinations at the outpatient clinic of the Istituto Nazionale Tumori at planned 4- to 6-month intervals during the first 5 years and at 12-month intervals thereafter. In addition to a routine clinical examination, disease assessment included mammography, chest roentgenogram, skeletal survey, and liver ultrasonography. Primary treatment failure, considered for computing disease-free survival (DFS), was defined as the first documented evidence of local recurrence or regional axillary relapse (183 cases), distant metastasis (225 cases), contralateral breast cancer (119 cases), or other second primary cancer (75 cases). Local, regional, contralateral, and distant failures were accurately assessed by clinical, radiologic, and, whenever possible, histopathologic evaluations. Of the patients included in the study, 349 had died at the time of this writing, 281 because of the relapse of breast cancer and 68 for other causes without evidence of relapse. The median observed follow-up time was 127 months (25th percentile, 79; 75th percentile, 148), with 4% of cases lost to follow-up during this period.
Receptor Determination
Statistical Analysis Prognostic relationships. The effect of age, ER and PgR contents, tumor size, and histology on DFS was investigated by multivariate analysis using a Cox regression model. In fact, the goal of the study was to evaluate the prognostic effect of ER and PgR content taking into account the concomitant effects of other known prognostic factors. Histology was categorized as follows: lobular/lobular plus associated histologic types, ductal/ductal plus associated histologic type, mixed (ductal plus lobular), and other histologic types (tubular, medullary, papillary, and mucinous). In the regression models, steroid receptors content and age were considered as continuous variables, the latter in its original measure scale and the former in terms of its natural logarithms because of the positive skew of its distribution. Null values for steroid receptor content were arbitrarily set to 1 considering a sensitivity threshold value of 2 fmol/mg of cytosolic protein. On the basis of the findings of Veronesi et al,25 who in a clinical trial found superimposable DFS curves for radical mastectomy and quadrantectomy, the type of surgery was not considered in the statistical analysis. Models were performed on 1,715 (95.6%) cases, with complete information on the considered variables. The analysis was based on a minimum of 10 years of follow-up information. The constant effects of continuous variables were prespecified according to prior considerations. Four- and three-knots restricted cubic spline regression were used to model in a flexible way the relationship between the logarithm of hazards ratio (HR) and age and tumor size,26 respectively. A more complex effect was hypothesized for age because of the relationships with hormone receptor contents. On the basis of prior knowledge,27 only linear terms were considered for ER and PgR. In the basic form of the Cox regression model, the coefficients corresponded to the logarithm of the HR and were constant in time. This assumption was graphically evaluated by means of smoothed Schoenfeld residuals and tested as suggested by Grambsch and Therneau.28 Because time-dependent covariates can be included in the Cox regression model, nonproportional effects were modelled in a flexible way as the cross-product of the restricted cubic spline terms of time and those of the considered variables.29 In the initial model, we included time-dependent effects for all of the considered variables. From graphical evaluations, possible nonlinear time effects were suggested for all continuous variables, whereas histology nonlinear time effects were not evident. Thus, a four-knots cubic spline on time was adopted for continuous variables, and only the linear term on time was adopted for histology. In the initial model, all variables were investigated and their time-dependent effects were considered. A final model was obtained by applying a backward selection procedure only on time-dependent effect terms. A likelihood ratio test at the conventional significance level of P = .05 was adopted as a stopping criterion. Because of the application of a model selection procedure, the significance levels of the final model may be overstated and should be considered with caution. The results from the final model were graphically provided, and the HR was plotted as a function of covariate values and time. Statistical analysis was performed by S-plus (Statistical Sciences, Seattle, WA) and SAS (SAS Institute, Cary, NC) software packages. The programs written by Harrell et al30 and the macros written by Heinzl and Kaider29 were applied for some of the steps of the model-building procedure.
Relationships Among ER and PgR Contents, Age, Tumor Size, and Histology Concerning MCA, the plane identified by the first two factorial axes proved to explain 92% of the total inertia of the data (Fig 1). The categories for the variables are displayed in Fig 1 as points: ER content ( 10, 11 to 36, 37 to 83, 84 to 198, and > 198 fmol/mg cytosolic protein), PgR content ( 6, 7 to 36, 37 to 115, 116 to 307, and > 307 fmol/mg cytosolic protein), age ( 43, 44 to 49, 50 to 57, 58 to 65, and > 65 years), tumor size ( 1.1, 1.2 to 1.5, 1.6 to 2.0, 2.1 to 2.5, and > 2.5 cm), and histology. The first factorial axis was mainly determined by ER (40%) and PgR contents (34%), whereas the contributions of age, tumor size, and histology were 3%, 10%, and 13%, respectively. The second factorial axis was mainly determined by age (39%) and ER content (45%), whereas the contributions of tumor size, PgR content, and histology were 6%, 8%, and 2%, respectively. According to the expected correlation between the two variables, the projections of ER and PgR categories showed similar patterns. Young age was characterized by tumors having a low-to-intermediate ER content (category with ER 83 fmol/mg protein), whereas old age was associated with a high ER content (category with ER content > 198 fmol/mg protein). Lobular histologies were characterized by an intermediate ER/PgR content (categories with ER content from 37 to 83 fmol/mg protein and with PgR content from 37 to 115 fmol/mg protein) and a small tumor size (category with diameter 1.1 cm), whereas mixed histologies were associated with a high ER/PgR content (categories with ER content from 84 to 198 fmol/mg protein and with PgR content > 307) and an intermediate tumor size (category with diameter 1.2 to 1.5 cm). Medullary histology was characterized by a low ER/PgR content (categories with ER content 10 fmol/mg protein and with PgR content 6) and a large tumor size (category with diameter > 2.5 cm). All the reported category values should be considered only as indicative of the general association trends because the grouping of variables was arbitrary.
Prognostic Relationships In the final multivariate model, ER content, PgR content, age, tumor size, and histology were significantly related to prognosis (Table 1), and a time-dependent effect for ER content, tumor size, and histology was identified. In particular, for ER content and histology, the pattern of dependence of log(HR) with time was linear, whereas a complex time pattern was observed for tumor size. The time-dependent relationship between HR and prognostic variables was shown by plotting the former as a function of covariate values at different follow-up times (ie, 36 months, representative of an early event, and 72 months, the most common follow-up reported in translational studies) and over time for some selected comparisons of covariate values.
For ER content at 36 months of follow-up (Fig 2a) and considering 1 fmol/mg of cytosolic protein as a reference, no prognostic effect of receptor content was observed, whereas at 72 months (Fig 2b), HR was positively related to ER content. The effect on the hazard of disease of a small increase in ER concentration over follow-up time was plotted by taking into account two different ER levels: low (10 v 1 fmol/mg of cytosolic protein, Fig 2c), which was considered as a negative value according to the commonly used cutoff, and intermediate (20 v 10 fmol/mg of cytosolic protein, Fig 2d), which was considered as a positive value according to any cutoff currently used. At intermediate ER content levels, a small increase in ER concentration did not seem to affect HR, whereas at low levels the presence of a small but detectable amount of ER showed a positive relation with the HR. Because the effect of ER and PgR content is modeled as a linear term on the logarithm of the concentration, constant increments in their values have the highest effect when starting from low concentrations. Such a finding became clinically relevant for late relapse, HR being greater than 1 after approximately 48 months of follow-up time. In contrast, HR did not show a time-dependent relation with PgR content (Table 1) because, considering the value of 1 fmol/mg cytosolic protein as a reference, a decrease in the HR by increasing receptor concentration was always observed (Fig 2e).
Considering as a reference category the median age (57 years), the relation between patient age at time of surgery and HR was nonlinear (Fig 2f). In fact, an HR of approximately 2 for patients younger than 30 years progressively decreased to 1 for those aged 45 years, remained stable for women aged between 45 and 60 years, and progressively decreased below 1 for women older than 60 years. A nonlinear and time-dependent relation between HR and tumor size was also found. By considering 0.5 cm (ie, the smallest tumor size on which steroid receptor determination was performed in our laboratory) as a reference value, at 36 months the HR seemed to be directly related to a tumor diameter of up to 2.5 cm (Fig 3a). Thereafter, no further increase in the HR was found. At a longer follow-up time (ie, 72 months), the impact of tumor size on prognosis decreased (Fig 3b). Considering a constant size increment of 0.5 cm, the effect of an equal absolute difference in tumor size proved to be greater for small (1.5 v 1.0 cm, Fig 3c) than for large tumors (2.5 v 2.0 cm, Fig 3d). For tumors smaller than 2 cm in diameter (Fig 3c), the HR of relapse increased to a value of 1.5 for up to 40 months; thereafter, it decreased and approximated 1.0 after approximately 60 months. For tumors larger than 2 cm in diameter (Fig 3d), the HR was slightly greater than 1 only within 2 years.
Time-dependent effects were also detected for histologic categories (data not shown). In fact, the risk of relapse for lobular and ductal carcinomas was similar until approximately 80 months, whereas for longer follow-up times, the HR of lobular was slightly greater than that of ductal carcinoma. When ductal carcinoma was compared with mixed lobular plus ductal histologic type, the HR of the latter tumors was significantly greater than that of ductal carcinoma after only approximately 70 months of follow-up, which thus suggests that even in a mixed tumor the presence of a lobular component is associated with an increased risk of relapse. No prognostic difference was found when other histologies were compared with ductal.
The availability of a long follow-up time and the determination of steroid receptors by the biochemical method in a substantial series of N- breast cancer patients who were given only local-regional treatment allowed a thorough exploration of the prognostic role of ER and PgR content on the continuous scale accounting for the clinically relevant patient and tumor characteristics. Moreover, the application of flexible statistical models allowed us to elicit time-dependent relations between pathobiologic variables and clinical outcome generally not disclosed otherwise. In fact, ER content, which did not provide any prognostic information within the first years of follow-up, was thereafter positively associated with relapse. In contrast, high PgR concentrations were negatively related to risk of relapse, and this finding was consistent over the entire follow-up period. The positive time-dependent relation between ER content and risk of relapse could seem unexpected and inconsistent according to the general finding that patients with a poor prognosis prevalently present at diagnosis with ER-poor tumors. However, our results are supported by similar findings of a time-dependency reported by other authors.15 The difference in developing a relapse as a function of ER content could be explained by a faster non–estrogen-related growth rate of subclinical metastases from ER-poor tumors31 and by a counteracting effect of estrogens in the development of metastases for an initial period after the removal of primary ER-rich tumors.32-34 Relapse after primary tumor removal represents one of the major problems in the management of breast cancer patients, because latent micrometastases often remain asymptomatic and clinically undetectable for a long time. The present findings suggest a pivotal role of ER in the mechanism through which quiescent tumor cells shift to a metastatic behavior, with a possible link between estrogen-related mechanisms and acquisition of metastatic properties.35 In fact, to migrate within the extracellular matrix, tumor cells produce proteolytic enzymes, such as cathepsin D and urokinase-type plasminogen activator (u-PA), which are both estrogen-inducible through a receptor-mediated pathway36,37 and which are coexpressed in approximately two thirds of breast cancers.38 In particular, experimental and clinical studies have shown the following: (1) interaction of u-PA and its membrane-bound receptor elicits endothelial cell chemotaxis and chemokinesis during angiogenesis39,40; (2) u-PA and its receptor are especially expressed by ER-positive tumor cells41; and (3) u-PA is associated with poor prognosis.42 Micrometastasis growth is controlled by several factors in the microenvironment. In particular, it is favored by low levels of ovarian estrogens in young and nonovarian estrogens in elderly women and counteracted by natural antiangiogenic factors, the most investigated of which is angiostatin.43,44 In experimental systems, continuous daily administration of angiostatin is required to maintain the dormancy of residual microscopic tumor nodules, which resume growth within weeks on cessation of therapy.45 Therefore, the angiogenic process seems to be regulated by the relative balance of a complex interaction of positive and negative signals whose disruption can shift the equilibrium toward angiogenesis and metastatic spread. We can thus speculate that long-term relapses, mainly observed in patients with ER-rich tumors, may occur because of an imbalance between the estrogen-induced plasminogen activator and antiangiogenic factors such angiostatin and that ER-rich clones might benefit from the growth stimulation of low but continuous levels of ovarian or extraovarian estrogens. Tumor size also showed a time-dependent relationship with risk of relapse. The early prognostic importance of size, which is particularly evident within small tumors, could be explained by the experimental evidence that a small tumor mass is prevalently constituted of living and actively proliferating cells, which initiate angiogenesis and, subsequently, metastatic spread after exposure to a hypoxic stimulus. However, a further increase in tumor diameter does not imply a proportional increase in aggressiveness because of a decreased ratio between living and dead cells. In fact, histologic types characterized by a large tumor mass and a negative ER status (for example, medullary carcinoma) are generally associated with a favorable prognosis, whereas lobular carcinomas (known to be associated with intermediate values of ER content according to the present and other findings14) have been shown to be related to a time-dependent increase in risk of relapse. The nonlinear relation between patient age and risk of relapse was also expected according to the relation between age and endocrine patient milieu, with a better prognosis for elderly than for young women. In conclusion, the complex functional relation between pathobiologic features and clinical outcome, disclosed by considering continuous variables and using innovative statistical approaches, emphasizes the need to study the involvement of steroid receptor–related mechanisms in tumor progression and the need for a more intensive follow-up at different time points for ER-positive tumors and for specific histologies such as invasive lobular. Moreover, from a biologic point of view the proposed interaction with angiogenetic mechanisms thus represents a hypothesis that merits further investigation.
Supported in part by grants from the Italian Association for Cancer Research, Ministero della Sanità, and Consiglio Nazionale delle Ricerche (99.02444.CTO), Rome, Italy. We thank B. Johnston for editing the article.
D.C. and M.G.D. contributed equally to the planning and execution of this article.
1. De Sombre ER, Smith S, Block GE, et al: Prediction of breast cancer response to endocrine therapy. Cancer Chemother Rep 58:5113-5119, 1974
2.
Horwitz KB, McGuire WL: Predicting response to endocrine therapy in human breast cancer: A hypothesis. Science 189:726-727, 1975
3.
Knight WA, Livinston RB, Gregory EJ, et al: Estrogen receptor as an independent prognostic factor for early recurrence in breast cancer. Cancer Res 37:4669-4671, 1977 4. Knight WA, Osborne CK, Yochmowitz MG, et al: Steroid hormone receptors in the management of human breast cancer. Ann Clin Res 12:202-207, 1980[Medline] 5. Allegra JC, Lippman ME, Thompson EB, et al: Estrogen receptor status: An important variable in predicting response to endocrine therapy in metastatic breast cancer. Eur J Cancer 16:323-331, 1980
6.
Mason BH, Holdaway IM, Mullins PR, et al: Progesterone and estrogen receptors as prognostic variables in breast cancer. Cancer Res 43:2985-2990, 1983 7. Ciatto S, Bravetti P, Cardona G, et al: Prognostic role of estrogen receptor determination in breast cancer. Tumori 69:527-530, 1983[Medline] 8. Aamdal S, Bormer O, Jorgensen O, et al: Estrogen receptors and long-term prognosis in breast cancer. Cancer 53:2525-2529, 1984[Medline] 9. Gelbfish GA, Davidson AL, Kopel S, et al: Relationship of estrogen and progesterone receptors to prognosis in breast cancer. Ann Surg 207:75-79, 1988[Medline] 10. Butler JA, Bretsky S, Menendez-Botet C, et al: Estrogen receptor protein of breast cancer as a predictor of recurrence. Cancer 55:1178-1181, 1985[Medline] 11. Broet P, Pichon MF, Magdelenat H, et al: Long-term prognostic role of steroid receptors in cancer of the breast. Bull Cancer 85:347-352, 1998[Medline] 12. Altman DG: Suboptimal analysis using ‘optimal’ cutpoints. Br J Cancer 78:550-557, 1998[Medline] 13. Biganzoli E, Boracchi P, Daidone MG, et al: Flexible modelling in survival analysis: Structuring biological complexity from the information provided by tumor markers [review]. Int J Biol Markers 13:107-123, 1998[Medline] 14. Hupperets PS, Vovovics L, Schouten LJ, et al: The prognostic significance of steroid receptor activity in tumor tissues of patients with primary breast cancer. Am J Clin Oncol 20:546-551, 1997[Medline] 15. Hilsenbeck SG, Ravdin PM, de Moor CA, et al: Time-dependence of hazard ratios for prognostic factors in primary breast cancer. Breast Cancer Res Treat 52:227-237, 1998[Medline] 16. Johns A, Freay AD, Fraser W, et al: Disruption of estrogen receptor gene prevents 17 beta estradiol-induced angiogenesis in transgenic mice. Endocrinology 137:4511-4513, 1996[Abstract] 17. Shekhar MP, Nangia-Makker P, Wolman SR, et al: Direct action of estrogen on sequence of progression of human preneoplastic breast disease. Am J Pathol 152:1129-1132, 1998[Abstract] 18. Veronesi U, Saccozzi R, Del Vecchio M, et al: Comparing radical mastectomy with quadrantectomy, axillary dissection and radiotherapy in patients with small cancers of the breast. N Engl J Med 305:6-11, 1981[Abstract] 19. Ronchi E, Granata G, Brivio M, et al: A double-labeling assay for simultaneous estimation and characterization of estrogen and progesterone receptors using radioiodinated estradiol and tritiated Org 2058. Tumori 72:251-257, 1986[Medline] 20. Revision of the standards for the assessment of hormone receptors in human breast cancer: Report of the second E.O.R.T.C. Workshop, held on 16–17 March, 1979, in the Netherlands Cancer Institute. Eur J Cancer 16:1513-1515, 1980 21. Piffanelli A, Pelizzola D, Giovannini G, et al: Characterization of laboratory working standard for quality control of immunometric and radiometric estrogen receptor assays: Clinical evaluation on breast cancer biopsies—Italian Committee for Hormone Receptor Assays Standardization. Tumori 75:550-556, 1989[Medline] 22. Greenacre MJ: Correspondence Analysis in Practice. London, United Kingdom,Academic Press, 1993 23. Benzecri JP: Sur le calcul des taux d’inertie dans l’analyse d’un questionnaire. Cah Anal Données 4:377-383, 1979 24. Lebart L, Morineau A, Piron M: Statistique Exploratoire Multidimensionelle. Paris, France,Dunod, 1995 25. Veronesi U, Banfi A, Del Vecchio M, et al: Comparison of Halsted mastectomy with quadrantectomy, axillary dissection, and radiotherapy in early breast cancer: Long-term results. Eur J Cancer Clin Oncol 22:1085-1098, 1986[Medline] 26. Durrleman S, Simon R: Flexible regression models with cubic splines. Stat Med 8:551-561, 1989[Medline] 27. Silvestrini R, Daidone MG, Benini E, et al: Validation of p53 accumulation as a predictor of distant metastasis at 10 years of follow-up in 1400 node-negative breast cancers. Clin Cancer Res 2:2007-2013, 1996[Abstract]
28.
Grambsch P, Therneau T: Proportional hazards tests and diagnostics based on weighted residuals. Biometrika 81:515-526, 1994 29. Heinzl H, Kaider A: Gaining more flexibility in Cox proportional hazards regression models with cubic spline functions. Comput Methods Programs Biomed 54:201-218, 1997[Medline] 30. Harrell FE Jr, Lee KL, Mark DB: Multivariable prognostic models: Issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 15:361-387, 1996[Medline] 31. Silvestrini R, Valagussa P, Di Fronzo G, et al: Cell kinetics as a prognostic indicator in node-negative breast cancer. Eur J Cancer Clin Oncol 25:1165-1171, 1989[Medline] 32. Sheikh MS, Garcia M, Pujol P, et al: Why are estrogen-receptor-negative breast cancers more aggressive than the estrogen-receptor-positive breast cancers? Invasion Metastasis 14:329-336, 1994-1995[Medline] 33. Petrangeli E, Lubrano C, Ortolani F, et al: Estrogen receptors: New perspectives in breast cancer management. J Steroid Biochem Mol Biol 49:327-331, 1994[Medline] 34. Garcia M, Rochefort H: Estrogens and breast cancer: From action mechanisms to clinical applications. Ann Endocrinol 56:543-545, 1995[Medline] 35. Folkman J: Tumor angiogenesis, in Mendelsohn J, Howley PM, Istrael MA, et al (eds): The Molecular Basis of Cancer. Philadelphia, PA,WB Saunders, 1995, pp 206-232 36. Yamashita J, Horiuchi S, Shigaki N, et al: Estrogen-dependent plasminogen activator in 7,12-dimethylbenz[a]anthracene-induced rat mammary tumors in vivo and in vitro. Gann 75:681-689, 1984[Medline]
37.
Ryan TJ, Seeger JI, Kumar SA, et al: Estradiol preferentially enhances extracellular tissue plasminogen activators of MCF7 breast cancer cells. J Biol Chem 259:14324-14327, 1984 38. Gohring UJ, Scharl A, Thelen U, et al: Comparative prognostic value of cathepsin D and urokinase plasminogen activator detected by immunohistochemistry in primary breast carcinoma. Anticancer Res 16:1011-1018, 1996[Medline] 39. Duffy MJ, O’Grady P: Plasminogen activator and cancer. Eur J Cancer Clin Oncol 20:577-582, 1984[Medline] 40. Schnaper HW, Barnathan ES, Mazaar A, et al: Plasminogen activators augment endothelial cell organization in vitro by two distinct pathways. J Cell Physiol 165:107-118, 1995[Medline] 41. Yamashita J, Horiuchi S, Kimura M, et al: Plasminogen activator as a functional marker for estrogen dependence in human breast cancer cells. Jpn J Cancer Res 77:177-181, 1986[Medline] 42. Shiba E, Kim SJ, Taguchi T, et al: A prospective study on the prognostic significance of urokinase-type plasminogen activator levels in breast cancer tissue. J Cancer Res Clin Oncol 123:555-559, 1997[Medline]
43.
Patterson BC, Sang QA: Angiostatin-converting enzyme activities of human matrilysin (MMP-7) and gelatinase B/type IV collagenase (MMP-9). J Biol Chem 272:28823-28825, 1997
44.
Gately S, Twardowski P, Stack MS, et al: The mechanism of cancer-mediated conversion of plasminogen to the angiogenesis inhibitor angiostatin. Proc Natl Acad Sci U S A 94:10868-10872, 1997 45. O’Reilly MS, Holmgren L, Chen C, et al: Angiostatin induces and sustains dormancy of human primary tumors in mice. Nat Med 2:689-692, 1996[Medline] Submitted June 28, 1999; accepted March 8, 2000. This article has been cited by other articles:
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2000 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
TOP
ABSTRACT
INTRODUCTION
2 metric, indicates the dissimilarities between categories. A high variability concentrated on the first two axes implies the presence of a relevant association among variables. In such a case, MCA allows low-dimensional images that well approximate the structure of the association among categories of the original high-dimensional space of the variables. Although the application of MCA requires the categorization of continuous variables in an arbitrary number of classes (in the present study, by the quintile values of their distribution), the technique has the advantage of implying neither linearity nor specific distribution assumptions.
