A Pooled Analysis of Bone Marrow Micrometastasis in Breast Cancer
http://www.100md.com
《新英格兰医药杂志》
ABSTRACT
Background We assessed the prognostic significance of the presence of micrometastasis in the bone marrow at the time of diagnosis of breast cancer by means of a pooled analysis.
Methods We combined individual patient data from nine studies involving 4703 patients with stage I, II, or III breast cancer. We evaluated patient outcomes over a 10-year follow-up period (median, 5.2 years), using a multivariable piecewise Cox regression model.
Results Micrometastasis was detected in 30.6 percent of the patients. As compared with women without bone marrow micrometastasis, patients with bone marrow micrometastasis had larger tumors and tumors with a higher histologic grade and more often had lymph-node metastases and hormone receptor-negative tumors (P<0.001 for all variables). The presence of micrometastasis was a significant prognostic factor with respect to poor overall survival and breast-cancer–specific survival (univariate mortality ratios, 2.15 and 2.44, respectively; P<0.001 for both outcomes) and poor disease-free survival and distant-disease–free survival during the 10-year observation period (incidence-rate ratios, 2.13 and 2.33, respectively; P<0.001 for both outcomes). In the multivariable analysis, micrometastasis was an independent predictor of a poor outcome. In the univariate subgroup analysis, breast-cancer–specific survival among patients with micrometastasis was significantly shortened (P<0.001 for all comparisons) among those receiving adjuvant endocrine treatment (mortality ratio, 3.22) or cytotoxic therapy (mortality ratio, 2.32) and among patients who had tumors no larger than 2 cm in diameter without lymph-node metastasis and who did not receive systemic adjuvant therapy (mortality ratio, 3.65).
Conclusions The presence of micrometastasis in the bone marrow at the time of diagnosis of breast cancer is associated with a poor prognosis.
Data from experiments in animals1 performed in the 1960s and from more recent immunocytochemical2,3 and molecular4,5 studies suggest that lymph-node involvement does not accurately predict hematogenous dissemination of cancer cells, nor is hematogenous dissemination necessarily associated with lymph-node involvement.6,7 During the past two decades, several studies have assessed the prevalence and prognostic value of hematogenous dissemination of tumor cells in patients with node-positive and node-negative breast cancer.3,8,9,10,11,12,13,14,15 The influence of the presence of micrometastasis in the bone marrow on prognosis has been shown in patients with identical stages of breast cancer, as defined by tumor size, histologic grade, presence or absence of lymph-node metastasis, and expression of hormone receptors.3,9,10,11,12,13 However, the clinical usefulness of finding such micrometastasis is limited by the low statistical power of published studies and the lack of clinical trials specifically investigating the predictive role of bone marrow micrometastasis. To date, only two small studies have reported the outcome of patients with bone marrow micrometastasis10,12 well beyond a median observation time of five years. In this study, we investigated the long-term outcome of patients with and those without bone marrow micrometastasis. We also explored the effect of bone marrow micrometastasis on prognosis in clinically relevant subgroups. To accomplish these goals, we analyzed pooled data from nine independent studies with updated follow-up data and numbers of patients; these studies involved 4703 patients with stage I, II, or III breast cancer who were treated in Augsburg and Munich, Germany (two independent studies that were initially published together),11 Paris,8 Oslo,9 Rostock, Germany,3 New York,15 Erlangen, Germany,10 Heidelberg, Germany,13 and London.12 (These studies are referred to hereinafter by the names of the cities.)
Methods
Data Collection
The National Library of Medicine of the National Center for Biotechnology Information (www.ncbi.nlm.nih.gov) was searched for studies related to bone marrow micrometastasis and survival of patients with breast cancer. Six such studies were identified.3,10,11,12,13 Furthermore, on the basis of personal contact, we knew of three studies that were in the process of manuscript preparation15 or submission8,9 at the time we were collecting our data. We defined eligible patients as women with complete baseline clinical records, data on follow-up examinations, and histologically confirmed and completely removed primary stage I, II, or III breast cancer, with information on tumor size and the presence or absence of axillary lymph-node metastasis. We further required documented validation of the immunoassays used to detect micrometastasis. Patients were excluded from the analysis if they had in situ carcinomas only, if they had either distant metastases or local recurrence within 3 months after diagnosis, or both, or if the duration of follow-up was less than 12 months at the time of data collection.
We asked the principal investigators of the nine studies to submit the original data collected for each patient.3,8,9,10,11,12,13,15 Owing to differences in criteria for inclusion and exclusion, the numbers of patients in the original publications may differ from those reported here. Survival results differ if the follow-up information was updated after publication.
We asked the collaborating groups to code data that had been rendered anonymous in a standardized fashion for inclusion in a database. In a signed letter, all principal investigators stated that local institutional review boards had agreed to the bone marrow–aspiration procedure and the study and that all patients whose data were submitted had agreed to bone marrow aspiration and statistical analysis in accordance with international regulations regarding data safety.
Bone Marrow Aspiration and Immunocytochemistry
The criteria for designating a case positive for the presence of micrometastasis and the details of the immunocytochemical assays used by the contributing groups have been described in detail elsewhere.3,8,9,10,11,12,13,14,15,16
Statistical Analysis
We tested for associations between the presence of bone marrow micrometastasis and the baseline characteristics of patients, as well as established prognostic factors, using the chi-square test. Categorical variables with more than two categories were analyzed for trend.
Hazard ratios and 95 percent confidence intervals for recurrence or death with micrometastasis as the sole variable were calculated for each of the nine studies by means of meta-analysis (with the use of the random-effects model based on individual patient data). The Q-test was performed to assess interstudy heterogeneity.17 For the sensitivity analysis, meta-analytic hazard ratios and confidence intervals were computed with the omission of one study at a time.
For the survival analysis, we considered in separate analyses the following primary end points: death due to any cause; death due to causes related to breast cancer (i.e., metastasis-dependent organ failure or progression of breast cancer); distant or local disease recurrence, or both; and distant metastasis. Survival intervals were measured from the time of surgery and bone marrow aspiration to the time of death or of the first clinical or radiographic evidence of disease recurrence. Incidence rates and mortality were calculated as the number of disease recurrences or deaths per 1000 person-years; mortality ratios, incidence-rate ratios, and 95 percent confidence intervals were estimated.
For patients surviving 10 years or more (412 patients), the follow-up data were censored after 120 months. Data for women in whom the envisaged end point was not reached were censored as of the last follow-up. We constructed Kaplan – Meier curves18 and used the log-rank test19 to determine the univariate significance of the study variables.
We used a Cox proportional-hazards regression model to examine simultaneously the effects of multiple covariates on survival.20 In all models, the categorical variables were tested for trend and the proportional-hazards assumption was assessed. If separate categories did not improve the fit of the model, a linear trend was preferred. A test for interaction between pairs of variables in the final models was performed. The effect of each variable in these models was assessed with the use of the Wald test and described by the hazard ratio, with a 95 percent confidence interval. All estimates were stratified according to study center, and all reported P values are two-sided.
The initial model included age at diagnosis, menopausal status, tumor size and grade, and information on lymph-node metastases as well as hormone-receptor expression. Since progesterone-receptor expression was not routinely assessed in all participating centers, a binary variable was created to indicate that at least one hormone receptor was positive. Subjects with missing values for this hormone-receptor variable or for tumor grade were excluded from modeling. The final model was developed by dropping each variable in turn from the model and conducting a likelihood-ratio test to compare the full and the nested models. We used a significance level of 0.05 as the cutoff to exclude a variable from the model. Finally, the variable of bone marrow micrometastasis (present vs. absent) was added to the model in order to test the resultant model against that without the variable.
On the basis of the observation that curves on the Kaplan – Meier graphs dispersed during the first years of follow-up and then showed less divergence, the assumption of proportional hazards for the final model was not met over the entire follow-up period. We therefore opted for a piecewise Cox model,21 with a cutoff point set at five years for overall survival and breast-cancer–specific survival and at four years for disease-free survival and distant-disease–free survival. We fit separate Cox models for both the first and second intervals. The proportional-hazards assumption was formally tested for each interval,22 and separate regression estimates are given.
Results
Prevalence of Bone Marrow Micrometastasis
Table 1 gives a summary of the original studies, the patients enrolled in them, and the technical variables used in the studies. A total of 4703 patients with invasive breast cancer were included in our analysis. Bone marrow micrometastasis was present in 1438 patients (30.6 percent). As compared with women without bone marrow micrometastasis, patients with bone marrow micrometastasis had larger tumors, tumors with a higher histologic grade, more frequent lymph-node metastasis, and more hormone-receptor–negative tumors (Table 2).
Table 1. Baseline Characteristics of the Patients and Study Variables, According to Study Center (City) and Technical Variables.
Table 2. Prevalence of Bone Marrow Micrometastasis According to Clinical Variables.
Sensitivity Analysis
In the meta-analysis, using a random-effects model, we found a hazard ratio of 2.26 (95 percent confidence interval, 1.72 to 2.97; P<0.001) for death from any cause and for any disease recurrence associated with the presence of micrometastasis. For these survival end points, the hazard ratios calculated in eight studies ranged from 1.36 to 4.04 and from 1.23 to 3.16, respectively; in the ninth study (Paris), the hazard ratios were 8.58 and 8.23, respectively. For each outcome, the 95 percent confidence intervals were significant in all but two studies (Munich and New York) and showed considerable overlap, indicating a similar effect of micrometastasis on outcome in all nine studies. The Q-test for statistical heterogeneity showed significant interstudy variation among the estimated hazard ratios (P=0.007 for death from any cause; P<0.001 for disease recurrence), which was further investigated by sensitivity analysis. The exclusion of any one study did not markedly change the estimates of the hazard ratios or confidence intervals found in the meta-analysis (for details, see the Supplementary Appendix, available with the full text of this article at www.nejm.org). However, we found that the large Heidelberg study (hazard ratio, 4.04) had the most influence on the outcome of death from any cause. The omission of this study resulted in a marginally lower but still significant hazard ratio (2.02; 95 percent confidence interval, 1.62 to 2.77; P=0.18 according to the Q-test) for death from any cause.
Survival
In the pooled data, the median follow-up time among survivors was 62 months. Of 889 patients who died during follow-up, 667 (75.0 percent) died from breast cancer and 222 (25.0 percent) from other causes; 76.9 percent of all deaths occurred during the first five years. Both the overall rate of death and the rate of death from breast cancer among patients with micrometastasis were significantly higher than the rate of death among patients without micrometastasis in bone marrow (Figure 1A and Figure 1B). The presence of micrometastasis remained a significant prognostic factor with respect to survival when we controlled for tumor size, grade, lymph-node metastasis, and hormone-receptor expression in the multivariable analysis. In the piecewise multivariable analysis, hazard ratios for death from any cause and death from breast cancer among patients with micrometastasis, as compared with those among patients without micrometastasis, were significantly increased during the first five years of follow-up and thereafter (Table 3).
Figure 1. Kaplan–Meier Estimates of Long-Term Survival and Outcome in the Complete Patient Group According to the Presence or Absence of Bone Marrow Micrometastasis.
Dotted lines indicate the cutoff point at five or four years used for piecewise Cox regression modeling. MR denotes mortality ratio (the ratio of the mortality rate among women with micrometastasis as compared with that among those without micrometastasis), IRR incidence-rate ratio (the ratio of the incidence of recurrence or death among women with micrometastasis as compared with that among those without micrometastasis), and CI confidence interval. P values were calculated by the log-rank test.
Table 3. Multivariable Hazard Ratios for Death from Any Cause, Death from Breast Cancer, Disease Recurrence, and Distant Metastasis at Different Follow-up Intervals (Adjusted for the Study Center).
Recurrence of Disease
During the follow-up period, breast cancer recurred in 1192 patients (25.3 percent). Of these, 969 patients (81.3 percent) had a recurrence only in the form of distant disease, whereas 447 patients (37.5 percent) had a local relapse (in the breast or the chest wall) or a recurrence in regional lymph nodes (alone or in combination with distant metastases); 80.9 percent of all recurrences occurred within the first four years. Both the disease-free interval and the distant-disease–free interval (Figure 1C and Figure 1D) were significantly shorter among patients with micrometastasis (P<0.001 for all comparisons, by the log-rank test); for these two end points, piecewise multivariable Cox regression modeling showed that the presence of micrometastasis was a significant predictor of recurrence only during the first four years of follow-up (Table 3).
Subgroup Analyses
We analyzed subgroups of patients who had received endocrine treatment alone or chemotherapy alone and patients considered to be at low risk who had tumors no larger than 2 cm (pT1N0) and no lymph-node metastasis who did not receive systemic adjuvant therapy. Patients in the endocrine-therapy and chemotherapy subgroups had significantly poorer outcomes for all investigated end points if micrometastasis was present, as compared with patients in these subgroups in whom micrometastasis was absent (Figure 2). Remarkably, among 1036 patients in the low-risk subgroup, the presence of micrometastasis was associated with an increase by a factor of 3.65 (95 percent confidence interval, 1.94 to 6.89; P<0.001) in mortality from breast cancer and a factor of 2.00 (95 percent confidence interval, 1.20 to 3.35; P=0.007) in the risk of distant metastasis during the first five years, as compared with patients in whom micrometastasis in the bone marrow was absent (Figure 2).
Figure 2. Kaplan–Meier Estimates of Breast-Cancer–Specific and Distant-Disease–free Survival among Predefined Patient Subgroups According to the Presence or Absence of Bone Marrow Micrometastasis.
Dotted lines indicate the cutoff point at five or four years used for piecewise Cox regression modeling. MR denotes mortality ratio (the ratio of the mortality rate among women with micrometastasis as compared with that among those without micrometastasis), IRR incidence-rate ratio (the ratio of the incidence of recurrence or death among women with micrometastasis as compared with that among those without micrometastasis), and CI confidence interval. P values were calculated by the log-rank test.
Discussion
This pooled analysis of data on 4703 patients with breast cancer who were enrolled in nine clinical studies found strong evidence of the independent, adverse prognostic significance of the presence of bone marrow micrometastasis at the time of the initial diagnosis of operable breast cancer. Interstudy heterogeneity was influenced by a single large study, but it introduced no significant bias with respect to overall survival or disease-free survival. Further sources of heterogeneity were differences in patients' characteristics and in the immunoassays used to detect micrometastasis. Stratification according to center and the inclusion of patients' characteristics in the regression models took these sources of heterogeneity into account. Variability in treatment over time was overcome by conducting a pooled analysis of data on individual patients. The use of these data allowed us to standardize inclusion and exclusion criteria and to update the numbers of patients and follow-up information after the appearance of the original published reports. Others have suggested that the ideal way to perform a meta-analysis of survival data is to use individual patient data.23,24
In the multivariable analysis, the presence of micrometastasis was associated with the highest estimates of relative risk for each end point during the first follow-up interval of five years (for death from any cause and death from breast cancer) and four years (disease recurrence and distant metastasis) (Table 3). A plausible explanation for the failure to demonstrate a significant association between micrometastasis and recurrence or distant metastasis during the second interval (i.e., years 5 to 10 of follow-up) is that the presence of micrometastasis is associated with the recurrence of breast cancer before the second interval of follow-up, thereby selecting out patients at risk for recurrence during the second interval.
Not all bone marrow cells that stain with an anticytokeratin antibody or with antibodies against polymorphic epithelial mucins (the technical definition of micrometastatic cells) can be unequivocally or uncritically defined as malignant.6,7 Convincing molecular data, however, point to numerous signs of malignancy in cytokeratin-positive cells.5,25,26,27
We did not identify a subgroup of patients in whom micrometastasis appeared to be prognostically irrelevant. The data presented here may therefore help in planning clinical trials aimed at determining whether the presence or absence of micrometastasis suffices for a decision on the need for therapy and to predict the outcome of treatment in certain subgroups. In our study, the group of 807 patients with tumors no larger than 2 cm in diameter and without lymph-node metastasis, who had no detectable micrometastasis and who did not receive systemic adjuvant treatment, had a 94 percent five-year survival (Figure 2) and might be considered cured. Treatment stratification based on the presence or absence of micrometastasis may therefore be useful in trials of systemic adjuvant therapy in patients with pT1N0 tumors and bone marrow micrometastasis.
In summary, our data support the prognostic value of the presence of bone marrow micrometastasis and could be useful in the design of trials of the adjuvant treatment of breast cancer.
Drs. Braun and Pantel report having received lecture fees from Veridex; Drs. Pantel and Schlimok report having equity ownership in Micromet; and Dr. Pantel reports having received consulting fees from Clarient.
We are indebted to the 4703 women who gave informed consent for bone marrow aspiration to make possible this study and the previous studies without benefit to themselves but to support scientific progress.
* Additional investigators who contributed to this study are listed in the Appendix.
Source Information
From the Department of Obstetrics and Gynecology, Innsbruck Medical University, Innsbruck, Austria (S.B., C.M.); Department of Obstetrics and Gynecology, General Hospital, Merano, Italy (F.D.V.); Department of Oncology, Norwegian Radium Hospital, Oslo (B.N.); Department of Obstetrics and Gynecology, Ludwig-Maximilians University, Munich, Germany (W.J., B.S.); Department of Surgery, New York Presbyterian Hospital, Cornell University, New York (M.P.O.); Division of Medicine, Imperial College, London (R.C.C.); Department of Hematology and Oncology, Central Hospital, Augsburg, Germany (G.S., D.O.); Department of Obstetrics and Gynecology, University Hospital, Heidelberg, Germany (I.J.D., E.-F.S.); Department of Obstetrics and Gynecology, University Hospital, Rostock, Germany (B.G., G.K.); Department of Obstetrics and Gynecology, Nuremberg-Erlangen University Hospital, Erlangen, Germany (G.G., T.F.); Department of Hematology and Oncology, Institut Curie, Paris, (J.-Y.P., A.V.-S.); Department of Surgery, Ullev?l University Hospital, Oslo (G.W.); Strang Cancer Prevention Center, Cornell Medical Center, New York (G.Y.C.W.); Institute of Cancer Research, Sutton, United Kingdom (J.B.); and Institute of Tumor Biology, Eppendorf University, Hamburg, Germany (K.P.).
Drs. Braun and Vogl contributed equally to this manuscript.
Address reprint requests to Dr. Braun at the Department of Obstetrics and Gynecology, Innsbruck Medical University, Anichstr. 35, A-6020 Innsbruck, Austria, or at stephan.braun@uklibk.ac.at.
References
Fisher B, Fisher ER, Guzman C, Copeland CE, Caceres E. The dissemination of subcutaneously inoculated tumor cell suspensions. Arch Surg 1969;98:347-351.
Braun S, Cevatli BS, Assemi C, et al. Comparative analysis of micrometastasis to the bone marrow and lymph nodes of node-negative breast cancer patients receiving no adjuvant therapy. J Clin Oncol 2001;19:1468-1475.
Gerber B, Krause A, Muller H, et al. Simultaneous immunohistochemical detection of tumor cells in lymph nodes and bone marrow aspirates in breast cancer and its correlation with other prognostic factors. J Clin Oncol 2001;19:960-971.
Woelfle U, Cloos J, Sauter G, et al. Molecular signature associated with bone marrow micrometastasis in human breast cancer. Cancer Res 2003;63:5679-5684.
Klein CA, Blankenstein TJF, Schmidt-Kittler O, et al. Genetic heterogeneity of single disseminated tumour cells in minimal residual cancer. Lancet 2002;360:683-689.
Braun S, Naume B. Circulating and disseminated tumor cells. J Clin Oncol 2005;23:1623-1626.
Pantel K, Brakenhoff RH. Dissecting the metastatic cascade. Nat Rev Cancer 2004;4:448-456.
Pierga J-Y, Bonneton C, Vincent-Salomon A, et al. Clinical significance of immunocytochemical detection of tumor cells using digital microscopy in peripheral blood and bone marrow of breast cancer patients. Clin Cancer Res 2004;10:1392-1400.
Wiedswang G, Borgen E, Karesen R, et al. Detection of isolated tumor cells in bone marrow is an independent prognostic factor in breast cancer. J Clin Oncol 2003;21:3469-3478.
Gebauer G, Fehm T, Merkle E, Beck EP, Lang N, Jager W. Epithelial cells in bone marrow of breast cancer patients at time of primary surgery: clinical outcome during long-term follow-up. J Clin Oncol 2001;19:3669-3674.
Braun S, Pantel K, Müller P, et al. Cytokeratin-positive cells in the bone marrow and survival of patients with stage I, II, or III breast cancer. N Engl J Med 2000;342:525-533.
Mansi JL, Gogas H, Bliss JM, Gazet JC, Berger U, Coombes RC. Outcome of primary-breast-cancer patients with micrometastases: a long-term follow-up study. Lancet 1999;354:197-202.
Diel IJ, Kaufmann M, Costa SD, et al. Micrometastatic breast cancer cells in bone marrow at primary surgery: prognostic value in comparison with nodal status. J Natl Cancer Inst 1996;88:1652-1658.
Cote RJ, Rosen PP, Lesser ML, Old LJ, Osborne MP. Prediction of early relapse in patients with operable breast cancer by detection of occult bone marrow micrometastases. J Clin Oncol 1991;9:1749-1756.
Wong GYC, Yu QQ, Osborne MP. Bone marrow micrometastasis is a significant predictor of long-term relapse-free survival for breast cancer by a non-proportional hazards model. Breast Cancer Res Treat 2003;82:Suppl 1:S99-S99. abstract.
Osborne M, Wong G, Asina S, Old LJ, Cote RJ, Rosen PP. Sensitivity of immunocytochemical detection of breast cancer cells in human bone marrow. Cancer Res 1991;51:2706-2709.
DerSimonian R. Meta-analysis in the design and monitoring of clinical trials. Stat Med 1996;15:1237-1248.
Kaplan EL, Meier P. Nonparametric estimation from incomplete observations. J Am Stat Assoc 1958;53:457-481.
Peto R, Peto J. Asymptotically efficient rank invariant test procedures. J R Stat Soc 1972;135:185-206.
Cox DR. Regression models and life-tables. J R Stat Soc 1972;34:187-220.
Collett D. Modelling survival data in medical research. 2nd ed. Boca Raton, Fla.: Chapman & Hall/CRC, 2003.
Grambsch P, Therneau TM. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika 1994;81:515-526.
Hunink MG, Wong JB. Meta-analysis of failure-time data with adjustment for covariates. Med Decis Making 1994;14:59-70.
Clarke MJ, Stewart LA. Obtaining data from randomised controlled trials: how much do we need for reliable and informative meta-analyses? BMJ 1994;309:1007-1010.
Gangnus R, Langer S, Breit E, Pantel K, Speicher MR. Genomic profiling of viable and proliferative micrometastatic cells from early-stage breast cancer patients. Clin Cancer Res 2004;10:3457-3464.
Schmidt-Kittler O, Ragg T, Daskalakis A, et al. From latent disseminated cells to overt metastasis: genetic analysis of systemic breast cancer progression. Proc Natl Acad Sci U S A 2003;100:7737-7742.
Fehm T, Sagalowsky A, Clifford E, et al. Cytogenetic evidence that circulating epithelial cells in patients with carcinoma are malignant. Clin Cancer Res 2002;8:2073-2084.(Stephan Braun, M.D., Flor)
Background We assessed the prognostic significance of the presence of micrometastasis in the bone marrow at the time of diagnosis of breast cancer by means of a pooled analysis.
Methods We combined individual patient data from nine studies involving 4703 patients with stage I, II, or III breast cancer. We evaluated patient outcomes over a 10-year follow-up period (median, 5.2 years), using a multivariable piecewise Cox regression model.
Results Micrometastasis was detected in 30.6 percent of the patients. As compared with women without bone marrow micrometastasis, patients with bone marrow micrometastasis had larger tumors and tumors with a higher histologic grade and more often had lymph-node metastases and hormone receptor-negative tumors (P<0.001 for all variables). The presence of micrometastasis was a significant prognostic factor with respect to poor overall survival and breast-cancer–specific survival (univariate mortality ratios, 2.15 and 2.44, respectively; P<0.001 for both outcomes) and poor disease-free survival and distant-disease–free survival during the 10-year observation period (incidence-rate ratios, 2.13 and 2.33, respectively; P<0.001 for both outcomes). In the multivariable analysis, micrometastasis was an independent predictor of a poor outcome. In the univariate subgroup analysis, breast-cancer–specific survival among patients with micrometastasis was significantly shortened (P<0.001 for all comparisons) among those receiving adjuvant endocrine treatment (mortality ratio, 3.22) or cytotoxic therapy (mortality ratio, 2.32) and among patients who had tumors no larger than 2 cm in diameter without lymph-node metastasis and who did not receive systemic adjuvant therapy (mortality ratio, 3.65).
Conclusions The presence of micrometastasis in the bone marrow at the time of diagnosis of breast cancer is associated with a poor prognosis.
Data from experiments in animals1 performed in the 1960s and from more recent immunocytochemical2,3 and molecular4,5 studies suggest that lymph-node involvement does not accurately predict hematogenous dissemination of cancer cells, nor is hematogenous dissemination necessarily associated with lymph-node involvement.6,7 During the past two decades, several studies have assessed the prevalence and prognostic value of hematogenous dissemination of tumor cells in patients with node-positive and node-negative breast cancer.3,8,9,10,11,12,13,14,15 The influence of the presence of micrometastasis in the bone marrow on prognosis has been shown in patients with identical stages of breast cancer, as defined by tumor size, histologic grade, presence or absence of lymph-node metastasis, and expression of hormone receptors.3,9,10,11,12,13 However, the clinical usefulness of finding such micrometastasis is limited by the low statistical power of published studies and the lack of clinical trials specifically investigating the predictive role of bone marrow micrometastasis. To date, only two small studies have reported the outcome of patients with bone marrow micrometastasis10,12 well beyond a median observation time of five years. In this study, we investigated the long-term outcome of patients with and those without bone marrow micrometastasis. We also explored the effect of bone marrow micrometastasis on prognosis in clinically relevant subgroups. To accomplish these goals, we analyzed pooled data from nine independent studies with updated follow-up data and numbers of patients; these studies involved 4703 patients with stage I, II, or III breast cancer who were treated in Augsburg and Munich, Germany (two independent studies that were initially published together),11 Paris,8 Oslo,9 Rostock, Germany,3 New York,15 Erlangen, Germany,10 Heidelberg, Germany,13 and London.12 (These studies are referred to hereinafter by the names of the cities.)
Methods
Data Collection
The National Library of Medicine of the National Center for Biotechnology Information (www.ncbi.nlm.nih.gov) was searched for studies related to bone marrow micrometastasis and survival of patients with breast cancer. Six such studies were identified.3,10,11,12,13 Furthermore, on the basis of personal contact, we knew of three studies that were in the process of manuscript preparation15 or submission8,9 at the time we were collecting our data. We defined eligible patients as women with complete baseline clinical records, data on follow-up examinations, and histologically confirmed and completely removed primary stage I, II, or III breast cancer, with information on tumor size and the presence or absence of axillary lymph-node metastasis. We further required documented validation of the immunoassays used to detect micrometastasis. Patients were excluded from the analysis if they had in situ carcinomas only, if they had either distant metastases or local recurrence within 3 months after diagnosis, or both, or if the duration of follow-up was less than 12 months at the time of data collection.
We asked the principal investigators of the nine studies to submit the original data collected for each patient.3,8,9,10,11,12,13,15 Owing to differences in criteria for inclusion and exclusion, the numbers of patients in the original publications may differ from those reported here. Survival results differ if the follow-up information was updated after publication.
We asked the collaborating groups to code data that had been rendered anonymous in a standardized fashion for inclusion in a database. In a signed letter, all principal investigators stated that local institutional review boards had agreed to the bone marrow–aspiration procedure and the study and that all patients whose data were submitted had agreed to bone marrow aspiration and statistical analysis in accordance with international regulations regarding data safety.
Bone Marrow Aspiration and Immunocytochemistry
The criteria for designating a case positive for the presence of micrometastasis and the details of the immunocytochemical assays used by the contributing groups have been described in detail elsewhere.3,8,9,10,11,12,13,14,15,16
Statistical Analysis
We tested for associations between the presence of bone marrow micrometastasis and the baseline characteristics of patients, as well as established prognostic factors, using the chi-square test. Categorical variables with more than two categories were analyzed for trend.
Hazard ratios and 95 percent confidence intervals for recurrence or death with micrometastasis as the sole variable were calculated for each of the nine studies by means of meta-analysis (with the use of the random-effects model based on individual patient data). The Q-test was performed to assess interstudy heterogeneity.17 For the sensitivity analysis, meta-analytic hazard ratios and confidence intervals were computed with the omission of one study at a time.
For the survival analysis, we considered in separate analyses the following primary end points: death due to any cause; death due to causes related to breast cancer (i.e., metastasis-dependent organ failure or progression of breast cancer); distant or local disease recurrence, or both; and distant metastasis. Survival intervals were measured from the time of surgery and bone marrow aspiration to the time of death or of the first clinical or radiographic evidence of disease recurrence. Incidence rates and mortality were calculated as the number of disease recurrences or deaths per 1000 person-years; mortality ratios, incidence-rate ratios, and 95 percent confidence intervals were estimated.
For patients surviving 10 years or more (412 patients), the follow-up data were censored after 120 months. Data for women in whom the envisaged end point was not reached were censored as of the last follow-up. We constructed Kaplan – Meier curves18 and used the log-rank test19 to determine the univariate significance of the study variables.
We used a Cox proportional-hazards regression model to examine simultaneously the effects of multiple covariates on survival.20 In all models, the categorical variables were tested for trend and the proportional-hazards assumption was assessed. If separate categories did not improve the fit of the model, a linear trend was preferred. A test for interaction between pairs of variables in the final models was performed. The effect of each variable in these models was assessed with the use of the Wald test and described by the hazard ratio, with a 95 percent confidence interval. All estimates were stratified according to study center, and all reported P values are two-sided.
The initial model included age at diagnosis, menopausal status, tumor size and grade, and information on lymph-node metastases as well as hormone-receptor expression. Since progesterone-receptor expression was not routinely assessed in all participating centers, a binary variable was created to indicate that at least one hormone receptor was positive. Subjects with missing values for this hormone-receptor variable or for tumor grade were excluded from modeling. The final model was developed by dropping each variable in turn from the model and conducting a likelihood-ratio test to compare the full and the nested models. We used a significance level of 0.05 as the cutoff to exclude a variable from the model. Finally, the variable of bone marrow micrometastasis (present vs. absent) was added to the model in order to test the resultant model against that without the variable.
On the basis of the observation that curves on the Kaplan – Meier graphs dispersed during the first years of follow-up and then showed less divergence, the assumption of proportional hazards for the final model was not met over the entire follow-up period. We therefore opted for a piecewise Cox model,21 with a cutoff point set at five years for overall survival and breast-cancer–specific survival and at four years for disease-free survival and distant-disease–free survival. We fit separate Cox models for both the first and second intervals. The proportional-hazards assumption was formally tested for each interval,22 and separate regression estimates are given.
Results
Prevalence of Bone Marrow Micrometastasis
Table 1 gives a summary of the original studies, the patients enrolled in them, and the technical variables used in the studies. A total of 4703 patients with invasive breast cancer were included in our analysis. Bone marrow micrometastasis was present in 1438 patients (30.6 percent). As compared with women without bone marrow micrometastasis, patients with bone marrow micrometastasis had larger tumors, tumors with a higher histologic grade, more frequent lymph-node metastasis, and more hormone-receptor–negative tumors (Table 2).
Table 1. Baseline Characteristics of the Patients and Study Variables, According to Study Center (City) and Technical Variables.
Table 2. Prevalence of Bone Marrow Micrometastasis According to Clinical Variables.
Sensitivity Analysis
In the meta-analysis, using a random-effects model, we found a hazard ratio of 2.26 (95 percent confidence interval, 1.72 to 2.97; P<0.001) for death from any cause and for any disease recurrence associated with the presence of micrometastasis. For these survival end points, the hazard ratios calculated in eight studies ranged from 1.36 to 4.04 and from 1.23 to 3.16, respectively; in the ninth study (Paris), the hazard ratios were 8.58 and 8.23, respectively. For each outcome, the 95 percent confidence intervals were significant in all but two studies (Munich and New York) and showed considerable overlap, indicating a similar effect of micrometastasis on outcome in all nine studies. The Q-test for statistical heterogeneity showed significant interstudy variation among the estimated hazard ratios (P=0.007 for death from any cause; P<0.001 for disease recurrence), which was further investigated by sensitivity analysis. The exclusion of any one study did not markedly change the estimates of the hazard ratios or confidence intervals found in the meta-analysis (for details, see the Supplementary Appendix, available with the full text of this article at www.nejm.org). However, we found that the large Heidelberg study (hazard ratio, 4.04) had the most influence on the outcome of death from any cause. The omission of this study resulted in a marginally lower but still significant hazard ratio (2.02; 95 percent confidence interval, 1.62 to 2.77; P=0.18 according to the Q-test) for death from any cause.
Survival
In the pooled data, the median follow-up time among survivors was 62 months. Of 889 patients who died during follow-up, 667 (75.0 percent) died from breast cancer and 222 (25.0 percent) from other causes; 76.9 percent of all deaths occurred during the first five years. Both the overall rate of death and the rate of death from breast cancer among patients with micrometastasis were significantly higher than the rate of death among patients without micrometastasis in bone marrow (Figure 1A and Figure 1B). The presence of micrometastasis remained a significant prognostic factor with respect to survival when we controlled for tumor size, grade, lymph-node metastasis, and hormone-receptor expression in the multivariable analysis. In the piecewise multivariable analysis, hazard ratios for death from any cause and death from breast cancer among patients with micrometastasis, as compared with those among patients without micrometastasis, were significantly increased during the first five years of follow-up and thereafter (Table 3).
Figure 1. Kaplan–Meier Estimates of Long-Term Survival and Outcome in the Complete Patient Group According to the Presence or Absence of Bone Marrow Micrometastasis.
Dotted lines indicate the cutoff point at five or four years used for piecewise Cox regression modeling. MR denotes mortality ratio (the ratio of the mortality rate among women with micrometastasis as compared with that among those without micrometastasis), IRR incidence-rate ratio (the ratio of the incidence of recurrence or death among women with micrometastasis as compared with that among those without micrometastasis), and CI confidence interval. P values were calculated by the log-rank test.
Table 3. Multivariable Hazard Ratios for Death from Any Cause, Death from Breast Cancer, Disease Recurrence, and Distant Metastasis at Different Follow-up Intervals (Adjusted for the Study Center).
Recurrence of Disease
During the follow-up period, breast cancer recurred in 1192 patients (25.3 percent). Of these, 969 patients (81.3 percent) had a recurrence only in the form of distant disease, whereas 447 patients (37.5 percent) had a local relapse (in the breast or the chest wall) or a recurrence in regional lymph nodes (alone or in combination with distant metastases); 80.9 percent of all recurrences occurred within the first four years. Both the disease-free interval and the distant-disease–free interval (Figure 1C and Figure 1D) were significantly shorter among patients with micrometastasis (P<0.001 for all comparisons, by the log-rank test); for these two end points, piecewise multivariable Cox regression modeling showed that the presence of micrometastasis was a significant predictor of recurrence only during the first four years of follow-up (Table 3).
Subgroup Analyses
We analyzed subgroups of patients who had received endocrine treatment alone or chemotherapy alone and patients considered to be at low risk who had tumors no larger than 2 cm (pT1N0) and no lymph-node metastasis who did not receive systemic adjuvant therapy. Patients in the endocrine-therapy and chemotherapy subgroups had significantly poorer outcomes for all investigated end points if micrometastasis was present, as compared with patients in these subgroups in whom micrometastasis was absent (Figure 2). Remarkably, among 1036 patients in the low-risk subgroup, the presence of micrometastasis was associated with an increase by a factor of 3.65 (95 percent confidence interval, 1.94 to 6.89; P<0.001) in mortality from breast cancer and a factor of 2.00 (95 percent confidence interval, 1.20 to 3.35; P=0.007) in the risk of distant metastasis during the first five years, as compared with patients in whom micrometastasis in the bone marrow was absent (Figure 2).
Figure 2. Kaplan–Meier Estimates of Breast-Cancer–Specific and Distant-Disease–free Survival among Predefined Patient Subgroups According to the Presence or Absence of Bone Marrow Micrometastasis.
Dotted lines indicate the cutoff point at five or four years used for piecewise Cox regression modeling. MR denotes mortality ratio (the ratio of the mortality rate among women with micrometastasis as compared with that among those without micrometastasis), IRR incidence-rate ratio (the ratio of the incidence of recurrence or death among women with micrometastasis as compared with that among those without micrometastasis), and CI confidence interval. P values were calculated by the log-rank test.
Discussion
This pooled analysis of data on 4703 patients with breast cancer who were enrolled in nine clinical studies found strong evidence of the independent, adverse prognostic significance of the presence of bone marrow micrometastasis at the time of the initial diagnosis of operable breast cancer. Interstudy heterogeneity was influenced by a single large study, but it introduced no significant bias with respect to overall survival or disease-free survival. Further sources of heterogeneity were differences in patients' characteristics and in the immunoassays used to detect micrometastasis. Stratification according to center and the inclusion of patients' characteristics in the regression models took these sources of heterogeneity into account. Variability in treatment over time was overcome by conducting a pooled analysis of data on individual patients. The use of these data allowed us to standardize inclusion and exclusion criteria and to update the numbers of patients and follow-up information after the appearance of the original published reports. Others have suggested that the ideal way to perform a meta-analysis of survival data is to use individual patient data.23,24
In the multivariable analysis, the presence of micrometastasis was associated with the highest estimates of relative risk for each end point during the first follow-up interval of five years (for death from any cause and death from breast cancer) and four years (disease recurrence and distant metastasis) (Table 3). A plausible explanation for the failure to demonstrate a significant association between micrometastasis and recurrence or distant metastasis during the second interval (i.e., years 5 to 10 of follow-up) is that the presence of micrometastasis is associated with the recurrence of breast cancer before the second interval of follow-up, thereby selecting out patients at risk for recurrence during the second interval.
Not all bone marrow cells that stain with an anticytokeratin antibody or with antibodies against polymorphic epithelial mucins (the technical definition of micrometastatic cells) can be unequivocally or uncritically defined as malignant.6,7 Convincing molecular data, however, point to numerous signs of malignancy in cytokeratin-positive cells.5,25,26,27
We did not identify a subgroup of patients in whom micrometastasis appeared to be prognostically irrelevant. The data presented here may therefore help in planning clinical trials aimed at determining whether the presence or absence of micrometastasis suffices for a decision on the need for therapy and to predict the outcome of treatment in certain subgroups. In our study, the group of 807 patients with tumors no larger than 2 cm in diameter and without lymph-node metastasis, who had no detectable micrometastasis and who did not receive systemic adjuvant treatment, had a 94 percent five-year survival (Figure 2) and might be considered cured. Treatment stratification based on the presence or absence of micrometastasis may therefore be useful in trials of systemic adjuvant therapy in patients with pT1N0 tumors and bone marrow micrometastasis.
In summary, our data support the prognostic value of the presence of bone marrow micrometastasis and could be useful in the design of trials of the adjuvant treatment of breast cancer.
Drs. Braun and Pantel report having received lecture fees from Veridex; Drs. Pantel and Schlimok report having equity ownership in Micromet; and Dr. Pantel reports having received consulting fees from Clarient.
We are indebted to the 4703 women who gave informed consent for bone marrow aspiration to make possible this study and the previous studies without benefit to themselves but to support scientific progress.
* Additional investigators who contributed to this study are listed in the Appendix.
Source Information
From the Department of Obstetrics and Gynecology, Innsbruck Medical University, Innsbruck, Austria (S.B., C.M.); Department of Obstetrics and Gynecology, General Hospital, Merano, Italy (F.D.V.); Department of Oncology, Norwegian Radium Hospital, Oslo (B.N.); Department of Obstetrics and Gynecology, Ludwig-Maximilians University, Munich, Germany (W.J., B.S.); Department of Surgery, New York Presbyterian Hospital, Cornell University, New York (M.P.O.); Division of Medicine, Imperial College, London (R.C.C.); Department of Hematology and Oncology, Central Hospital, Augsburg, Germany (G.S., D.O.); Department of Obstetrics and Gynecology, University Hospital, Heidelberg, Germany (I.J.D., E.-F.S.); Department of Obstetrics and Gynecology, University Hospital, Rostock, Germany (B.G., G.K.); Department of Obstetrics and Gynecology, Nuremberg-Erlangen University Hospital, Erlangen, Germany (G.G., T.F.); Department of Hematology and Oncology, Institut Curie, Paris, (J.-Y.P., A.V.-S.); Department of Surgery, Ullev?l University Hospital, Oslo (G.W.); Strang Cancer Prevention Center, Cornell Medical Center, New York (G.Y.C.W.); Institute of Cancer Research, Sutton, United Kingdom (J.B.); and Institute of Tumor Biology, Eppendorf University, Hamburg, Germany (K.P.).
Drs. Braun and Vogl contributed equally to this manuscript.
Address reprint requests to Dr. Braun at the Department of Obstetrics and Gynecology, Innsbruck Medical University, Anichstr. 35, A-6020 Innsbruck, Austria, or at stephan.braun@uklibk.ac.at.
References
Fisher B, Fisher ER, Guzman C, Copeland CE, Caceres E. The dissemination of subcutaneously inoculated tumor cell suspensions. Arch Surg 1969;98:347-351.
Braun S, Cevatli BS, Assemi C, et al. Comparative analysis of micrometastasis to the bone marrow and lymph nodes of node-negative breast cancer patients receiving no adjuvant therapy. J Clin Oncol 2001;19:1468-1475.
Gerber B, Krause A, Muller H, et al. Simultaneous immunohistochemical detection of tumor cells in lymph nodes and bone marrow aspirates in breast cancer and its correlation with other prognostic factors. J Clin Oncol 2001;19:960-971.
Woelfle U, Cloos J, Sauter G, et al. Molecular signature associated with bone marrow micrometastasis in human breast cancer. Cancer Res 2003;63:5679-5684.
Klein CA, Blankenstein TJF, Schmidt-Kittler O, et al. Genetic heterogeneity of single disseminated tumour cells in minimal residual cancer. Lancet 2002;360:683-689.
Braun S, Naume B. Circulating and disseminated tumor cells. J Clin Oncol 2005;23:1623-1626.
Pantel K, Brakenhoff RH. Dissecting the metastatic cascade. Nat Rev Cancer 2004;4:448-456.
Pierga J-Y, Bonneton C, Vincent-Salomon A, et al. Clinical significance of immunocytochemical detection of tumor cells using digital microscopy in peripheral blood and bone marrow of breast cancer patients. Clin Cancer Res 2004;10:1392-1400.
Wiedswang G, Borgen E, Karesen R, et al. Detection of isolated tumor cells in bone marrow is an independent prognostic factor in breast cancer. J Clin Oncol 2003;21:3469-3478.
Gebauer G, Fehm T, Merkle E, Beck EP, Lang N, Jager W. Epithelial cells in bone marrow of breast cancer patients at time of primary surgery: clinical outcome during long-term follow-up. J Clin Oncol 2001;19:3669-3674.
Braun S, Pantel K, Müller P, et al. Cytokeratin-positive cells in the bone marrow and survival of patients with stage I, II, or III breast cancer. N Engl J Med 2000;342:525-533.
Mansi JL, Gogas H, Bliss JM, Gazet JC, Berger U, Coombes RC. Outcome of primary-breast-cancer patients with micrometastases: a long-term follow-up study. Lancet 1999;354:197-202.
Diel IJ, Kaufmann M, Costa SD, et al. Micrometastatic breast cancer cells in bone marrow at primary surgery: prognostic value in comparison with nodal status. J Natl Cancer Inst 1996;88:1652-1658.
Cote RJ, Rosen PP, Lesser ML, Old LJ, Osborne MP. Prediction of early relapse in patients with operable breast cancer by detection of occult bone marrow micrometastases. J Clin Oncol 1991;9:1749-1756.
Wong GYC, Yu QQ, Osborne MP. Bone marrow micrometastasis is a significant predictor of long-term relapse-free survival for breast cancer by a non-proportional hazards model. Breast Cancer Res Treat 2003;82:Suppl 1:S99-S99. abstract.
Osborne M, Wong G, Asina S, Old LJ, Cote RJ, Rosen PP. Sensitivity of immunocytochemical detection of breast cancer cells in human bone marrow. Cancer Res 1991;51:2706-2709.
DerSimonian R. Meta-analysis in the design and monitoring of clinical trials. Stat Med 1996;15:1237-1248.
Kaplan EL, Meier P. Nonparametric estimation from incomplete observations. J Am Stat Assoc 1958;53:457-481.
Peto R, Peto J. Asymptotically efficient rank invariant test procedures. J R Stat Soc 1972;135:185-206.
Cox DR. Regression models and life-tables. J R Stat Soc 1972;34:187-220.
Collett D. Modelling survival data in medical research. 2nd ed. Boca Raton, Fla.: Chapman & Hall/CRC, 2003.
Grambsch P, Therneau TM. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika 1994;81:515-526.
Hunink MG, Wong JB. Meta-analysis of failure-time data with adjustment for covariates. Med Decis Making 1994;14:59-70.
Clarke MJ, Stewart LA. Obtaining data from randomised controlled trials: how much do we need for reliable and informative meta-analyses? BMJ 1994;309:1007-1010.
Gangnus R, Langer S, Breit E, Pantel K, Speicher MR. Genomic profiling of viable and proliferative micrometastatic cells from early-stage breast cancer patients. Clin Cancer Res 2004;10:3457-3464.
Schmidt-Kittler O, Ragg T, Daskalakis A, et al. From latent disseminated cells to overt metastasis: genetic analysis of systemic breast cancer progression. Proc Natl Acad Sci U S A 2003;100:7737-7742.
Fehm T, Sagalowsky A, Clifford E, et al. Cytogenetic evidence that circulating epithelial cells in patients with carcinoma are malignant. Clin Cancer Res 2002;8:2073-2084.(Stephan Braun, M.D., Flor)