当前位置: 首页 > 期刊 > 《临床肿瘤学》 > 2005年第11期 > 正文
编号:11332110
Nomograms to Predict Pathologic Complete Response and Metastasis-Free Survival After Preoperative Chemotherapy for Breast Cancer
http://www.100md.com 《临床肿瘤学》
     the Departments of Breast Medical Oncology, Biostatistics and Applied Mathematics, and Pathology, The University of Texas M.D. Anderson Cancer Center, Houston, TX

    Breast Cancer Unit and UPRES EA, Institut Gustave Roussy, Villejuif, France

    Departments of Gynecologic Oncology and Breast Cancer Surgery (DA), Hotel-Dieu, Beirut, Lebanon.

    ABSTRACT

    PURPOSE: To combine clinical variables associated with pathologic complete response (pCR) and distant metastasis–free survival (DMFS) after preoperative chemotherapy (PC) into a prediction nomogram.

    PATIENTS AND METHODS: Data from 496 patients treated with anthracycline PC at the Institut Gustave Roussy were used to develop and calibrate a nomogram for pCR based on multivariate logistic regression. This nomogram was tested on two independent cohorts of patients treated at the M.D. Anderson Cancer Center. The first cohort (n = 337) received anthracycline; the second cohort (n = 237) received a combination of paclitaxel and anthracycline PC. A separate nomogram to predict DMFS was developed using Cox proportional hazards regression model.

    RESULTS: The pCR nomogram based on clinical stage, estrogen receptor status, histologic grade, and number of preoperative chemotherapy cycles had good discrimination and calibration in the training and the anthracycline-treated validation sets (concordance indices, 0.77, 0.79). In the paclitaxel plus anthracycline group, when the predicted pCR rate was less than 14%, the observed rate was 7.5%; for a predicted rate of 38%, the actual rate was 85%. For a predicted rate between 14% to 38%, the observed rates were 50% with weekly and 27% with 3-weekly paclitaxel. This indicates that patients with intermediate chemotherapy sensitivity benefit the most from the optimized schedule of paclitaxel. Patients unlikely to achieve pCR to anthracylines remain at low probability for pCR, even after inclusion of paclitaxel. The nomogram for DMFS had a concordance index of 0.72 in the validation set and outperformed other prediction tools (P = .02).

    CONCLUSION: Our nomograms predict pCR accurately and can serve as a basis to integrate future molecular markers into a clinical prediction model.

    INTRODUCTION

    Preoperative chemotherapy is increasingly used in the management of breast cancer.1 It is considered to be the standard of care for locally advanced and inoperable tumors. The main advantages of preoperative therapy include induction of tumor shrinkage that may render inoperable tumors amenable to surgery and may allow smaller resection and better cosmetic outcome for patients with operable tumors.2,3 Pathologic complete eradication of the invasive cancer (pathologic complete response; pCR) also provides a powerful early surrogate of long-term survival and could be considered as a marker of benefit from chemotherapy.4,5 Administration of chemotherapy before surgery is an ideal clinical setting to discover molecular predictors of response to therapy.6 There are several clinical and pathologic variables that are consistently associated with a better response to preoperative chemotherapy. These include estrogen receptor (ER)–negative status, high grade, high proliferative activity, and smaller tumor size, among others.7,8 It is likely that a combination of these clinical variables is a more accurate predictor of response than any of the variables alone. Clinical and pathologic variables have successfully been combined into accurate prognostic predictors to estimate the probability of breast cancer recurrence for individual patients with and without adjuvant systemic therapy.9,10 No similar multivariable predictor exists to estimate the probability of pCR to preoperative chemotherapy. Also, all of the current prognostic predictors were optimized for surgical pathologic data that were available at the time of diagnosis. However, tumor size and nodal status change in response to preoperative chemotherapy. Therefore, the extent to which the standard prognostic predictors such as the Adjuvantonline (www.adjuvantonline.com) or the Nottingham Prognostic Index provide accurate predictions for patients treated with preoperative chemotherapy remains unclear.11

    The aim of this current study was to combine clinical pathologic variables that are associated with pCR and distant metastasis–free survival after preoperative chemotherapy into prediction nomograms. Nomograms are statistical tools that enable users to calculate the overall probability of a specific clinical outcome for an individual patient.12,13 We developed and validated nomograms that predict probability of pCR to anthracycline and paclitaxel plus anthracycline preoperative chemotherapies, respectively. We also developed and validated a prognostic model that can be used to predict metastasis-free survival for patients who received primary chemotherapy. These nomograms could be used to estimate the probability of benefit from preoperative chemotherapy for an individual patient and can serve as basis to integrate future molecular markers into clinical prediction models.

    PATIENTS AND METHODS

    Study Population

    Using the institutional clinical databases from the Institut Gustave Roussy (IGR) and The University of Texas M.D. Anderson Cancer Center (MDACC), we identified 1,147 women who were treated with anthracycline-based or paclitaxel plus anthracycline preoperative chemotherapies on various clinical trials. Primary results from these trials have previously been reported.3,14,15 These patients were divided into four cohorts according to treatment center and regimen. The first cohort included 496 patients treated at IGR with three (n = 229) or four (n = 267) courses of anthracycline-based preoperative therapy (doxorubicin 50 mg/m2, vincristine 1 mg/m2, cyclophosphamide 600 mg/m2, methotrexate 30 mg/m2, and fluorouracil 900 mg/m2; doxorubicin 50 mg/m2 or epirubicin 50/60 mg/m2 and cyclophosphamide 600 mg/m2 and fluorouracil 900 mg/m2; and epirubicin 100 mg/m2, cyclophosphamide 1,000 mg/m2, and fluorouracil 1,000 mg/m2) and was used as a training set to develop predictive and prognostic models. The second cohort included 337 patients treated at MDACC with four courses of fluorouracil 500 mg/m2 and doxorubicin 50 mg/m2 given as a 72-hour continuous infusion and cyclophosphamide 500 mg/m2 (FAC) and was used as a validation set. The third cohort included 258 patients from MDACC who received preoperative paclitaxel either weekly (150 mg/m2 for lymph node-positive and 80 mg/m2 node-negative disease x 12 courses; n = 131) or every 3 weeks (225 mg/m2 given as a 24-hour continuous infusion x four courses; n = 127) followed by four additional courses of FAC chemotherapy. Data from these patients was used for validation of the prediction models developed from cohort 1 and subsequently to optimize the pCR nomogram for paclitaxel-containing preoperative chemotherapy. Cohort 4 included 111 patients treated with paclitaxel/FAC at MDACC and was used to estimate the accuracy of the paclitaxel-optimized pCR predictor.

    The clinical and histologic characteristics of all patients were prospectively entered into institutional clinical databases. No central pathology review was performed; routine diagnostic results from the medical record were entered into the database. Clinical tumor size at diagnosis was determined by physical examination and imaging, including mammograms. ER measurements were available for all patients; however, HER-2 results were not available for most of these cases. Patients were considered to be ER-positive if more than 10% of cells stained positive for ER by immunohistochemistry or if ER receptor was more than 10 fmol in ligand binding assay. The patients at IGR were treated between 1987 and 2000; patients at MDACC were treated between 1989 and 2003. All patients underwent mastectomy or breast-conserving surgery plus radiation therapy and axillary dissection after preoperative chemotherapy, and ER-positive patients also received 5 years of tamoxifen. The details of primary and adjuvant chemotherapy, radiation, and hormone therapy modalities used have been reported previously.3,14,15 Patient characteristics are reported in Table 1. In the IGR series, histologic grade was defined according to the modified Scarff, Bloom and Richardson (SBR) system described by Contesso et al.16 In the MDACC series, the grade was defined according to the modified Black's nuclear grade. In both institutions, patients without residual invasive cancer in the breast or in the lymph nodes were considered to have pCR. The number of histologically positive axillary lymph nodes was determined by examination of serial macroscopic sections of each lymph node. The median follow-up time was 96 months for the entire population.

    Statistical Analysis

    To develop well-calibrated and exportable nomograms for pCR and for freedom from distant metastases (FDM), we built each model in a training cohort and validated it in an independent validation cohort. Multivariate logistic regression analysis was used to test the association of patient age, initial clinical tumor diameter, multicentricity, histologic type of tumor, SBR grade, ER status, and initial clinical lymph node status to response to chemotherapy. This model was then used to predict individual patient probability of response to primary chemotherapy. FDM was estimated using the Kaplan-Meier method. Cox proportional hazards regression was used for multivariable analysis. Multivariate logistic regression analysis and the Cox proportional hazards regression model were used to construct the nomograms. Backward variable selection was performed to determine independent covariates. Continuous variables were also fit using restricted cubic splines to relax the linearity assumptions, but these models were not selected because continuous variables (ie, tumor size, number of metastatic nodes, and age) were linearly correlated with outcome. However, the square root of the number of metastatic nodes yielded a better linear correlation with FDM and was used in the survival model. Interactions between covariates were also tested. For the survival model, patients with complete eradication of breast tumor after preoperative chemotherapy were considered separately because their outcome could not be fit linearly with the Cox model.

    The model performance was quantified with respect to discrimination and calibration. Discrimination (ie, whether the relative ranking of individual predictions is in the correct order) was quantified with the area under the receiver operating characteristic curve or with the concordance index, which is similar to the area under the receiver operating characteristic curve (AUC) but appropriate for censored data. The concordance index is the probability that given two randomly selected patients, the patient with the worse outcome will in fact have a worse outcome prediction. The concordance index ranges from 0 to 1, with 1 indicating perfect concordance, 0.5 indicating no better concordance than chance, and 0 indicating perfect discordance. We used the bootstrapping method (200 repetitions) to obtain relatively unbiased estimates. Calibration (ie, agreement between observed outcome frequencies and predicted probabilities) was studied with graphical representations of the relationship between the observed outcome frequencies and the predicted probabilities (calibration curves): the grouped proportions versus mean predicted probability in groups defined by quantiles and the logistic calibration were represented. A calibration curve can be approximated by a regression line with intercept and slope ?. These parameters can be estimated in a logistic regression model with the event as outcome and the linear predictor as the only covariate. Well-calibrated models have = 0 and ? = 1. Therefore, a sensible measure of calibration is a likelihood ratio statistic testing the null hypothesis that = 0 and ? = 1. The statistic has a 2 distribution with 2 df (unreliability [U] -statistic).17 There is no accepted test to assess the calibration of a censored data prediction. We compared outcome prediction to those provided by Adjuvantonline: for each patient, we evaluated the 10-year relapse-free rate, then we calculated the average annual risk of breast cancer relapse and applied the proportional risk reductions provided by tamoxifen. All analyses were performed using the R package with the Survival, Design, Hmisc, Rpart, and Lexis libraries (http://lib.stat.cmu.edu/R/CRAN/).

    RESULTS

    Prediction of Probability of pCR to Anthracycline-Based Chemotherapy Based on Clinicopathologic Variables

    In the training set of 496 patients treated at IGR, 45 patients (9%) had pCR. In multivariate logistic regression analysis, clinical TNM stage at diagnosis, ER status, histologic grade, and number of preoperative chemotherapy cycles were independently associated with pCR. To construct the nomogram from these clinical pathologic variables, we added age at diagnosis because it improved the calibration of the logistic regression model. The final prediction model showed an AUC of 0.77 in the training set (Fig 1A and B). We performed internal validation with bootstrap sampling to determine calibration accuracy. The corrected AUC after bootstrapping remained high (0.76). The nomogram to predict probability of pCR to anthracycline-based preoperative chemotherapy is reported in Figure 2.

    We next examined the accuracy of the nomogram in an independent data set of 337 patients treated at MDACC. In this validation set, nuclear grade was used instead of the modified SBR grade. A high degree of concordance between histologic grade (which includes assessment of nuclear pleomorphism) and nuclear grade has been reported previously.18 We compared the nomogram-predicted pCR rates to the observed rates in the validation set. The nomogram accurately predicted response (Fig 1A). For the validation set, the calibration was good, with no significant difference (P = .25) between the predicted and the observed pCR proportions. The AUC for the validation data set was 0.79 (Fig 1B). These results demonstrate that the individual probability of achieving pCR to three to four courses of anthracycline-based combination chemotherapy can be predicted accurately by combining information from routinely available clinicopathologic variables.

    Prediction pCR After Anthracycline and Paclitaxel Preoperative Chemotherapy

    Having established that our nomogram accurately predicts probability of pCR to three to four courses of anthracycline-based preoperative therapy, we investigated whether the nomogram would also predict pCR to anthracycline and paclitaxel preoperative chemotherapy. We tested the predictor on patients from MDACC who were randomly assigned to receive either weekly or 3-weekly paclitaxel followed by four courses of FAC in a clinical trial (ID 98-240; n = 237). Nomogram-predicted and observed pCR rates in this cohort of patients are shown on Figure 3A (weekly paclitaxel + FAC) and 3B (3-weekly paclitaxel + FAC). The nomogram significantly underestimated the pCR rate in the paclitaxel-treated patients (P < .001). This discrepancy between the predicted and observed pCR rates may be used to define the group of patients whose chance of pCR was improved by the inclusion of paclitaxel. We performed recursive partitioning where patients were grouped by predicted pCR rate to anthracycline alone, and observed pCR rates were plotted by paclitaxel schedule. The resulting tree is shown in Figure 3C. The group of patients whose predicted probability of achieving pCR to anthracycline chemotherapy was less than 14% had an actual pCR rate of 7.5% (n = 9 of 121). This rate was not influenced by the schedule of paclitaxel. Patients with 38% probability of achieving pCR with anthracycline-based therapy alone had an actual pCR rate of 85% (n = 17 of 20). Again, this rate was not influenced by the schedule of paclitaxel. Most interestingly, patients with intermediate chemotherapy sensitivity whose predicted pCR rate was between 14% and 38% showed an observed rate of 50% (n = 25 of 51) when treated with weekly paclitaxel + FAC and 27% (n = 12 of 45) when treated with 3-weekly paclitaxel. These observations suggest that patients with intermediate chemotherapy sensitivity benefit the most from an optimized schedule of paclitaxel. Patients with highly chemotherapy-sensitive disease benefit from paclitaxel regardless of the schedule, whereas patients who are unlikely to achieve pCR with an anthracycline-based regimen remain at low probability for pCR, even after inclusion of paclitaxel.

    It could be clinically useful to be able to predict the probability of pCR to paclitaxel-anthracycline preoperative chemotherapy. This could be accomplished by optimizing and recalibrating (using the same linear predictor) the nomogram developed for anthracycline-treated patients. We used the 237 patients from the randomized study to build a new taxane-optimized nomogram. The concordance indices of this nomogram before and after bootstrapping were 0.81 (P < 10–12) and 0.81 (P < 10–12), respectively, in the training set. A new cohort of 111 patients treated with paclitaxel-FAC preoperative chemotherapy was used as independent validation set to estimate the accuracy of this new prediction model. The concordance index was 0.77 in the validation set. The calibration was also good with no significant difference (P = .14) between the predicted and the observed proportions of pCR (Fig 4).

    Prediction of Freedom From Distant Metastasis (FDM) After Preoperative Chemotherapy

    pCR is an important early surrogate of long-term disease-free survival and could be used as an early indicator of benefit from chemotherapy. However, a more important outcome to predict is the probability of FDM after completion of therapy. Therefore, we also developed a metastases-free survival nomogram based on clinical pathologic findings at the time of surgery, postchemotherapy. In the IGR cohort (n = 469 with 5-year follow-up), the metastasis-free survival rates were 69% and 50% at 5 and 10 years, respectively. In the Cox proportional hazards model, the number of metastatic lymph nodes (P < 10–9), the residual pathologic tumor size (P < 10–2), ER status (P < 10–3), histologic grade (P < 10–2), and histologic type (P = 10–2) were associated with metastasis-free survival. The concordance index for the model was 0.71 in the training set. A nomogram constructed from the Cox model appears in Figure 5. Figure 6A shows that the nomogram was also well calibrated to predict 5- and 10-year metastasis-free survival in the training set. Next, we tested the predictive accuracy of the survival nomogram on a validation set derived from patient data from MDACC. Only patients with a minimum of 5 years of follow-up were included in this analysis (n = 308, all patients received only anthracycline-based preoperative chemotherapy like patients of the training set). The predicted and observed 5-year metastasis-free survival rates were highly concordant (Fig 6B).

    Currently, probably the most commonly used and validated instrument to predict survival of a breast cancer patient with or without adjuvant chemotherapy in the United States is the AdjuvantOnline software program (http://www.adjuvantonline.com).10 This tool uses clinical and surgical pathologic information that is available at the time of diagnosis. However, tumor size and the number of lymph nodes involved with cancer can change substantially during preoperative chemotherapy. To examine how accurately AdjuvantOnline predicts metastasis-free survival in patients who received preoperative chemotherapy, we used the software to make survival predictions with and without including the type of preoperative chemotherapy that was given. For each patient, we calculated the risk of relapse at 5 years using the surgical pathologic results after preoperative chemotherapy. The relapse rates predicted by AdjuvantOnline without including chemotherapy effect were plotted against the observed proportions (Fig 6B). AdjuvantOnline slightly underestimated the relapse rate. Relapse predictions were even more underestimated when chemotherapy effectiveness was added to the prediction (data not shown). The concordance index was 0.63 with and 0.69 without considering the effect of chemotherapy in the prediction. In comparison, the concordance index for our metastasis-free survival nomogram was 0.72 in this validation set (P = .02). This discrepancy is not surprising because AdjuvantOnline was optimized for clinical pathologic data from newly diagnosed patients without preoperative chemotherapy. The prognostic significance of 2.0-cm tumor size is likely to be quite different if this represents residual cancer after chemotherapy as opposed to baseline tumor size.11,19

    Neo!adjuvant

    We developed a computer interface that uses the above prediction models to estimate the probability of pCR and 5-year metastasis-free survival for individual patients treated with four courses of anthracycline or 6 months of paclitaxel plus anthracycline preoperative chemotherapies. This internet-based tool may assist patients and physicians in decision making regarding preoperative chemotherapy. The predictor is called Neo!adjuvant (version 1) and was programmed in Java text. An Internet browser with Java enablement is required to run the applets. An example of a screen is shown in Figure 7. The applets will be available online through the MDACC and the IGR Web sites: http://www.mdanderson.org/care_centers/breastcenter/dIndex.cfm?pn=448442B2-3EA5-4BAC-98310076A9553E63.

    DISCUSSION

    We developed prediction models based on clinical and pathologic characteristics of the primary tumor to estimate the probability of pCR and metastasis-free survival for patients treated with preoperative chemotherapy with anthracyclines alone or with a paclitaxel and anthracycline combination. These models were validated on independent cohorts of patients treated with similar preoperative chemotherapy. We also observed that the pCR predictor nomogram calibrated for patients who received four courses of anthracycline therapy underestimated the pCR rates in patients treated with paclitaxel plus anthracycline therapy in general. However, further examination of the results with recursive partitioning revealed that patients who were predicted to have extremely low pCR rates to four courses of anthracycline preoperative chemotherapy indeed had low pCR rates even after the inclusion of paclitaxel, regardless of schedule. In contrast, patients with high probability of pCR to four courses of anthracyclines had an observed pCR rate that was twice the predicted rate after the inclusion of paclitaxel, regardless of schedule. The schedule of administration of paclitaxel became important for those patients only who had intermediate probability to achieve pCR with anthracyclines alone. In this group, inclusion of weekly paclitaxel improved the pCR rates, whereas every 3-week paclitaxel did not improve pCR rates in comparison to the predicted rate.

    These observations suggest that patients with high and intermediate chemotherapy sensitivity benefit the most from an optimized schedule of paclitaxel and that patients with extremely chemotherapy-resistant cancers may not benefit much from inclusion of paclitaxel. These results are consistent with the observations reported by the so-called Aberdeen neoadjuvant trial.20 This study randomly assigned patients who responded to four courses of anthracycline-based preoperative chemotherapy to either receive four more cycles of the same chemotherapy or receive four treatments with docetaxel. Patients who had no initial clinical response were all switched to docetaxel. The study reported substantially improved pCR rates among the anthracycline responders when switched to docetaxel compared with continuing with anthracycline therapy. However, pCR rates remained low in the clinically anthracycline-resistant group despite additional therapy with docetaxel. These observations have consequences for future adjuvant clinical trial design. It seems that further improvements in outcome by optimizing cytotoxic chemotherapy combinations will only be achieved for patients who have highly or moderately chemotherapy-sensitive cancers. Those who are predicted to have extremely low probability to achieve pCR after four courses of anthracycline therapy should be steered toward clinical trials that incorporate novel agents that may revert chemotherapy resistance. Inclusion of trastuzumab in preoperative regimens for HER-2 amplified cancer serves as an encouraging example that restoration of chemotherapy sensitivity may be possible with novel therapeutic agents.21

    We assume that pCR is a valid early surrogate of long-term survival and cure from breast cancer. A large body of evidence from retrospective analysis of several clinical trials supports this assumption. 4,5,18,22,23 However, how to translate improvements in pCR rate to improvement in long-term survival in prospective clinical trials remains uncertain.24 If long-term benefit from chemotherapy is mostly, though not exclusively, concentrated on patients who are predicted to achieve pCR, a 10% to 15% absolute increase in pCR rate may not translate into detectable improved overall survival unless large numbers of patients are included in the trial.

    We also developed and validated a nomogram to predict 5-year metastasis-free survival after preoperative chemotherapy. The predictive accuracy of this nomogram in the independent validation set outperformed AdjuvantOnline both in terms of calibration and discrimination. This is not surprising because this widely used and validated tool was optimized for clinical and surgical pathologic data obtained at diagnosis without preoperative chemotherapy.

    In conclusion, we developed nomograms that can be used to predict outcome after primary chemotherapy in patients with breast cancer, including the probability of achieving pCR and the risk of disease recurrence. These nomograms may be useful when counseling patients about treatment options. For example, patients unlikely to achieve pCR (predicted rate < 10%) may not benefit much from preoperative (or indeed adjuvant) chemotherapy; however, patients with high (predicted pCR rate > 30%) and intermediate sensitivity may benefit the most from inclusion of paclitaxel in their preoperative chemotherapy regimen. Our clinical prediction model can also serve as basis to integrate future molecular markers into the model. It is expected that inclusion of molecular markers such as individual genes or predictive signatures will improve the regimen specificity of the clinical variable-based model.25

    Authors' Disclosures of Potential Conflicts of Interest

    The authors indicated no potential conflicts of interest.

    NOTES

    Supported by the Nellie B. Connally Breast Cancer Research Fund. R.R. was supported in part by a grant from Philippe Foundation, Paris and New York. L.P. is supported by RO1 CA106290-01 of the National Cancer Institute.

    Authors' disclosures of potential conflicts of interest are found at the end of this article.

    REFERENCES

    Schwartz GF, Hortobagyi GN: Proceedings of the consensus conference on neoadjuvant chemotherapy in carcinoma of the breast, April 26-28, 2003, Philadelphia, Pennsylvania. Cancer 100:2512-2532, 2004

    Mauri D, Pavlidis N, Ioannidis JP: Neoadjuvant versus adjuvant systemic treatment in breast cancer: A meta-analysis. J Natl Cancer Inst 97:188-194, 2005

    Rouzier R, Mathieu MC, Sideris L, et al: Breast-conserving surgery after neoadjuvant anthracycline-based chemotherapy for large breast tumors. Cancer 101:918-925, 2004

    Rouzier R, Extra JM, Klijanienko J, et al: Incidence and prognostic significance of complete axillary downstaging after primary chemotherapy in breast cancer patients with T1 to T3 tumors and cytologically proven axillary metastatic lymph nodes. J Clin Oncol 20:1304-1310, 2002

    Kuerer HM, Newman LA, Smith TL, et al: Clinical course of breast cancer patients with complete pathologic primary tumor and axillary lymph node response to doxorubicin-based neoadjuvant chemotherapy. J Clin Oncol 17:460-469, 1999

    Pusztai L, Gianni L: Prediction of response to preoperative chemotherapy in operable breast cancer. Nat Clin Pract Oncol 1:44-50, 2004

    Chang J, Powles TJ, Alfred DC, et al: Biologic markers as predictors of clinical outcome from systemic therapy for primary operable breast cancer. J Clin Oncol 17:3058-3063, 1999

    Ellis P, Smith I, Ashley S, et al: Clinical prognostic and predictive factors for primary chemotherapy in operable breast cancer. J Clin Oncol 16:107-114, 1998

    Galea MH, Blamey RW, Elston CE, et al: The Nottingham Prognostic Index in primary breast cancer. Breast Cancer Res Treat 22:207-219, 1992

    Ravdin PM, Siminoff LA, Davis GJ, et al: Computer program to assist in making decisions about adjuvant therapy for women with early breast cancer. J Clin Oncol 19:980-991, 2001

    Chollet P, Amat S, Belembaogo E, et al: Is Nottingham prognostic index useful after induction chemotherapy in operable breast cancer? Br J Cancer 89:1185-1191, 2003

    Eastham JA, Kattan MW, Scardino PT: Nomograms as predictive models. Semin Urol Oncol 20:108-115, 2002

    Kattan MW, Giri D, Panageas KS, et al: A tool for predicting breast carcinoma mortality in women who do not receive adjuvant therapy. Cancer 101:2509-2515, 2004

    Valero V, Buzdar AU, McNeese M, et al: Primary chemotherapy in the treatment of breast cancer: The University of Texas M. D. Anderson cancer center experience. Clin Breast Cancer 3:S63-S68, 2002 (suppl 2)

    Rajan R, Poniecka A, Smith T, et al: Change in tumor cellularity of breast cancer following neoadjuvant chemotherapy as a variable in the pathological assessment of response. Cancer 100:1365-1373, 2004

    Contesso G, Mouriesse H, Friedman S, et al: The importance of histologic grade in long-term prognosis of breast cancer: A study of 1,010 patients, uniformly treated at the Institut Gustave-Roussy. J Clin Oncol 5:1378-1386, 1987

    Cox DR: Two further applications of a model for binary regression. Biometrika 45:562-565, 1958

    Fisher B, Redmond C, Fisher ER, et al: Relative worth of estrogen or progesterone receptor and pathologic characteristics of differentiation as indicators of prognosis in node negative breast cancer patients: Findings from National Surgical Adjuvant Breast and Bowel Project Protocol B-06. J Clin Oncol 6:1076-1087, 1988

    van der Hage JA, van de Velde CJ, Julien JP, et al: Preoperative chemotherapy in primary operable breast cancer: Results from the European Organization for Research and Treatment of Cancer trial 10902. J Clin Oncol 19:4224-4237, 2001

    Smith IC, Heys SD, Hutcheon AW, et al: Neoadjuvant chemotherapy in breast cancer: Significantly enhanced response with docetaxel. J Clin Oncol 20:1456-1466, 2002

    Buzdar AU, Ibrahim NK, Francis D, et al: Significantly higher pathologic complete remission rate after neoadjuvant therapy with trastuzumab, paclitaxel, and epirubicin chemotherapy: Results of a randomized trial in human epidermal growth factor receptor 2-positive operable breast cancer. J Clin Oncol 23:3676-3685, 2005

    Fisher B, Bryant J, Wolmark N, et al: Effect of preoperative chemotherapy on outcome of women with operable breast cancer. J Clin Oncol 16:2672-2685, 1998

    Scholl SM, Pierga JY, Asselain B, et al: Breast tumour response to primary chemotherapy predicts local and distant control as well as survival. Eur J Cancer 31A:1969-1975, 1995

    Bear HD, Anderson S, Smith RE, et al: A randomized trial comparing preoperative (preop) doxorubicin/cyclophosphamide (AC) to preop AC followed by preop docetaxel (T) and to preop AC followed by postoperative T in patients with operable carcinoma of the breast: Results of NSABP B-27. Breast Cancer Res Treat 88:16, 2004 (abstr 26)

    Roman R, Rajan R, Hess KR, et al: Microtubule associated protein tau is a predictive marker and modulator of response to paclitaxel-containing preoperative chemotherapy in breast cancer. Proc Natl Acad Sci U S A 102:8315-8320, 2005(Roman Rouzier, Lajos Pusz)