|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Originally published as JCO Early Release 10.1200/JCO.2007.10.6823 on September 4 2007 © 2007 American Society of Clinical Oncology. Measurement of Residual Breast Cancer Burden to Predict Survival After Neoadjuvant Chemotherapy
From the Departments of Pathology, Surgery, and Breast Medical Oncology, University of Texas M.D. Anderson Cancer Center, Houston, TX; and Nuvera Biosciences Inc, Woburn, MA Address reprint requests to W. Fraser Symmans, MD, Department of Pathology, Unit 85, The University of Texas M.D. Anderson Cancer Center,1515 Holcombe Blvd, Houston, TX 77030-4009; e-mail: fsymmans{at}mdanderson.org
Purpose To measure residual disease after neoadjuvant chemotherapy in order to improve the prognostic information that can be obtained from evaluating pathologic response. Patients and Methods Pathologic slides and reports were reviewed from 382 patients in two different treatment cohorts: sequential paclitaxel (T) then fluorouracil, doxorubicin, and cyclophosphamide (FAC) in 241 patients; and a single regimen of FAC in 141 patients. Residual cancer burden (RCB) was calculated as a continuous index combining pathologic measurements of primary tumor (size and cellularity) and nodal metastases (number and size) for prediction of distant relapse-free survival (DRFS) in multivariate Cox regression analyses. Results RCB was independently prognostic in a multivariate model that included age, pretreatment clinical stage, hormone receptor status, hormone therapy, and pathologic response (pathologic complete response [pCR] v residual disease [RD]; hazard ratio = 2.50; 95% CI 1.70 to 3.69; P < .001). Minimal RD (RCB-I) in 17% of patients carried the same prognosis as pCR (RCB-0). Extensive RD (RCB-III) in 13% of patients was associated with poor prognosis, regardless of hormone receptor status, adjuvant hormone therapy, or pathologic American Joint Committee on Cancer stage of residual disease. The generalizability of RCB for prognosis of distant relapse was confirmed in the FAC-treated validation cohort. Conclusion RCB determined from routine pathologic materials represented the distribution of RD, was a significant predictor of DRFS, and can be used to define categories of near-complete response and chemotherapy resistance.
A central tenet of neoadjuvant clinical trials is that tumor response, as a surrogate end point, should be strongly correlated with long-term patient survival.1,2 Pathologic complete response (pCR) is associated with long-term survival, and has been adopted as the primary end point for neoadjuvant trials.3-12 While it is generally held that a definition of pCR should include patients without residual invasive carcinoma in the breast (pT0), the presence of nodal metastasis, minimal residual cellularity, and residual in situ carcinoma are not consistently defined as pCR or residual disease (RD).11-15 When there is no residual invasive cancer in the breast, the number of involved axillary lymph nodes is inversely related to survival.11 Conversely, patients who convert to node-negative status after treatment have excellent survival, even if there is RD in the breast.17 Consequently, the combination of tumor size and nodal status after neoadjuvant treatment is prognostic.18 Alternatively, the Miller and Payne classification ignores tumor size and nodal status altogether, and estimates only the decrease in cancer cellularity after treatment.10 However, the reduction in cellularity is often greatest when the residual tumor is small, suggesting a relationship between residual size and cellularity.19 While microscopic RD, altered cytologic appearance, and estimated tumor volume less than 1 cm3 also indicate good response, these tend to be descriptive parameters and are also difficult to apply to tumor beds with dispersed microscopic foci of carcinoma.3-6,9,20 Finally, there is no evidence that residual in situ carcinoma alone increases risk of future distant relapse.12,21,22 Stronger prognostic information from pathologic response can increase the clinical and scientific information learned from neoadjuvant clinical trials. Dichotomization of response as pCR or RD is overly simplistic for these objectives because RD after neoadjuvant treatment includes a broad range of actual responses from near pCR to frank resistance. More effective or prolonged neoadjuvant treatments should reduce the extent of RD in many patients, possibly blurring the prognostic distinction between pCR and RD. In contrast, it should be possible to identify patients with resistant disease in order to develop predictive tests for this adverse outcome. Therefore, we proposed to measure residual cancer burden (RCB) as a continuous variable derived from the primary tumor dimensions, cellularity of the tumor bed, and axillary nodal burden.
Pathologic Review The authors proposed that the extent of RD in the post-treatment surgical resection specimen could be determined from bidimensional diameters of the primary tumor bed in the resection specimen (d1 and d2), the proportion of the primary tumor bed that contains invasive carcinoma (finv), the number of axillary lymph nodes containing metastatic carcinoma (LN), and the diameter of the largest metastasis in an axillary lymph node (dmet; Appendix 1, online only). Largest bidimensional measurements of the residual primary tumor bed were recorded from the macroscopic description in the pathology report and confirmed after review of corresponding slides. If multiple tumors were present, the dimensions of the largest were recorded. Bidimensional measurements of the primary tumor bed (millimeters) were combined as follows:
Patients and Materials
An independent validation cohort (Table 1) included 141 patients treated for 3 months with FAC alone, followed by surgical resection of the residual tumor and axillary dissection, and then 3 additional months of adjuvant chemotherapy (FAC for 129 patients and other noncrossresistant chemotherapy for 12 patients who had clinically stable or progressive disease; protocols MDACC DM 91-015, DM 94-002.25,26 This cohort included patients with more advanced disease (63% node positive v 47%; 100% stage II/III v 90%) and larger size tumors (mean diameter, 3.9 v 2.5 cm). The pCR rate was lower in the validation cohort (16%), consistent with the shorter duration of preoperative chemotherapy and the absence of a taxane. Patients with hormone receptor–positive breast cancer were offered 5 years of adjuvant tamoxifen according to treatment guidelines at the time. The institutional review board of MDACC approved these protocols and all patients signed an informed consent form before initiation of therapy. Pathologic review and data analyses were conducted in accordance with a separately approved institutional review board protocol (MDACC LAB02-010).
Statistical Analysis
Development of RCB Index The four parameters of residual tumor (dprim, finv, LN, and dmet) were individually associated with significantly higher risk of distant relapse (P < .001) after T/FAC chemotherapy in univariate Cox regression analyses and maintained significance as independent predictors in the main effects multivariate Cox regression model (Fig 1 and Appendix 3, online only). To calculate a single index of RCB, we first combined the covariates to terms that measure RCB in the primary tumor bed (RCBprim = finv dprim) and in regional metastases (RCBmet = 4 (1 – 0.75LN) dmet). The metastatic term is intended to be proportional to the sum of diameters of the affected lymph nodes, but since only the size of the largest metastasis is routinely measured we assumed that additional nodal metastases each have 75% of the diameter of the next-largest metastasis (Appendix 3, online only).
The distributions of the primary and metastatic RCB components were highly right skewed (Appendix 3, online only) and so an unconditional power transformation on the two components was applied.27 The transformed terms were then scaled to match the 95th percentiles of their respective distributions, and added to define the RCB index:
Residual Cancer Burden Index As a Predictor of Distant Relapse
There appeared to be a disproportionate increase in the risk of 5-year distant relapse with increasing RCB values after T/FAC chemotherapy (Fig 2A). A similar analysis stratified by hormone treatment status demonstrated an overall increased risk of relapse with increasing RCB levels for patients who did not receive adjuvant hormone therapy (Fig 2B). The likelihood of 5-year relapse in patients who received hormone treatment was lower for the entire range of RCB values, and it increased more gradually through the lower spectrum of RCB values (Figs 2A and 2B). However, both groups had similar gradients of increasing risk through the higher spectrum of RCB values, indicating comparatively greater risk of relapse with more extensive RD.
RCB Index Identifies Near pCR and Resistant Groups We identified two cutoff points to assign patients with RD (RCB > 0) after T/FAC treatment into one of three classes: RCB-I (minimal RD), RCB-II (moderate RD), and RCB-III (extensive RD). Two cutoff points were determined sequentially by maximizing the profile log-likelihood of a multivariate Cox model that included the clinical covariates and the dichotomized RCB index (Appendix 2, online only). The first cutoff point (RCB-III v RCB-I/II) was selected as the 87th percentile (RCB, 3.28), and the second (RCB-I v RCB-II) corresponds to the 40th percentile (RCB, 1.36). The cutoff points defined subgroups of RCB-0 to RCB-III with increasingly poor prognosis (Appendix Table A1, online only, and Fig 3A). The cumulative incidence estimate of the overall probability of relapse within 5 years adjusted for the competing risk of death events was 5.4% for the pCR group and 2.4% for the group with minimal RD (RCB-I), whereas it was 53.6% for the group with extensive RD (RCB-III). The difference in the rates of distant relapse at 5 years between the groups with the worst (RCB-III) and best (RCB-0) prognoses was 48.2% (95% CI, 28.1 to 65.6), providing sufficient separation to reliably classify patients into groups with different prognosis.28
Because adjuvant hormone therapy likely affects relapse-free survival, we evaluated the risk of relapse within groups who did or did not receive adjuvant hormone treatment. All hormone receptor–positive patients were eligible for treatment and 91% of them underwent adjuvant hormone therapy. Women with RCB-0 or RCB-I after neoadjuvant T/FAC had excellent 5-year relapse-free prognosis irrespective of whether or not they received adjuvant hormone treatment (Figs 3B and 3C). It is noteworthy that nine patients with hormone receptor–negative breast cancer and RCB-III after neoadjuvant T/FAC chemotherapy all relapsed within 27 months (Fig 3B). The prognosis of those with RCB-II was improved in the group treated with adjuvant hormone therapy (Fig 3C).
RCB Groups Stratify Prognosis of Revised yAJCC Stage After Chemotherapy
Validation of RCB As Predictor of Distant Relapse We evaluated the intrinsic prognostic accuracy of the RCB-based survival model by calibrating the predicted probabilities of distant relapse at 5 years produced by the full multivariate Cox regression model (including RCB group) to the observed probabilities of relapse (Appendix 2, online only).30 The calibration plot suggested that the predicted probabilities of distant relapse by the RCB survival model were similar to the empirical Kaplan-Meier estimates (Fig A1, online only; cross symbols). Next, we adjusted for potential overoptimism in the predictions from bias introduced by "using the data twice" first, for selecting cutoff points and subsequently for evaluating the model's predictive accuracy.31,32 The estimated global shrinkage factor of 0.871 indicated only moderate overfitting. The adjusted prognostic model and its calibration is shown in Figure A1, online only (filled symbols), and appears to predict accurately the relapse-free rates at 5 years in the T/FAC cohort.
Discrimination of this prognostic model between relapsed and nonrelapsed patients was measured using Harrell's c index (Appendix 2, online only).30 The bias-adjusted c-index in the development cohort was estimated to be 0.77 (95% CI, 0.69 to 0.84), indicating statistically significant discrimination (c = 0.5 for random predictions, c = 1 for perfectly discriminating model).32 Generalizability of the RCB system was evaluated in the independent validation cohort of patients treated with neoadjuvant FAC chemotherapy. RCB defined groups with increasingly poor 5-year and 10-year prognoses (Appendix Table A1 and Appendix 5, online only). The difference in the rates of distant relapse between the worst (RCB-III) and best (RCB-0) prognosis groups was 36.3% (95% CI, 21.4 to 51.4) at 5 years and 52.2% (95% CI, 35.1 to 66.9) at 10 years. The separation of the 5-year relapse rates is somewhat smaller in the FAC cohort than for the T/FAC cohort (48.2%), indicating some optimism in those predictions and possibly benefit from additional postoperative chemotherapy. No systematic bias was apparent in the calibration plot (Fig A1B, online only), especially for the optimism-adjusted model. The c-index of the prognostic model on the validation cohort was 0.70 (95% CI, 0.61 to 0.79) suggesting similar discriminatory ability. Taken together, these results validate the prognostic ability of the RCB system for predicting distant relapse in breast cancer patients treated with neoadjuvant T/FAC or FAC chemotherapy.
The lack of uniform methods to report pathologic response is a contributing factor in the recent erosion of confidence in the value of neoadjuvant trials to anticipate the results of larger adjuvant trials.11 Although pCR (including node-negative status) has consistently imparted an excellent prognosis in published studies, meaningful reporting of RD has been an elusive goal. This problem is accentuated when evaluation of pathologic response is limited to the review of archival pathology reports. This is because asymmetry of RD and variable hypocellularity after treatment are not usually quantified in the report, and are not captured by tumor diameter and nodal status alone. We have attempted to combine relevant pathological characteristics of RD into a composite index of RCB. Each variable in the equation for RCB has prognostic significance, and the calculated primary and metastatic terms in the equation are equivalently and independently prognostic. As a result, RCB is strongly prognostic, and represents the continuum of RD in a treated population. Our analyses of intrinsic prognostic accuracy (Fig A1, online only), discrimination, and generalizability (Tables 2 and A1 [online only], Fig A1 [online only]) did not demonstrate any major bias in our model of RCB to predict distant relapse, even though the FAC-treated patients received additional postoperative chemotherapy. Furthermore, RCB extends the prognostic value of our current dichotomous assessment of response as pCR or RD (Table 1), and the revised yAJCC stage classifications of RD (Fig 4). We have also been careful to employ methods of pathologic assessment that could feasibly be incorporated in routine diagnostic practice without adding to the cost of patient care. The variables used to calculate RCB can be simply obtained from pathologic review and entered into a calculation script that is freely available on the internet (www.mdanderson.org/breastcancer_RCB). A stepwise guide for the pathologic evaluation of post-treatment breast specimens is provided, along with links to illustrative examples. This Web site could be a useful tool for pathologists, and could also be employed in multicenter trials of neoadjuvant treatment to standardize sampling and reporting of pathologic findings from post-treatment specimens. Further studies should address interobserver variability of RCB measurements (and prognostic power), and evaluate RCB when used by other groups in other study populations. One must also consider whether incomplete pathologic data might invalidate the utility of RCB. For example, assessment of the residual primary tumor bed in patients who had pretreatment surgical biopsy might overestimate the response in the breast. Alternatively, assessment of the residual nodal cancer burden in patients who had a positive lymph node excised before neoadjuvant treatment might overestimate the nodal response. RCB measurements provide a continuous parameter of response, so that all subject responses contribute to the analysis. Therefore, small, phase II studies, treatment regimens with low pCR rates (such as hormone therapy), or with similar pCR rates, can be compared to identify differences in the extent of RD. RCB can also be divided into four classes (RCB-0 to RCB-III). We note that patients with minimal RD (RCB-I) had the same 5-year prognosis as those with pCR (RCB-0), irrespective of the type of neoadjuvant chemotherapy administered, adjuvant hormone therapy (Fig 3), or the pathologic stage of RD (Fig 4). Therefore, the combination of RCB-0 (pCR) and RCB-I expands the subset of patients who can be identified as having benefited from neoadjuvant chemotherapy. Extensive RD (RCB-III) was associated with poor prognosis, irrespective of the type of neoadjuvant chemotherapy administered, adjuvant hormone therapy (Fig 3), or the pathologic stage of RD (Fig 4). In particular, all patients with RCB-III after T/FAC chemotherapy, who did not receive adjuvant hormone therapy, suffered distant relapse within 3 years (Fig 3B). However, it should also be noted that 13% of patients with receptor-positive disease had RCB-III after T/FAC chemotherapy (21 of 160), with a 5-year distant relapse rate of 40% despite receiving adjuvant hormone treatment (Fig 3C). This identifies an important subset of patients with combined insensitivity to chemotherapy and hormone therapy, or with RD (after surgery) that is too extensive to be controlled by hormone therapy alone. Conversely, even a moderate response from chemotherapy (RCB-II) appears to improve the survival benefit from subsequent hormone therapy (Figs 2B, 3C). This illustrates how identification of the subset of receptor-positive patients who might correctly be spared (denied) adjuvant chemotherapy despite consensus treatment recommendations will require very careful selection based on the tumor's predicted chemosensitivity and the predicted endocrine sensitivity.33,34 It has been recommended that the predictive ability of a new marker should be evaluated based on whether the marker improves an already optimized multivariate model of available risk factors.35 On this basis, the RCB index is an independent new risk factor that improves the prediction of distant relapse after neoadjuvant chemotherapy compared with currently used risk factors. Although RCB could supplement existing methods to define pathologic response, independent validation of RCB is needed before it can be broadly used as a surrogate end point for patient survival.36
Appendix 1: Detailed Pathology Methods Residual cancer burden is estimated from routine pathologic sections of the primary breast tumor site and the regional lymph nodes. The number of positive lymph nodes (LN) and the diameter of the largest nodal metastasis (dmet) were obtained from microscopic evaluation of routine diagnostic slides. In general terms, pathologic evaluation of the primary tumor bed in the breast requires that the pathologist make three judgments: identify the cross-sectional dimensions of residual tumor bed (d1 and d2); estimate of the proportion of that residual tumor bed area that is involved by cancer (% CA), and estimate the proportion of the cancer that is in situ component (% CIS). Identification of the residual tumor bed requires gross (with or without radiologic) evaluation of the breast specimen to identify the macroscopic tumor bed. Using the schematic illustrations, the macroscopic tumor bed dimensions may also define the dimensions of the residual tumor bed (Figs A2A, A2C, A2D). However, the macroscopic tumor bed dimensions can overestimate the extent of residual cancer (Figs A2B, A3), and then the dimensions of the residual tumor bed (d1 and d2) are defined from microscopic evaluation of the extent of residual cancer in the corresponding slides from that macroscopic tumor bed. Microscopic residual cancer can extend beyond the confines of the macroscopic tumor bed (Fig A2E), and then the dimensions of the residual tumor bed (d1 and d2) are also defined from microscopic evaluation of the recognizable extent of residual cancer beyond the macroscopic tumor bed combined with the extent of residual cancer in the corresponding slides from the macroscopic tumor bed.
In summary, the information to define the largest cross-sectional dimensions of residual tumor bed (d1 and d2) was obtained from our comparison of the macroscopic tumor bed dimensions reported in the gross description with our microscopic study of the slides that were reported in the section code to represent the macroscopic tumor bed and immediately adjacent tissues. As noted in this article, the largest residual tumor was evaluated in cases of multicentric residual disease. The proportion of cancer (% CA) within the defined tumor bed area was estimated from microscopic evaluation of the corresponding slides from the residual tumor bed (Figs A2, A3). The estimated % CA could be high in a small area (Figs A2A, A2B), or lower in a larger area (Figs A2C, A2D). Also the estimated % CA could be similar in two tumor beds (Figs A2C, A2D), even though the pattern of distribution is different. In each microscopic field, % CA could be estimated by comparing the proportion of residual tumor bed area containing cancer (invasive or in situ). For example, if one half of the area in the field contained cancer, then % CA is 50%; one fifth is 20%, one tenth is 10%, one twentieth is 5%, and so on. To assign the overall value for % CA, we scanned across the defined residual tumor bed using the microscope and estimated an average of the readings for % CA in the cross-sectional area. The % CA in each microscopic field, and the estimated average for overall % CA, were assigned one of the following values: 0%, 1%, 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, or 90%. We next estimated the proportion of residual cancer that was in situ component (% CIS) as one of the following values: 0%, 1%, 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, or 100%. This applied the same methods to determine in situ component that are commonly used for identification of extensive intraductal carcinoma component in usual pathologic evaluation of breast cancer specimens. An example of the pathologic evaluation of the primary tumor bed is illustrated in Figure A3. Within the specimen slices there was an ill-defined macroscopic tumor bed that had greatest cross-sectional dimensions of 27 x 10 mm. The macroscopic tumor bed was entirely submitted as sections B1 to B3. Microscopic evaluation of those corresponding slides identified a single focus of residual cancer that measured 8 x 6 mm (d1 = 8 mm; d2 = 6 mm). Across the area of residual cancer there was average cancer cellularity of 20% (% CA = 20), of which 1% was in situ cancer (% CIS = 1). This particular case was node-negative, so LN = 0 and dmet = 0 mm. Additional examples with photomicrographs are linked to our web site (http://www.mdanderson.org/breastcancer_RCB).
Appendix 2: Detailed Statistical Methods Survival analysis. Initial analyses to evaluate the association of the four residual tumor measurements with DRFS were based on univariate Cox models that included each variable as a continuous covariate term untransformed in its original scale, and in a multivariate Cox model that included all untransformed terms. Covariate effects on distant relapse risk were evaluated in multivariate Cox proportional hazards analyses. None of the covariates exhibited significant deviations from the proportionality assumption or had time-dependent effects (Therneau TM, Grambsch PM: New York, NY, Springer-Verlag, 2000). The incremental contribution of the continuous RCB index was evaluated by comparing multivariate Cox models without or with a linear RCB term and its statistical significance was assessed based on the likelihood ratio test. Competing risks analysis of distant relapse rates. Death was treated as a censoring event in most analyses. However, death is a competing risk for distant metastasis because death precludes the occurrence of distant metastases. Competing risks do not interfere with the computation of the Kaplan-Meier estimators of survival probabilities but alter their interpretation in terms of the probability of failure for end points that are subject to competing risks (Gooley TE, Leisenring W, Crowley J, Storer BE: Stat Med 18:695-706, 1999). Similarly, competing risks do not invalidate the Cox proportional hazards model, but will affect the interpretation of the estimated hazard ratios (Lunn M, McNeil D: Biometrics 51:524-532, 1995). We used cumulative incidence estimators to estimate the overall probability of recurrence within RCB classes. In this analysis, death was treated as a competing risk, whereas local recurrence was treated as a censoring event since it does not preclude distant relapse. The 95% CIs of the relapse rates were calculated based on the Aalen estimate of asymptotic variance (Aalen O: Annals of Statistics 6:534-545, 1978) and a log-transformation of the cumulative hazard rate. Additional results from this analysis are provided in Appendix 4. Determining the functional form of relapse risk on RCB index. The functional form of the dependence of the 5-year distant relapse risk on the RCB index was determined through a univariate Cox proportional hazards model for DRFS having as the only covariate a penalized spline (p-spline) approximation of the RCB index with 2 df (Therneau TM, Grambsch PM: New York, NY, Springer-Verlag, 2000). The baseline cumulative hazard rate was estimated from the Cox model based on the Nelson-Aalen estimator and the predicted rate of distant relapse at 5 years was then obtained from the Breslow-type estimator of the survival function. Point-wise CIs of the survival estimate were calculated based on the Tsiatis variance estimates of the cumulative log- hazards (Therneau TM, Grambsch PM: New York, NY, Springer-Verlag, 2000). Determination of RCB cutoff points for defining residual disease classes associated with distinct prognosis. We determined thresholds to define four categories of residual disease associated with distinct prognosis: class RCB-0 included those with no traces of residual disease (RCB = 0; ie, those who achieved complete pathologic response); class RCB-I included those with minimal residual disease; class RCB-II those with moderate residual disease; and class RCB-III those with extensive residual disease. To determine the first cutoff point (between RCB-III and less residual disease), we fit a multivariate Cox regression model that included all clinical and demographic covariates (dichotomized age, clinical stage before treatment, hormone receptor status, and hormone treatment) and a dichotomous RCB factor based on cutoff points selected between the 5% and the 95% quantiles of the RCB distribution. The optimal cutoff point was selected as the quantile that maximized the profile log-likelihood of this model. A second cutoff point (between RCB-I and RCB-II) was determined similarly by maximizing the profile log-likelihood of a Cox model that included all clinical covariates and the first dichotomous RCB factor (ie, RCB-III v RCB-I/II). Validation of RCB as predictor of distant relapse. The significance of RCB index as an independent predictor of disease relapse was also assessed based on the separation probability, defined as the difference between the relapse probabilities for a patient in the group with the worst prognosis (RCB-III) and a patient in the group with the best prognosis (RCB-0; Altman DG, Royston P: Stat Med 19:453-473, 2000). The relapse probabilities were obtained from the Kaplan-Meier survival estimates and the CIs were based on the cumulative log-hazard estimate of the variance. To evaluate whether knowledge of a tumor's RCB class after chemotherapy adds new independent prognostic information to the revised yAJCC stage, we performed separate Kaplan-Meier analyses by RCB class within each AJCC stage stratum. The significance of the additional stratification provided by the RCB class was evaluated based on the log-rank test. We evaluated the accuracy of RCB-based model for prognosis of disease relapse by assessing its calibration and discrimination (Altman DG, Royston P: Stat Med 19:453-473, 2000). Calibration of the full multivariate Cox model was evaluated by comparing the predicted probabilities of distant relapse to the observed Kaplan-Meier survival estimates at 5 years in patient groups defined by the quintiles of predicted survival. We used bootstrap resampling with 300 replications to evaluate potential overoptimism in the predictions of RCB prognostic model, which refers to the bias introduced by "using the data twice", that is, for selecting cutoff points and also for evaluating the model's predictive accuracy (Schumacher M, Hollander N, Sauerbrei W: Stat Med 16:2813-2827, 1997; Harrell FE Jr: New York, Springer-Verlag, 2001). The bootstrap-estimated global shrinkage factor indicates the extent of overfitting (a shrinkage factor of 1 indicates no overfitting). The prognostic model was adjusted by multiplying the linear risk predictor of the Cox model by the above shrinkage factor, and was calibrated against the observed Kaplan-Meier survival estimates at 5 years in patient groups defined by the quintiles of predicted survival. The same estimated shrinkage factor was subsequently used to obtain unbiased estimates of relapse-free survival in the independent validation cohort of FAC-treated patients. Model discrimination was evaluated based on Harrell's concordance index, or c index, which is a generalized area under the receiver operating curve (AUC) for censored observations and is equal to the probability of concordance between the predicted probability of relapse and the relapse outcome (Harrell FE Jr: New York, Springer-Verlag, 2001). The concordance index was adjusted for bias using bootstrap resampling with 300 replications. The CI for the c index was obtained based on approximate normality using the variance estimate of the unadjusted index.
Appendix 3: Characterization of the Residual Cancer Burden Index
Metastatic contribution to residual disease. The contribution to residual disease from the primary tumor bed or the total tumor left at the primary site was estimated as the product of the average tumor diameter and the fraction of invasive disease in the primary tumor bed. The contribution of nodal metastases to the overall residual disease could be estimated from the number of positive lymph nodes and the size of each node. The metastatic term would be proportional to the sum of the diameters of all positive nodes. However, because only the diameter of the largest metastasis is routinely measured, we need to make an assumption about the size of the additional nodal metastases. If we assume that all metastases are similar in size, the metastatic contribution term would be equal to LN x dmet, where LN is the number of positive nodes. However, empirical observations suggest that additional nodal metastases are smaller than the next-largest metastasis, thus we can assume that each have (a x 100) % of the diameter of the next-largest metastases. Under these assumptions, the total metastatic contribution becomes dmet x (1 + a + a2 + + a(LN – 1)) = dmet x (1 – aLN) /(1 – a). Notice that for a = 1 the above term becomes dmet x LN. Figure A4 shows the dependence of the metastatic term on the number of positive nodes for different values of parameter a. Clearly, smaller values of a have the effect to dampen the dependence on the number of positive nodes. Appendix Table A3 (online only) presents the results from multivariate Cox models of DRFS that included interaction terms for different a. The results for a = 1, 0.75, 0.5 were qualitatively similar, and since a = approximately 0.75 corresponded to our empirical observations, we chose this value in the metastatic term definition. Additional characteristics of the RCB index and its components are provided in Figures A5 to A10.
Appendix 4: Analysis of Residual Cancer Burden in the T/FAC Cohort Competing risk analysis of probability of distant relapse. Four potential outcomes were possible for each patient in the study at any given time (t): (1) a patient experienced failure from distant relapse (metastasis) before or at t; (2) a patient experienced failure from local recurrence before or at t; (3) a patient died before or at t; (4) or a patient did not have distant relapse or local recurrence and was still alive at time t. The first outcome is the event of interest, whereas the second and third outcomes may represent competing risks. A patient who had the fourth outcome was censored at time t. Because presence of local recurrence does not fundamentally alter the probability of developing distant metastases, outcome (2) is not necessarily a competing risk for the event of distant relapse and therefore these observations can be censored. In the T/FAC-treated cohort, one patient had local recurrence within 5 years after initial biopsy and another one died in the same interval (Table 1). Overall, two patients died throughout the duration of follow-up. Therefore the number of observations with competing risks in the T/FAC cohort is too small to affect in any significant way the Kaplan-Meier estimated probabilities of relapse in the T/FAC group. The estimate of the probability of treatment failure among all patients under study should be consistent with the simple ratio estimate of the number of failures divided by the number of patients under study within each group defined by a covariate. For the T/FAC cohort, the probabilities of distant relapse estimated by the simple failure ratios within the four RCB classes were 5.4%, 2.4%, 15.8%, and 53.3% for RCB-0, RCB-I, RCB-II, and RCB-III, respectively. These are in very good agreement with those estimated by the corresponding unadjusted Kaplan-Meier estimates of survival probabilities: 5.4%, 2.4%, 16.2%, and 53.6%, respectively. Therefore, for the T/FAC data set, competing risks do not affect the outcome of survival analysis due to the small frequency of such events. Figure A11 shows the cumulative incidence estimates of the probability or distant relapse and death for the four RCB classes in both data sets. These curves are the complements of the in Fugure 3A but now adjusted for the competing risk of death. It is interesting to point out that the death rate was 0 in the T/FAC treated group (only one death event by 5 years), but the risk of death at 5 years for the FAC group was independent of the RCB status: 0%, 0%, 4.8%, and 5.2% at 5 years and 5.0%, 6.2%, 4.8%, and 7.9% at 10 years for RCB-0, RCB-I, RCB-II, and RCB-III, respectively. Additional results from the analysis of the T/FAC cohort are shown in Appendix Figures A12 and A13 and in Appendix Table A4, online only.
Appendix 5 In the FAC-treated validation cohort, one patient experienced failure due to local recurrence at 110 months and for our analysis this patient was censored (Figs A14 to A17). However, there were five deaths within 5 years (4%) and eight deaths within 10 years (6%), which could bias the Kaplan-Meier estimates of the probability of distant relapse for this group of patients (Appendix Table A4, online only). In fact, for FAC-treated patients within groups defined by RCB category the simple ratio estimates of the probability of relapse were 0%, 0%, 22.2%, and 35.9% at 5 years, and 0%, 12.5%, 34.9%, and 51.3% at 10 years (Appendix Table A1). The corresponding probabilities derived from the Kaplan-Meier estimates of distant relapse-free survival were 0%, 0%, 22.2%, and 36.3% at 5 years and 0%, 21.6%, 36.2%, and 52.2% at 10 years. Therefore, the Kaplan-Meier probabilities appear to overestimate the probability of relapse at 10 years when there are a moderate number of deaths, which represent a competing risk.
Although all authors completed the disclosure declaration, the following authors or their immediate family members indicated a financial interest. No conflict exists for drugs or devices used in a study if they are not being evaluated as part of the investigation. For a detailed description of the disclosure categories, or for more information about ASCO's conflict of interest policy, please refer to the Author Disclosure Declaration and the Disclosures of Potential Conflicts of Interest section in Information for Contributors. Employment: Christos Hatzis, Nuvera Biosciences Inc Leadership: N/A Consultant: W. Fraser Symmans, Nuvera Biosciences Inc; Lajos Pusztai, Nuvera Biosciences Inc Stock: W. Fraser Symmans, Nuvera Biosciences Inc; Christos Hatzis, Nuvera Biosciences Inc; Lajos Pusztai, Nuvera Biosciences Inc Honoraria: N/A Research Funds: N/A Testimony: N/A Other: N/A
Conception and design: W. Fraser Symmans, Christos Hatzis Administrative support: Gabriel N. Hortobagyi Provision of study materials or patients: W. Fraser Symmans, Radhika Rajan, Henry M. Kuerer, Vicente Valero, Lina Assad, Anna Poniecka, Marjorie C. Green, Aman U. Buzdar, S. Eva Singletary, Gabriel N. Hortobagyi, Lajos Pusztai Collection and assembly of data: W. Fraser Symmans, Florentia Peintinger, Radhika Rajan, Lina Assad, Anna Poniecka, Bryan T.J. Hennessy, Lajos Pusztai Data analysis and interpretation: W. Fraser Symmans, Florentia Peintinger, Christos Hatzis, Lajos Pusztai Manuscript writing: W. Fraser Symmans, Florentia Peintinger, Christos Hatzis, Henry M. Kuerer, Gabriel N. Hortobagyi, Lajos Pusztai Final approval of manuscript: W. Fraser Symmans, Florentia Peintinger, Christos Hatzis, Radhika Rajan, Henry M. Kuerer, Vicente Valero, Lina Assad, Anna Poniecka, Bryan T.J. Hennessy, Marjorie C. Green, Aman U. Buzdar, S. Eva Singletary, Gabriel N. Hortobagyi, Lajos Pusztai
Published online ahead of print at www.jco.org on September 4, 2007. Supported by a research Grant No. DAMD17-02-1-0458 01 from Department of Defense Breast Cancer Research Program (W.F.S.) and the Nellie B. Connally Breast Cancer Research Fund. Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
1. Feldman LD, Hortobagyi GN, Buzdar AU, et al: Pathological assessment of response to induction chemotherapy in breast cancer. Cancer Res 46:2578-2581, 1986 2. Hortobagyi GN, Ames FC, Buzdar AU, et al: Management of stage III breast cancer with primary chemotherapy, surgery, and radiation therapy. Cancer 62:2507-2516, 1988[CrossRef][Medline] 3. Chevallier B, Roche H, Olivier JP, et al: Inflammatory breast cancer: Pilot study of intensive induction chemotherapy (FEC-HD) results in a high histologic response rate. J Clin Oncol 16:223-228, 1993[Medline] 4. Sataloff DM, Mason BA, Prestipino AJ, et al: Pathologic response to induction chemotherapy in locally advanced carcinoma of the breast: A determinant of outcome. J Am Coll Surg 180:297-304, 1995[Medline] 5. Honkoop AH, Pinedo HM, De Jong JS, et al: Effects of chemotherapy on pathologic and biologic characteristics of locally advanced breast cancer. Am J Clin Pathol 107:211-218, 1997[Medline] 6. Honkoop AH, van Diest PJ, de Jong JS, et al: Prognostic role of clinical, pathological and biological characteristics in patients with locally advanced breast cancer. Br J Cancer 77:621-626, 1998[Medline] 7. Bonadonna G, Valagussa P, Brambilla C, et al: Primary chemotherapy in operable breast cancer: Eight-year experience at the Milan Cancer Institute. J Clin Oncol 16:93-100, 1998 8. Fisher B, Bryant J, Wolmark N, et al: Effect of preoperative chemotherapy on the outcome of women with operable breast cancer. J Clin Oncol 16:2672-2685, 1998[Abstract] 9. 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 10. Ogston KN, Miller ID, Payne S, et al: A new histological grading system to assess response of breast cancers to primary chemotherapy: Prognostic significance and survival. Breast 12:320-327, 2003[CrossRef][Medline] 11. Bear HD, Anderson S, Smith RE, et al: Sequential preoperative or postoperative docetaxel added to preoperative doxorubicin plus cyclophosphamide for operable breast cancer: National Surgical Adjuvant Breast and Bowel Project Protocol B-27. J Clin Oncol 24:2019-2027, 2006 12. Kaufmann M, Hortobagyi GN, Goldhirsch A, et al: Recommendations from an international expert panel on the use of neoadjuvant (primary) systemic treatment of operable breast cancer: An update. J Clin Oncol 24:1940-1949, 2006 13. Kurosumi M: Significance of histopathological evaluation in primary therapy for breast cancer–recent trends in primary modality with pathological complete response (pCR) as endpoint. Breast Cancer 11:139-147, 2004[Medline] 14. Kuroi K, Toi M, Tsuda H, et al: Unargued issues on the pathological assessment of response in primary systemic therapy for breast cancer. Biomed Pharmacother 59:S387-S392, 2005[CrossRef] 15. von Minckwitz G, Raab G, Caputo A, et al: Doxorubicin with cyclophosphamide followed by docetaxel every 21 days compared with doxorubicin and docetaxel every 14 days as preoperative treatment in operable breast cancer: The GEPARDUO study of the German Breast Group. J Clin Oncol 23:2676-2685, 2005 17. Hennessy BT, Hortobagyi GN, Rouzier R, et al: Outcome after pathologic complete eradication of cytologically proven breast cancer axillary node metastases following primary chemotherapy. J Clin Oncol 23:9304-9311, 2005 18. Carey LA, Metzger R, Dees EC, et al: American Joint Committee on Cancer tumor-node-metastasis stage after neoadjuvant chemotherapy and breast cancer outcome. J Natl Cancer Inst 97:1137-1142, 2005 19. Rajan R, Poniecka A, Smith TL, et al: Change in tumor cellularity of breast carcinoma after neoadjuvant chemotherapy as a variable in the pathologic assessment of response. Cancer 100:1365-1373, 2004[CrossRef][Medline] 20. Thomas E, Holmes FA, Smith TL, et al: The use of alternate, non-cross-resistant adjuvant chemotherapy on the basis of pathologic response to a neoadjuvant doxorubicin-based regimen in women with operable breast cancer: Long-term results from a prospective randomized trial. J Clin Oncol 22:2294-2302, 2004 21. Jones RL, Lakhani SR, Ring AE, et al: Pathological complete response and residual DCIS following neoadjuvant chemotherapy for breast carcinoma. Br J Cancer 94:358-362, 2006[CrossRef][Medline] 22. Mazouni C, Peintinger F, Wan-Kau S, et al: Residual ductal carcinoma in situ in patients with complete eradication of invasive breast cancer after neoadjuvant chemotherapy does not adversely affect patient outcome. J Clin Oncol 25:2650-2655, 2007 23. Singletary SE, Allred C, Ashley P, et al: Revision of the American Joint Committee on Cancer staging system for breast cancer. J Clin Oncol 20:3628-3636, 2002 24. Green MC, Buzdar AU, Smith T, et al: Weekly paclitaxel improves pathologic complete remission in operable breast cancer when compared with paclitaxel once every 3 weeks. J Clin Oncol 23:5983-5992, 2005 25. Buzdar AU, Singletary SE, Theriault RL, et al: Prospective evaluation of paclitaxel versus combination chemotherapy with fluorouracil, doxorubicin, and cyclophosphamide as neoadjuvant therapy in patients with operable breast cancer. J Clin Oncol 17:3412-3417, 1999 26. Hortobagyi GN, Buzdar AU, Theriault RL, et al: Randomized trial of high-dose chemotherapy and blood cell autografts for high-risk primary breast carcinoma. J Natl Cancer Inst 92:225-233, 2000 27. Draper NR, Cox DR: Distributions and their transformation to normality. J R Statist Soc B 31:472-476, 1969 28. Altman DG, Royston P: What do we mean by validating a prognostic model? Stat Med 19:453-473, 2000[CrossRef][Medline] 29. Simon R: Roadmap for developing and validating therapeutically relevant genomic classifiers. J Clin Oncol 23:7332-7341, 2005 30. Justice AC, Covinsky KE, Berlin JA: Assessing the generalizability of prognostic information. Ann Intern Med 130:515-524, 1999 31. Schumacher M, Hollander N, Sauerbrei W: Resampling and cross-validation techniques: A tool to reduce bias caused by model building? Stat Med 16:2813-2827, 1997[CrossRef][Medline] 32. Harrell FE Jr: Regression modelling strategies: With applications to linear models, logistic regression, and survival analysis. New York, Springer-Verlag, 2001 33. Berry DA, Cirrincione C, Henderson IC, et al: Estrogen-receptor status and outcomes of modern chemotherapy for patients with node-positive breast cancer. JAMA 295:1658-1667, 2006 34. Swain SM: A step in the right direction. J Clin Oncol 24:3717-3718, 2006 35. Kattan MW: Judging new markers by their ability to improve predictive accuracy. J Natl Cancer Inst 95:634-635, 2003 36. Fleming TR: Surrogate endpoints and FDA's accelerated approval process. Health Aff (Millwood) 24:67-78, 2005 Submitted January 5, 2007; accepted June 29, 2007.
Related Correspondence
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2007 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
TOP
ABSTRACT
INTRODUCTION
s = 0.837) compared to the primary term (
