Introduction
Ductal carcinoma in situ (DCIS) is a noninvasive breast cancer but some women will go on and develop invasive breast cancer [
1]. Our inability to elucidate which DCIS lesions will progress to invasion and which ones will remain indolent culminate in recommendations that women with DCIS undergo treatment. Most women will be treated by breast-conserving surgery (BCS) followed by the administration of whole breast radiotherapy (RT), which has been proven to lower the risk of local recurrence (LR) (DCIS or invasive) after BCS [
2]. Subset analyses from randomized trials demonstrate a similar relative (50%) reduction in LR risk with RT, but the absolute benefit from RT is not uniform for all patients. Some women will derive no or a very small absolute benefit from RT, resulting in unnecessary exposure to radiation and its potential toxicities (over-treatment), while in others the omission of RT may result in a higher risk of LR (and invasive LR) that might have been avoided by treatment (under-treatment) [
3]. To reduce over-treatment and under-treatment of DCIS, ascertainment of more precise estimates of individualized LR risk after BCS is desirable to help clinicians and patients more accurately assess the risks of LR with the potential absolute benefits of treatment.
The Oncotype DCIS score (DS) is a 12-gene expression assay based on the Oncotype DX Recurrence score [
4]. The DS reports a numeric value ranging from 0 to 100 and a categorical risk group: low risk (0–38), intermediate risk (39–54) and high risk (55–100). The ECOG-ACRIN E5194 (E5194) prospective cohort study initially reported the significance of the DS as an independent predictor of LR in selected women treated by BCS alone [
5‐
7]. More recently, the DS was validated as a predictor of LR in the Ontario population-based DCIS cohort [
8]. Multivariable analyses from both studies found that in addition to the DS, age at diagnosis and tumor size were also significant predictors of LR; however, current estimates of local and invasive LR risks associated with the DS do not adjust for these effects.
The objective of this analysis is to combine the data from the E5194 (with extended 12-year follow-up data) and Ontario cohorts to provide refined and more precise estimates of recurrence risk after BCS alone for DCIS. We performed a patient-specific meta-analysis [
9] to evaluate the impact of the DS alone, age at diagnosis and tumor size alone or integration of all three parameters, on the predicted 10-year risks of LR and invasive LR. In addition, we report 10-year local and invasive recurrence risk estimates for each DS risk group (low, intermediate, high) adjusting for the effects of age, tumor size and year of diagnosis to provide more accurate estimates of recurrence risk to aid treatment decision making following BCS for DCIS.
Discussion
This combined analysis provides refined estimates of the 10-year risk of LR and invasive LR of DCIS lesions treated by BCS alone. Integrating the effects of tumor size and age at diagnosis with the DS provides improved prediction and substantially better separation of low-risk from high-risk patients than either DS alone or information based on tumor size and age alone (without the DS).
Treatment decision making relies on estimating an individual’s risk of recurrence after BCS weighed against the potential benefits of treatment. Regression models estimate the relationships among individual variables and the likelihood of developing LR. Predictiveness curves combine the effects of risk modeling with the distribution of the risk-prediction covariates in the patient population [
13]. They illustrate the range and distribution of risk estimates within populations and provide a way to compare the performance of different models. We compared the performance of three models in predicting the risk of LR at 10 years after treatment by BCS alone with clear margins using data from the E5194 and Ontario cohorts; one model included the DS alone, one model was based on tumor size, age at diagnosis, and a third model combined all three parameters, adjusting for year of diagnosis. We found that integration of the effects of tumor size, age at diagnosis and year of diagnosis with the DS significantly improves LR risk prediction compared with estimates based on the DS alone or one based on tumor size, age and diagnosis year alone (without the DS).
The extent to which predicted risk estimates will influence treatment decision making relies on the thresholds of LR risk that determine if additional treatment is warranted. In this regard, models that best classify individuals at very low or high risk of recurrence have the greatest clinical utility. We evaluated the ability of each model to predict cases at very low risk of LR (defined as 10-year risk of LR ≤ 8%) or those at high risk (defined as 10-year risk of LR > 15%) after BCS alone. The model integrating tumor size and age at diagnosis with the DS performed better at both extremes. The DS/tumor size/age model identified 25.9% of the cohort as having a 10-year LR risk ≤ 8%. By comparison, the DS alone classified only 17.7% of cases while the model based on tumor size and age alone did not identify any cases as having a 10-year LR risk ≤ 8%. This suggests that the DS adjusted for the effects of tumor size and age at diagnosis can help reduce over-treatment by identifying significantly more women with a very low risk of LR after treatment by BCS alone for whom the benefit of RT would be extremely small. If the threshold for additional treatment is (a 10-year LR risk of) > 10%, then almost half the cases in the cohort (47%) would avoid additional treatment. These are mostly women ≥ 50 years of age with lesions ≤ 1 cm and DS ≤ 38 (68%) but they also include women ≥ 50 years of age with lesions 1–2.5 cm and DS ≤ 21 (23%) and women ≤ 50 years of age with lesions ≤ 1 cm and DS ≤ 17 (9%). Table
4 lists the average refined estimates of LR risk within DS groups and demonstrates the impact of each parameter on the 10-year LR risk by DS risk group. Women ≥ 50 years of age at diagnosis with lesions ≤ 2.5 cm and a low-risk DS or those aged < 50 years at diagnosis and tumor size ≤ 1 cm and a low-risk DS had average estimated 10-year risks of LR < 10.2% following treatment by BCS alone.
In addition, we found that integrating the impact of tumor size and age at diagnosis with the DS performed better at predicting cases with a high risk of LR (> 15%) after BCS alone where additional treatment would be warranted [21.1% compared to 18.4% classified by the DS alone and only 11% classified using tumor size and age without the DS (Fig.
2)]. There were women in all categories of age at diagnosis and lesion size with estimated LR > 15%.
This analysis has several strengths. It was derived from a prospective cohort and a population-based cohort treated by BCS alone with negative margins, includes large numbers of annotated samples with expert pathology assessment and DS molecular testing; therefore, the risk model is generalizable to similar patients in the general population. The baseline risks of LR and the HRs associated with the clinicopathological parameters were similar in the two cohorts (Table
2), indicating it is appropriate to apply a combined analysis. LR risks have declined over time [
14]; therefore, adjusting the risk estimates to reflect outcomes beyond year 2000 to provide more accurate prediction of expected outcomes of women treated in the current era.
This analysis has several limitations. The study population includes few women (N = 37) with tumors > 2.5 cm treated by BCS alone (6% of Ontario cohort); therefore, risk estimates in women with DCIS lesions > 2.5 cm should be interpreted with caution.
This analysis does not account for the impact of tamoxifen. Approximately one-third of the E5194 and 17% of those > 65 years in Ontario cohort received tamoxifen. Tamoxifen was used more frequently by patients diagnosed in 2000 or later (48.9%) than patients diagnosed before 2000 (15.0%). A sensitivity analysis of E5194 data was conducted to assess the effect of tamoxifen regression parameter estimates. A multivariate model was fit with the DS, tumor size, age, diagnosis year and a time-dependent indicator for tamoxifen use (Table S-2). The values of the HRs are similar to those in the main analysis, indicating that tamoxifen use did not greatly influence the estimates in this study.
In summary, this combined analysis provides refined estimates of the 10-year LR and invasive LR risk after treatment by BCS alone. Integrating the effects of tumor size and age at diagnosis with the DS provides improved prediction and better separation of very low-risk from high-risk patients (Table
6, Fig.
2). Specifically, these refined estimates identify a greater proportion of women with a 10-year LR risk ≤ 8% after BCS alone who could safely avoid additional treatment since the absolute benefit from additional interventions would be low and a greater proportion of women with a higher 10-year LR risk > 15% in whom efficacious treatments are needed to lower the risk of future recurrence. This can improve clinical decision making and the management of DCIS patients by helping clinicians and patients more accurately weigh risk of recurrence with the potential benefits of treatment.
Acknowledgement
This study is a joint collaboration by the ECOG-ACRIN Cancer Research Group (Peter J. O’Dwyer, MD and Mitchell D. Schnall, MD, PhD, Group Co-Chairs), Genomic Health, Inc., and the Institute for Clinical Evaluative Sciences and is supported by the National Cancer Institute of the National Institutes of Health under the following award numbers: CA180820, CA180794, CA180864, CA180795, CA189859, CA180828, CA180844 and the Canadian Cancer Society Research Institute (Grant Number 18491). Dr. Rakovitch is the LC Campbell Chair for Breast Cancer Research. We would like to acknowledge the contributions of pathologists Dr. A. Tuck, Dr. S. Robertson, Dr. S. Sengupta, Dr. M. Bonin, Dr. M.C. Chang, Dr. L. Elavathil and Dr. E. Slodkowska in the pathology review of the Ontario cohort. The content of this study is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health, nor does mention of trade names, commercial products or organizations imply endorsement by the US government. This study was supported by the Institute for Clinical Evaluative Sciences (ICES), which is funded by an annual grant from the Ontario Ministry of Health and Long-Term Care (MOHLTC). The opinions, results and conclusions reported in this paper are those of the authors and are independent from the funding sources. No endorsement by ICES or the Ontario MOHLTC is intended or should be inferred. These datasets were linked using unique encoded identifiers and analyzed at the Institute for Clinical Evaluative Sciences (ICES). Parts of this material are based on data and information compiled and provided by CIHI. However, the analyses, conclusions, opinions and statements expressed herein are those of the author, and not necessarily those of CIHI. Parts of this material are based on data and information provided by Cancer Care Ontario (CCO). The opinions, results, view and conclusions reported in this paper are those of the authors and do not necessarily reflect those of CCO. No endorsement by CCO is intended or should be inferred.