21-Gene Assay and Breast Cancer Mortality in Ductal Carcinoma in Situ

Eileen Rakovitch, MD, MSc, FRCPC; Rinku Sutradhar, PhD; Sharon Nofech-Mozes, MD, FRCPC; Sumei Gu, MSc; Cindy Fong, BSc, CCRP; Wedad Hanna, MD, FRCPC; Lawrence Paszat, MD, MSc, FRCPC

Disclosures

J Natl Cancer Inst. 2021;113(5):572-579. 

In This Article

Methods

Population Cohort

The Ontario DCIS Cohort has been previously described.[15] To identify all women diagnosed with pure DCIS treated with BCS from 1994 to 2003, we reviewed 129 140 breast pathology reports obtained from the Ontario Cancer Registry (OCR). Personal identifiers were removed to protect patient confidentiality. Lesions less than 2 mm (insufficient RNA) and individuals aged older than 75 years (because of statistically significant competing risks of death) were excluded. Expert pathology review confirmed the absence of invasion and provided standardized assessment of pathological features for every case. Tumor size was abstracted when present or imputed as a function of the largest focus of DCIS on a given slide and the number of blocks involved.[15]

This study was approved by the institutional review board at Sunnybrook Health Sciences Centre, Toronto, Canada. This is a population-based retrospective analysis. All personal identifiers for each case in this population cohort were removed. This study was facilitated through ICES, a not-for-profit research institute, which is named as a prescribed entity in Section 45 of PHIPA (Personal Health Information Protection Act Regulation 329/04, Section 18), which allows access and utilization of administrative data for research purposes with a waived requirement for consent.

Multigene Expression Assay

Details of the assay have been previously described.[18] RNA was extracted from 30 μm sections if DCIS measured 5.5 mm or more or from 60 μm sections if DCIS measured less than 5.5 mm. The RS is scaled as a continuous variable from 0 to 100 and in 2 risk categories: low risk (≤25) and high risk (>25).[19] The DS is scaled as a continuous variable from 0 to 100 or in 3 risk categories: low risk (<39), intermediate risk (39–54), and high risk (≥55).[13,14]

Treatment

Methods used to identify treatments and outcomes have been reported.[15] Each case in the cohort was linked by deterministic linkage to administrative databases of physician billings, the Registered Persons Database (vital statistics), the OCR, and the Canadian Institute of Health Information (CIHI) databases of hospital discharge summaries. Tamoxifen utilization in the cohort during this time interval was limited (<15%).

Local Recurrence or Contralateral Breast Cancer

Breast surgical procedures (and laterality) performed 6 months or more after initial diagnosis were ascertained from CIHI. Pathological diagnosis was determined from the pathology report when available or from the International Classification of Diseases (ICD) version diagnostic code (DCIS = ICD10 D05; invasive = ICD10 C50), from CIHI or OCR.

Cause of Death

We linked the cohort to the Registered Persons Database and OCR to identify date and cause of death. For each breast cancer death identified, we reviewed the records in CIHI and OCR to confirm the absence of a second (nonbreast) malignant diagnosis and to confirm that the clinical course of each case was consistent with metastatic adenocarcinoma of the breast.

Statistical Analysis

Risk of breast cancer death was examined in relation to patient and tumor characteristics, with the competing risk of death from other causes.[21] Modeling using a Fine-Gray approach[22] evaluated the relationship between independent variables and breast cancer death and other causes of death (competing event). Assumptions of the Fine-Gray model have been verified. Time to death was measured from the time of initial treatment of DCIS. Baseline covariates included age at diagnosis (50 years or younger or older than 50 years), margin status, nuclear grade, presence of multifocality, tumor size, RS, and post-BCS RT as previously defined.[14,15] To account for systematic differences between women treated with and without RT, a propensity score was calculated for each patient, which was then used to adjust the models.[23] Characteristics that may confound the relationship between RT and mortality were incorporated into the propensity score, similar to our prior work; this included tumor size, subtype, margin status, year of diagnosis, multifocality, socioeconomic status, RS, age at diagnosis, and nuclear grade. Because older women are more likely to die from comorbid conditions, and may derive less benefit from treatment, interaction terms between RS and treatment and RS and age were included (a priori) into the final multivariable models. Additional models were implemented using the main effects listed above and interactions between RS and treatment, nuclear grade, and multifocality, separately. Estimates of risk for breast cancer mortality and for nonbreast cancer mortality, along with 95% confidence intervals (CIs), were obtained and illustrated using the cumulative incidence function approach.[24] All analyses were performed using SAS 9.3 for Unix (SAS Institute, Cary, NC, USA) and R for Unix.

Statistical tests were 2-sided, and a P value less than .05 was considered statistically significant.

processing....