Specific calcification patterns on imaging have long been recognized as a characteristic of benign or malignant lesions and recently have been associated with an increase in future breast cancer risk [1]. Multiple types of mammographic calcifications have been associated with breast cancer risk including calcifications identified by radiologists during clinical interpretation and subsequently associated with a false-positive mammogram [initial Breast Imaging Reporting and Data System (BI-RADS) score of 3, 4, 5, or 0 for which breast cancer was not detected after recall assessment] [1], microcalcifications (calcification clusters with a malignant morphology) associated with a negative/benign mammogram assessment [2, 3], and any type of mammographic calcification reported during clinical interpretation [4].

Mammographic calcifications have been associated with subsequent diagnosis of invasive breast cancer and ductal carcinoma in situ (DCIS) with a similar twofold increase in risk associations [3, 4]. In addition, mammographic calcifications have been included in risk prediction models and are associated with short- and long-term breast cancer risk among premenopausal and postmenopausal women [3,4,5]. Whether calcifications increase advanced invasive breast cancer risk is not known. Advanced breast cancer is a surrogate for breast cancer mortality [6] and has increasingly become an outcome examined when evaluating breast imaging effectiveness and risk [7,8,9,10]. For example, cancer detection with supplemental screening ultrasound has been shown to be higher among women with dense (heterogeneously or extremely dense) breasts and high advanced cancer risk compared to high invasive cancer risk [11].

The primary study goal was to evaluate whether mammographic calcifications noted by a radiologist in a mammography report with a BI-RADS initial or final assessment of negative or benign are a marker of advanced invasive breast cancer risk and whether associations between calcifications and advanced cancer risk vary by breast density, body mass index (BMI) and menopausal status using data from facilities in the Breast Cancer Surveillance Consortium (BCSC). As a secondary aim, we examined the association of mammographic calcifications with non-advanced invasive breast cancer.

Methods.

Data were from 113 facilities participating in one of four BCSC breast imaging registries that collected calcification data: San Francisco Mammography Registry, Carolina Mammography Registry, Vermont Breast Cancer Surveillance System, and Kaiser Permanente Washington system (https://www.bcsc-research.org/about/sites). We prospectively collected women’s characteristics and mammography information from radiology facilities. Breast cancer diagnoses and tumor characteristics were obtained by linking women to pathology databases; regional Surveillance, Epidemiology, and End Results programs; and state tumor registries. Deaths were obtained by linking to state death records. Registries and a central Statistical Coordinating Center (SCC) received Institutional Review Board approval for active or passive consenting processes or a waiver of consent to enroll participants, link data, and perform analyses. All procedures were Health Insurance Portability and Accountability Act compliant, and registries and the SCC received a Federal Certificate of Confidentiality and other protections for the identities of women, physicians, and facilities.

This study included screening mammograms with a negative/benign initial or final assessment conducted from January 1996 through December 2019 among women aged 40–74 years. Screening mammograms were defined using the BCSC standard definition, which is based on a radiologist’s report of screening indication and excludes mammograms in women with a history of breast cancer or a prior mammogram within 9 months [12]. Only examinations with an initial assessment (based on screening views only) of BI-RADS 1 (negative), 2 (benign finding), or 0 (needs additional imaging) with a final assessment after any diagnostic work-up of BI-RADS 1 or 2 were included to avoid inclusion of cancers directly related with the calcification (N = 177546 excluded). Women were followed from 3 months after the mammogram to the earliest of the following: breast cancer diagnosis (advanced or non-advanced cancer, DCIS), death, disenrollment, end of complete cancer capture, or 5 years after the screen. Screening mammograms with less than 3 months of follow-up time (N = 108223) and examinations with cancers diagnosed within 3 months of screening (N = 16816) were excluded. Thus, we identified 3,710,313 screening mammograms with a negative/benign final assessment among 991,911 women (Supplemental Fig. 1).

We collected demographic and breast health history information from self-administered surveys at the time of screening and/or from electronic health records. Women self-reported race and ethnicity separately reported as the following categories: Hispanic/Latinx and non-Hispanic/Latinx African American/Black, Asian/Pacific Islander, White, Other/multiracial (non-Hispanic/Latinx Native American/Alaskan Native, or with two or more reported races, or other).

Radiologists (N = 932) categorized breast density during clinical interpretation using BI-RADS [13] breast density categories (Table 1) with a single radiologist providing an interpretation per exam. Mammographic calcifications were reported during clinical interpretation by radiologists and recorded in clinical electronic radiology systems.

Postmenopausal women were those with both ovaries removed, whose periods had stopped naturally, who currently used postmenopausal hormone therapy, or who were age 60 or older [14]. Premenopausal women reported a period within the last 180 days or did not meet one of the postmenopausal criteria and used birth control hormones. If a woman did not meet any of these criteria, then age at screen was used to classify a woman as postmenopausal (age ≥ 52) or premenopausal (age < 52) [14,15,16,17]. BMI was categorized as < 18.5 kg/m2 = underweight, 18.5–24.9 kg/m2 = normal weight, 25.0–29.9 kg/m2 = overweight, 30.0–34.9 kg/m2 = obese I, and ≥ 35.0 kg/m2 = obese II/III [18].

The primary outcome was diagnosis of advanced invasive breast cancer, defined as prognostic pathologic stage II or higher [6]. We classified American Joint Committee on Cancer 8th edition prognostic pathologic stage [19] using anatomic staging elements, tumor grade, and estrogen, progesterone, and human epidermal growth factor receptor status. If prognostic stage variables were missing (34%), we used anatomic stage IIB or higher (27%) or summary stage or other information (6%) to classify as advanced cancer [6]. Advanced cancer status was imputed for the remaining 1.5% of screens using an imputation model including tumor characteristics. The secondary outcome was prognostic pathologic stage I (non-advanced cancer).

Patient characteristics were summarized across advanced and non-advanced breast cancer stage, and the presence of calcifications (no, yes). Multiple imputation using fully conditional specification methods [20] was used to impute 108,207 (2.9%) missing values of race/ethnicity, 287,956 (7.8%) missing values of time since last mammogram, 357,084 (9.6%) missing values of breast density, 431,556 (11.6%) missing values of first-degree family history of breast cancer, and 1,649,968 (44.5%) missing values in BMI in 45 imputed datasets [21]. Imputation models included the characteristics, time to event, and the Nelson-Aalen estimator [20], weighted by the inverse number of mammograms per woman [22]. Prevalence of calcifications and variances were estimated for each imputed dataset and combination of menopausal status, breast density, and BMI. Prevalences were averaged across the imputed datasets, and Rubin’s Rule [23] was used to compute pooled standard errors to estimate Wald-type 95% confidence intervals.

To obtain unadjusted cumulative incidence functions (CIF) and standard errors, one observation per woman was randomly chosen within each imputed dataset. The CIF for each tumor type and imputed dataset was estimated using SAS PROC LIFETEST, subdivided by presence of calcifications, menopausal status, breast density and BMI category; the other tumor type and DCIS were considered competing risks. Five-year risks were averaged across the imputed datasets, and Rubin’s Rule [23] was used to compute pooled standard errors and estimate Wald-type 95% confidence intervals.

To estimate the association between advanced and non-advanced cancer risk and presence of calcifications, we estimated adjusted hazard ratios (HRs) based on Fine and Gray subdistribution hazard models accounting for competing risks of the other tumor outcomes. Models included interactions between presence of calcifications and menopausal status, BMI, breast density, and initial assessment (BI-RADS 1,2 vs BI-RADS 0) depending on the HRs being estimated. All models adjusted for age at screen (quadratic), race/ethnicity, family history of breast cancer, history of benign biopsy, and time since last mammogram, and were stratified by BCSC registry. A robust sandwich variance estimator and inverse-weighting by the number of screening mammograms per woman were used to account for multiple screens per woman [22, 24]. Results from the 45 imputed datasets were combined using PROC MIANALYZE in SAS.

To evaluate whether the association between presence of calcifications and advanced cancer risk varied over time, we refit the survival model with time-varying indicators of the presence of calcifications within three-time intervals; < 1 year, 1–3 years, and 3–5 years.

Data were analyzed using R version 4.0.4 (R Foundation for Statistical Computing, Vienna, Austria) and SAS version 9.4 (SAS Institute, Cary, NC). Two-sided alpha = 0.05 was used to determine statistical significance.