Settings and study population
This retrospective cohort study used claims data obtained from the National Health Information Database of the National Health Insurance Service, a compulsory health insurance system covering the entire Korean population. The database includes information on demographics, healthcare utilization, vital statistics, and national health-screening results [13]. In addition, mammographic screening information was obtained from breast cancer screening program data, which were provided every two years for all women aged 40 years or older [14].
The initial cohort included 4,873,325 women aged 40–74 who underwent breast cancer screening between 2009 and 2010 and were followed up until the date of breast cancer diagnosis, date of death, or December 31, 2020, whichever came first (Fig. 1). If participants underwent mammographic screening more than once during the study period, we used data from the first screening. The following exclusion criteria were applied to select women included in the analysis: (1) participants who were diagnosed with any cancer or died within 3 months of screening to avoid the possibility of including screening-detected cancer; (2) participants whose mammographic screening results were suspicious abnormalities, highly suggestive of malignancy, or incomplete; and (3) participants whose main exposure information was missing (breast density or microcalcification). In total, 3,910,815 women who underwent negative screening were included in the analysis.
The study was approved by the Institutional Review Board (approval no. HYUIRB-202106-003-1). The requirement for informed consent was waived because the database was constructed after anonymization of individual identities. This work was supported by a National Research Foundation of Korea grant funded by the Korean government (MSIT) (grant no. 2021R1A2C1011958). This work was partly supported by Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No.2020-0-01,373, Artificial Intelligence Graduate School Program (Hanyang University)) and the research fund of Hanyang University (HY- 202,100,000,670,061).
Mammographic breast density and microcalcification
Information on mammographic breast density and microcalcification was extracted from the mammography screening results, which were read by trained radiologists at each screening center. The results of the Korean national breast cancer screening were recorded based on the BI-RADS 4th edition since 2009. Each breast’s density was assessed according to the BI-RADS 4th edition guidelines (< 25% glandular tissue, BI-RADS 1; 25–50% glandular tissue, BI-RADS 2; 50–75% glandular tissue, BI-RADS 3; > 75% glandular tissue, BI-RADS 4) and the presence and location of a mass, typically benign calcification, microcalcification, asymmetry, architectural distortion, and associated features, were recorded. Based on these findings, the final assessment of the BI-RADS category was recorded. Microcalcification was defined as a record of microcalcification in the mammography results.
Breast cancer cases
In Korea, the national health insurance policy has set a cost-sharing rate from 0 to 10% of the total medical expenditures for patients with high medical expenses, including severe diseases such as cancer, rare diseases, and incurable diseases [15]. Thus, patients with cancer have a special payment reduction program. Incident cancer cases are registered in this system, and special codes are given to cancer patients in the NHID. The ascertainment of cancer cases in our study was obtained from the healthcare utilization database using a combination of ICD-10 codes for breast cancer and catastrophic illness codes. The primary outcome was a breast cancer event, which was defined as a combination of the International Classification of Disease-10 code of invasive breast cancer (C50) or ductal carcinoma in situ (D05), in combination with the catastrophic illness code. This definition of cancer has a sensitivity of 98.1% compared with that of the Korean Central Cancer Registry, which contains a register of 90% of cancer cases nationwide [16].
Covariates
We considered the following variables for adjustment in the analysis: age at screening, the cut-off for BMI was applied according to the WHO Asia–Pacific recommendation, which defines overweight as BMI ≥ 23 kg/m2 and obesity as BMI ≥ 25 kg/m2 [17, 18], family history of breast cancer among first-degree relatives, number of delivered children, smoking (experience of smoking), alcohol consumption (drinking even once a week), physical activity (high-or moderate-intensity, or walking at least once a week), age at menarche, history of breastfeeding, and use of oral contraceptives. For postmenopausal women, the age at menopause, menopausal status, and history of hormone replacement therapy were also included as adjustment factors. Except for BMI, information on the above-mentioned covariates was collected using standardized questionnaires and self-reported by the participants at each screening center during health examinations and cancer screening.
Statistical analysis
The distribution of breast cancer events in our cohort was described with respect to the presence of microcalcifications (Additional file 1: Appendix 1). Descriptive statistics of baseline characteristics at the screening examination of study participants who developed breast cancer were compared using the chi-square test or Student’s t-test. The 5-year risk of developing breast cancer was calculated according to the BI-RADS density category for those with and without microcalcification, and menopausal status among all participants. The 5-year breast cancer risk was estimated as the number of cases diagnosed with breast cancer within 5 years of screening. Additionally, the Gray test was used to identify the equality of the cumulative incidence functions between the two groups. To quantify the association between microcalcification and the risk of breast cancer, Cox proportional hazard regression (HR) analysis was used to model the time from screening to breast cancer diagnosis with adjustment for other covariates. The assumption of proportional hazards was examined using Kaplan–Meier curves, and parallel lines of the log–log survival distribution function were identified. In the Cox regression model, breast cancer events were the primary outcomes. All participants were followed up until the date of any cancer diagnosis, including breast cancer diagnosis, date of death, or December 31, 2020, which ever came first. The censored cases were those who did not develop breast cancer (including other types of cancer development or death) or were alive until December 31, 2020. To quantify the independent association between microcalcification and breast cancer risk, analyses were performed with and without adjustments for breast density together with adjustment for other covariates. The analysis was conducted on the total population and further stratified by menopausal status. Finally, to quantify the joint associations between breast density, microcalcification, and breast cancer risk, the participants were classified into a combination of BI-RADS category and the presence of microcalcification, and the HR was presented with women with BI-RADS density category 1 and no microcalcification as the reference group. In addition, we assessed the effect of microcalcification on breast cancer risk by breast density category and then assessed the significant interaction of these two factors on breast cancer risk using the extra-sum of square F test. Additionally, to evaluate the trend of breast density categories across the development of breast cancer and the comparison by the presence of microcalcification, we obtained the p-value for trend from the asymptotic test to evaluate the trend between the exposure and the outcome. All reported p-values were two-sided with a type I error (α < 0.05) and were considered statistically significant. Statistical analyses were performed using the SAS statistical software (version 9.4; SAS Institute, Cary, NC, USA).