This nationwide population-based cohort included women with DCIS as the primary diagnosis (International Classification of Disease, 10^th revision [ICD-10]: D05). The enrollment period was between January 2007 and December 2021, and a 2-year wash-out period was used for any previous malignant disease. The HIRA collects data about general healthcare services, such as diagnoses, medical treatment, and medication prescriptions. Patients who redeemed the cost of at least two prescriptions after breast cancer diagnosis were defined as tamoxifen users (prescription code: 234501ATB and 234502ATB). Osteoporosis diagnosis was based on ICD-10 codes (M80, 81, and 82) and concomitant prescription data of osteoporosis that included risedronate, ibandronate, etidronate, pamidronate, alendronate, zoledronic acid, and denosumab ( Supplementary Table 1 ).

In the screening phase, we extracted the data of women aged ≥20 years in the NHIS database. The screening procedure was identical to that used in a previous study (13). We included patients diagnosed with DCIS between January 2009 and December 2015 who had not visited a physician for any other malignancies (ICD-10 code: any C code) during the 2-year washout period between 2007 and 2008. To minimize misclassification errors, patients with DCIS were defined as patients with a surgery code within 1 year of receiving their DCIS diagnosis. The following surgery codes were considered: N7133, wide excision; N7134, wide excision of the axillary breast; N7136, wide excision with axillary surgery; N7137, wide excision without axillary surgery; N7138, total mastectomy with axillary surgery; and N7139, total mastectomy without axillary surgery. The date of enrollment was defined as the day of surgery for DCIS. The follow-up period was based on the date of enrollment. Enrolled patients were monitored for cataracts until 2021. We evaluated the risk of osteoporosis in patients with DCIS who received tamoxifen. The risk of developing osteoporosis was analyzed on the basis of only diagnosis or diagnosis and treatment. Women who had osteoporosis prior to DCIS and those who had been diagnosed within a year of DCIS diagnosis were excluded.

The primary endpoint of this study was the cumulative incidence of osteoporosis diagnosis and treatment after adjuvant tamoxifen administration in patients with DCIS. Secondary analysis included subgroup analyses according to age: <45, 45–55, and >55 years, representing pre-, peri-, and postmenopausal status.

The diagnostic data on the following comorbidities were also collected to determine the confounding variables: diabetes (ICD-10 codes: E10, E11, E12, E13, and E14), hypertension (I10), hyperlipidemia (E78), chronic obstructive pulmonary disease (COPD; J44), chronic kidney disease (CKD; N18), liver cirrhosis (K74 and K703), and heart failure (I50). Comorbidities were defined using these codes in the 2 years prior to the enrollment date.

The baseline demographic and clinical characteristics of the two groups were compared using a t-test and chi-square test. The cumulative cataract incidence rates in both groups were displayed using Kaplan–Meier curves and compared using the log-rank test. Cox proportional hazard models were used to determine the hazard ratios (HRs) and 95% confidence intervals (Cis) and ascertain the occurrence of cataracts after adjusting for confounding variables. We applied the enter method. Statistical significance was set at a two-sided p-value of less than 0.05. Randomization was performed using an algorithm in the SAS software program (version 9.4, SAS Institute, Cary, NC, USA). To minimize bias, an estimated propensity score was used to match patients with DCIS who received tamoxifen with those who did not. Each patient was assigned a propensity score reflecting the probability of receiving tamoxifen. This was calculated for each patient using logistic regression analysis with variables such as age, breast surgery, and comorbidities. A nearest-neighbor greedy algorithm was used to match patients using propensity scores without replacement. The matching algorithm first selected a patient who received tamoxifen and then selected a patient who did not receive tamoxifen with the closest propensity score to that of the first selected patient. Patients were matched in a 1:1 ratio within 0.1 caliper width. In addition, 1:1 propensity score matching was used to maximize the number of patients with breast cancer (14).

Patients with diagnostic codes for DCIS between 2009 and 2015 were first sorted from the HIRA database. Out of a total of 43,434 patients with DCIS, 12,032 who had undergone curative breast surgery within 1 year after the first diagnosis were included. Patients who had not undergone curative breast surgery within 1 year were considered previously diagnosed and counted as duplicates ( Figure 1 ). Patients who had co-existent diagnostic codes for invasive breast carcinoma or other types of malignancy were excluded. Further, 712 patients who were diagnosed with osteoporosis and prescribed treatment for osteoporosis before DCIS diagnosis were excluded from the main analysis. Among the remaining 4,654 patients, 2,970 were prescribed tamoxifen, and 1,684 were not prescribed any endocrine treatment. Propensity score matching was applied for age at diagnosis, chronic disorders, and type of breast surgery, as a surrogate for radiation therapy.

Before matching, the tamoxifen-treated and control groups consisted of 2,970 and 1,684 patients, respectively ( Table 1 ). Among them, 356 (7.6%) patients were diagnosed with osteoporosis and were taking osteoporosis medication. After matching, the tamoxifen-treated and control groups consisted of 1,577 patients each ( Table 1 ). Of the 3154 patients in the matched group, osteoporosis occurred in 250 (7.9%). Table 1 shows the characteristics of the two groups in terms of chronic disorders, breast surgery type, and age. Before matching, the median age of the patients was 49 years. Patients who received tamoxifen were less likely to have osteoporosis than the control group (6.4% [191/2,970] vs. 9.8% [165/1,684], p<0.001). The tamoxifen group was slightly older and more likely to receive breast-conserving surgeries. In addition, diabetes and hyperlipidemia were more prevalent in the tamoxifen group (p=0.015 and p=0.020, respectively). After matching for age, type of surgery, and chronic disorder, all other variables were well balanced between the tamoxifen and control group, except for osteoporosis. A total of 95 patients on tamoxifen developed osteoporosis (95/1577, 6%), whereas 155 (155/1577, 9.8%) who were not receiving endocrine therapy developed osteoporosis (p<0.001). Osteoporosis remained significantly less prevalent in the tamoxifen group ( Table 1 ).

Before matching, during the median follow-up period of 87 months, the cumulative incidence of osteoporosis was significantly lower in the tamoxifen group (before matching, p=0.0016, Figure 2A ). During the median follow-up of 88 months after matching, the cumulative incidence of osteoporosis was also significantly lower in the tamoxifen group than in the control group (after matching, p=0.0023, Figure 2B ). Before matching, in the univariate Cox proportional hazard analysis, women who were on tamoxifen had a significantly lower risk of osteoporosis than tamoxifen users ( Table 2 ; HR=0.715, 95% CI=0.580–0.881, p=0.0017). After adjusting for the type of surgery, age, and comorbidities, tamoxifen use was consistently associated with a lower risk of osteoporosis in the multivariable model (HR=0.690, 95% CI=0.559–0.851, p=0.0005). Along with tamoxifen use, age, hypertension, and hyperlipidemia were also associated with the development of osteoporosis from multivariable analysis ( Table 2 ). In the univariable analysis, these factors remained significant in the matched cohort as well. Only hyperlipidemia and age were consistently significant from multivariate analysis ( Table 2 , HR=1.488, 95% CI=1.129–1.963, p=0.0049 and HR=1.086, 95% CI=1.072–1.101, p<0.001, respectively).

We also carried out subgroup analysis by patient age. Since the HIRA registry does not provide information on the menopausal status, we subdivided the patients into three groups by age: <45, 45–55, and >55 years, representing the premenopausal, perimenopausal, and postmenopausal state, respectively. Supplementary Tables 2-4 show the differences in variables between the groups according to tamoxifen use and age, before and after matching. There were few incidences of osteoporosis in the premenopausal group: 24 out of 1,553 patients and 11 out of 1,056 patients ( Supplementary Table 2 , 1.5% before matching and 1.0% after matching, respectively). The incidence of osteoporosis regardless of tamoxifen use increased gradually in the perimenopausal and postmenopausal groups from 6.8% and 7.6% ( Supplementary Table 3 , 135/1,978 before matching and 99/1,302 after matching, respectively) to 17.5% and 17.6% ( Supplementary Table 4 , 197/1,123 before matching and 140/796 after matching, respectively), respectively. Age-dependent subgroup analysis showed that the protective effect of tamoxifen was robust in the older age group ( Figure 3 and Supplementary Table 5 , before matching and after matching, p=0.0009 and p=0.0015, respectively). The results of the univariate analysis showed that compared to other age groups, only the postmenopausal group showed a definite protective effect of tamoxifen ( Supplementary Figure 3 ; before matching, HR=0.68, 95% CI=0.514–0.9, p=0.007, Supplementary Figure 1 ; after matching, HR=0.638, 95% CI=0.453–0.899, p=0.01). In the cohorts before and after matching, multivariable analysis revealed that tamoxifen was not associated with osteoporosis in premenopausal women (<45 years old) ( Supplementary Figure 4 [before matching, multivariable], HR=1.334, 95% CI=0.545–3.262, p=0.528, Supplementary Figure 2 [after matching, multivariable], HR=0.705, 95% CI=0.199–2.494, p=0.587). In perimenopausal women, tamoxifen was significantly associated with a reduced risk of developing osteoporosis before matching ( Supplementary Figure 4 , HR=0.678, 95% CI=0.482–0.955, p=0.026). After adjusting for age and comorbidities, the protective effect persisted in perimenopausal women, but statistical significance was marginal ( Supplementary Figure 2 , HR=0.684, 95% CI=0.454–1.029, p=0.068). In postmenopausal women, the protective effect of tamoxifen was shown evidently both before and after matching. The multivariable analysis showed a consistent protective effect of tamoxifen before and after the matching ( Supplementary Figure 4 ; before matching, HR=0.657, 95% CI=0.495–0.872, p=0.004; Supplementary Figure 2 ; after matching, HR=0.637, 95% CI=0.452–0.898, p=0.010, respectively). There was a 35% reduction in the risk of developing osteoporosis when the postmenopausal population was treated with tamoxifen. Age was the only significant factor in both the univariable and multivariable models.