In this population- and register-based cohort study, we used the research database Breast Cancer Data Base Sweden (BCBaSe) as data source. BcBaSe is a database derived from the linkage of the National Quality Registry for Breast Cancer covering three healthcare regions in Sweden (Stockholm-Gotland, Uppsala-Örebro, North Region), which is corresponding to nearly 60% of Swedish population, with other national Registries of interest (the Prescribed Drug Registry, the National Patient Registry, and the Swedish cause of death registry).

Through BCBaSe, we identified all patients with non-metastatic breast cancer diagnosed at least one year after initiation of Prescribed Drug Registry (launched on July 1, 2005). Patients with bilateral breast cancer were also included. The date of breast cancer diagnosis was served as the index date.

We excluded men with breast cancer, patients with prior exposure to thyroid hormones (ATC-code: H03AA) before index date, and patients with prior cancer diagnosis (ICD-codes: C00-C14, C30-C32, C34, C50, C73) up to 10 years before index date.

The primary outcome was the frequency of new onset hypothyroidism in patients diagnosed with non-metastatic breast cancer. New onset hypothyroidism was defined as initiation of thyroid hormones (ATC-code: H03AA) from index date plus 90 days and onwards with at least two prescriptions irrespective of defined daily dose (DDD) without any prescription of antithyroid preparations (ATC-code: H03B) or surgical procedure to thyroid gland (ICD-10 codes: BAA40, BAA50, BAA60, and BAA99) at any time during the follow-up.

Secondary outcome was the identification of potential risk factors for development of hypothyroidism after breast cancer diagnosis with special interest in different oncological treatment strategies.

The following data were extracted from the BCBaSe: age at diagnosis, date at breast cancer diagnosis, region, menopausal status, relevant autoimmune comorbidities that could be associated with increased risk for hypothyroidism and had a prevalence in the cohort enabling statistical analyses (diabetes mellitus type 1, seropositive or seronegative rheumatoid arthritis, systemic lupus erythematosus), any exposure to medications that can lead to hypothyroidism (lithium with ATC-code N05AN01; amiodarone with ATC-code C01BD01) after index date, T status, N status, histology, ER (estrogen receptor)-status, PgR (progesterone-receptor)- status, tumor grade, Her2-status (with IHC or FISH); type of primary surgery, oncological treatment including chemotherapy, radiotherapy, endocrine therapy, and anti-HER2 treatment.

In terms of oncological treatment, the retrieved information from BCBaSe corresponds to planned treatment. Due to a considerably high percentage of missing data regarding HER2 status and anti-HER2 treatment, we did not include this treatment strategy in our analyses. Regarding radiotherapy, the available information through BCBaSe was whether patients were planned for radiotherapy or not and the planned irradiated target as breast/chest wall, regional lymph nodes, or both. According to the Swedish guidelines during the study period, radiation therapy to regional lymph nodes included axillary stations level II to IV but not internal mammary nodal stations.

Summary statistics are presented as frequencies and percentages for categorical variables and medians and interquartile ranges for continuous variables. Time to diagnosis of hypothyroidism was analyzed using Cox proportional hazards models stratified on region, with breast cancer diagnosis date representing the index date. We allowed for competing risks by censoring in the event of emigration or death. Maximum follow-up date was 31 December 2018. Treatment variables were categorical and included: radiation therapy, chemotherapy and endocrine therapy. Separate univariate Cox models were fitted to each of these treatment variables to estimate unadjusted effects. A multivariable Cox model was used to estimate adjusted effects, which in addition to the above treatment variables included: exposure to amiodarone and lithium (any DDD) post index date; diagnoses of type 1 diabetes, rheumatoid arthritis with rheumatic factor, other rheumatoid arthritis, and systemic lupus erythematosus pre- or post- index date; and diagnosis date and age at diagnosis (continuous variables modeled as restricted cubic splines using three pre-specified knots). Effects were presented as Hazard Ratios (HR) and 95% Confidence Intervals (CI). The proportional hazards assumption was assessed using unique term and global Chi-squared hypothesis tests (α = 0.05) as well as plots of the Schoenfeld residuals. No violation of proportional hazards was detected, with the exception of amiodarone. We, therefore, fitted an additional multivariable model with the same parameterization as above, but also including an interaction between amiodarone and time to event, modeled as a restricted cubic spline with three pre-specified knots. We, then, computed cumulative incidence curves for radiotherapy (stratified on region) accounting for competing risks of emigration and death. All statistical analyses were performed in R studio version 2022.07.2, using R version 4.2.2 [15], relying heavily on the packages survival, tidycmprsk, and the tidyverse suite [16,17,18]