This retrospective observation cohort study used nationally representative data collected in the Korean National Health Insurance (NHI) claims database for a 12-year period from January 2009 to December 2018. In South Korea, it is mandatory for all residents to be affiliated to the national health insurance system, the NHI, which covers 97% of the population, except some foreigners who are ineligible for National Health Insurance. The database includes all medical claims provided by the Korean NHI Medical Services and the Korean Medical Aid program. The NHI database includes all medical data including: personal information, date of disease registry, diagnostic codes, procedure information, and prescription information. The disease codes specified in the International Classification of Disease, 10th revision (ICD-10), were used to record diagnoses in the NHI database. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies [17].

In this study, all individual information was anonymized prior to data processing to comply with the privacy guidelines of the Health Insurance Portability and Accountability Act in Korea. The study protocol was approved by the Institutional Review Board (IRB) of the Gangnam Severance Hospital (local IRB No. 3-2019-0425), which waived the need for informed consent based on the retrospective cohort design. This study was conducted in according with the ethical principles of the Declaration of Helsinki.

All female patients aged ≥ 50 years or older in the database were included in the initial screening stage. We included patients who had been diagnosed with breast cancer (ICD-10 code; C50) between 2009 and 2010, and who had not seen a physician for any type of cancer (ICD-10 code; C code) during the preceding 2-year period (2-year wash-out period prior to the study period). When extracting data among non-breast cancer patients from NHI database, massive data extraction was not permitted due to enormous data size. Inevitably, we chose senile cataract, a common benign disease, occurring in elderly patients. To select the generally healthy control group of subjects with conditions unrelated to the outcome of interest across the overall population, we extracted data of senile cataract (ICD-10 code; H25 and H26) under the same conditions as those for patients with breast cancer (2-year wash-out period between 2007 and 2008). In the breast cancer patients group, patients with a history of cataract were excluded from the analysis. The enrollment day for all subjects was the day of operation for breast cancer or cataract.

The main outcomes of interest were the risk of all types of dementia (ICD-10 code; F00.0, F00.1, F00.2, F00.9, F01.0, F01.1, F01.2, F01.3, F01.8, F01.9, F02.0, F02.1, F02.2, F02.3, F02.8, F03, G30.0, G30.1, G30.8, G30.9, G31.0, G31.01, G31.02, G31.03, G31.04, G31.08, and G31.1) in patients with breast cancer. These codes include Alzheimer’s dementia, vascular dementia, dementia with Lewy bodies, frontotemporal dementia, dementia in Parkinson disease, and other types of dementia. The Korean government provides additional financial support to patients with breast cancer confirmed by pathologic evaluation. Therefore, the diagnostic code for breast cancer is highly reliable. To minimize misclassification errors, patients with breast cancer (ICD-10 code; C50) were defined as those who underwent breast cancer surgery between 2009 and 2010 (Behavior code; N7136, wide excision with axillary surgery; N7137, wide excision without axillary surgery; N7138, total mastectomy with axillary surgery; N7139, total mastectomy without axillary surgery). In the cataract group, to minimize the possibility of misclassification error, we defined cataract patients (ICD-10 code: H25 and H26) as those who underwent cataract surgery from 2009 to 2010 (Behavior code; S511, phacoemulsification and intraocular lens insertion). When entering the diagnosis code of dementia into a patient’s medical record, a clinician in Korea usually makes a diagnosis based on both his/her diagnostic impressions and test results such as Mini-Mental State Examination [18], Clinical Dementia Rating [19], and Global Deterioration Scale [20]. Magnetic resonance imaging scan or neuropsychological tests are also performed in many cases, however, there is a limit to the information provided from NHI, so that information could not be obtained. In this background, in view of the medical environment in Korea with the relative ease of visiting a physician, to increase the diagnostic validity of neurodegenerative disease, dementia was defined as at least 10 contacts with a physician to obtain a corresponding diagnostic code between 2011 and 2018.

A total of 90,396 subjects were identified in the breast cancer group and 85,906 subjects were selected from the cataract group. Each study participant was followed up until 2018 for the onset of dementia. Surviving participants without events until December 31, 2018 were censored after this time-point. The overall risks of any type of dementia were compared with those of the control group, and risks from receipt of endocrine treatment or chemotherapy in breast cancer patients were also evaluated.

The confounding variable included several comorbidities and age. This study defined the confounding variables related to the development of dementia as follows: diabetes (ICD-10 code; E10, E11, E12, E13, and E14), hypertension (I10), hyperlipidemia (E78), chronic obstructive pulmonary disease (COPD; J44), chronic kidney disease (CKD; N18), liver cirrhosis (LC; K74 and K703), and heart failure (I50). Furthermore, we defined the presence of comorbidities as any diagnoses of the aforementioned codes in the 2 years preceding the enrollment date.

To reduce the selection bias, we estimated propensity scores for each of the all subjects to match the subjects who had breast cancer to those who had not. This was calculated for each subject with a logistic regression analysis, including variables of age and comorbidities. A nearest neighbor-matching algorithm with a “greedy” was performed to match the subjects using propensity score. We also used 1:1 propensity score matching to obtain the maximum number of patients with breast cancer and to minimize in the estimated results [21,22]. Randomization was performed using an algorithm within the SAS software program (ver. 9.4, SAS Institute Inc., Cary, NC). We explored the association of treatment modalities (chemotherapy, endocrine therapy, and use of aromatase inhibitor), comorbidities, and age with the incidence of dementia in the patients with breast cancer. In order to determine the effect of cataracts as a confounding factor for developing dementia, we also conduct sensitivity analysis the risk of dementia in the groups with concurrent cataract and breast versus with cataract only [23].

The primary endpoint was the overall risk of dementia in breast cancer patients compared to the age- and comorbidities-matched control group. The secondary endpoint was the difference in the incidence of dementia according to the treatment modalities for breast cancer. To compare the characteristics between the two groups, the Student’s t test was used for continuous variables, and the chi-square or Fisher exact test was used for categorical variables. The cumulative dementia incidences for two group were obtained using Kaplan-Meier curves and compared using the log-rank test. In a multivariable analysis, Cox proportional hazard models were used to determine the hazard ratio (HR) and 95% CI to investigate for the onset of dementia after adjustments for confounding variables. We applied the backward likelihood method (significance level for entering effect=0.05 and removing effects=0.05). A two-sided p-value less than 0.05 was considered to be statistically significant. Statistical analyses were performed using SAS software.

A total of 90,396 patients with breast cancer and 85,906 controls (patients with cataract) were assessed for inclusion in this study (Fig. 1). We excluded the patients without surgical behavior code (to select only new primary breast cancer patients), 20,168 patients with breast cancer, and 22,740 controls. After excluding patients who were diagnosed with cancer, cataract, and dementia prior to 2009 and those diagnosed with breast cancer and cataract simultaneously, there were 15,407 patients with breast cancer and 7,020 controls in the modified study sample. After matching, each group comprised 2,252 patients. Table 1 shows the between-group differences in the comorbidities and age of the patients. Before matching, the mean ages (±standard deviation), in patients with breast cancer and in controls were 53.73±7.59 and 73.39±7.78 years, respectively (Table 1). After matching, the mean ages in patients with breast cancer and in controls were 66.27±7.32 and 66.28±7.37 years, respectively (Table 1). Both groups were well-balanced with regard to comorbidities, such as diabetes, hypertension, hyperlipidemia, COPD, CKD, LC, heart failure, and age.

Before matching, the median follow-up time was 104.1±24.0 months. Patients with breast cancer had a significantly low incidence of dementia (p < 0.001) (Fig. 2A). After matching, with a medial follow-up time of 103.3±25.4 months, patients with breast cancer had a statistically low incidence of dementia compared to controls (p < 0.001) (Fig. 2B). Of the 4,073 patients with dementia (166 and 3,907 dementia patients with and without breast cancer, before matching, respectively) (Table 1), dementia subtypes were classified as follows: 3,592 Alzheimer’s dementia, 185 vascular dementia, two frontotemporal dementia, nine dementia in Parkinson’s disease, 12 dementia with Lewy bodies, and 273 other types of dementia.

In univariable Cox proportional hazard analyses of the unmatched cohorts, diabetes, hypertension, hyperlipidemia, COPD, CKD, heart failure, age, and history of breast cancer were significantly associated inversely with incidence of dementia (Table 2). In the multivariable analysis of the unmatched cohorts, history of breast cancer (HR, 0.036; 95% CI, 0.030 to 0.043; p < 0.001) (Table 2) was a significant negative determinant of dementia after adjustment for comorbidities and age. In the cohort analyses of the Cox proportional hazard model after matching for comorbidities and age, the results showed that diabetes, hypertension, COPD, CKD, age, and history of breast cancer were significantly associated with a lower incidence of dementia (HR, 0.098; 95% CI, 0.081 to 0.119; p < 0.001) (Table 2). In the multivariable analysis, history of breast cancer was a significant lower predictor of dementia in the matched cohorts (HR, 0.091; 95% CI, 0.075 to 0.111; p < 0.001) (Table 2).

The impact of treatment modalities on the incidence of dementia was evaluated in patients with breast cancer. Prior to matching, receiving chemotherapy (HR, 0.566; 95% CI, 0.417 to 0.767; p=0.002) (Fig. 3A); administration of endocrine therapy (HR, 1.981; 95% CI, 1.046 to 3.755; p=0.036) (Fig. 3A); the presence of diabetes, hypertension, hyperlipidemia, COPD, heart failure; and age were statistically significant factors for the occurrence of dementia among 15,407 breast cancer patients. The use of an aromatase inhibitor was not significantly associated with dementia (HR, 1.794; 95% CI, 0.665 to 4.839; p=0.248) (Fig. 3A). However, after adjustment for significant factors in the univariable analysis of unmatched cohort, receiving chemotherapy (HR, 1.371; 95% CI, 0.975 to 1.929; p=0.070; model 1: null model+chemotherapy) (Fig. 3B) and receiving endocrine therapy (HR, 1.386; 95% CI, 0.727 to 2.644; p=0.321; model 2: null model+endocrine therapy) (Fig. 3C) were not significantly associated with the risk for developing dementia. After matching, receiving chemotherapy (HR, 0.834; 95% CI, 0.575 to 1.209; p=0.337) (Fig. 3D) and endocrine therapy (HR, 1.549; 95% CI, 0.721 to 3.328; p=0.263) (Fig. 3D) did not significantly affect the incidence of dementia.

To rule out whether cataracts affect the occurrence of dementia, a total of 7,291 patients (7,020 cataract patients without breast cancer and 272 patients with coexisting breast cancer and cataract) were included in sensitivity analysis (Fig. 1). Before matching, the median age was 73.26±7.80 and 67.78±9.01 years in the groups without and with breast cancer, respectively (p < 0.001) (S1 Table). In the univariable analysis, a history of breast cancer (HR, 0.113; 95% CI, 0.076 to 0.170; p < 0.001) (S2 Table), hypertension, hyperlipidemia, COPD, and age were significant factor for dementia. After adjustment, the history of breast cancer was a significant negative predictor of dementia in the multivariable analysis (HR, 0.138; 95% CI, 0.092 to 0.206; p < 0.001) (S2 Table). After matching, the two groups were well-balanced in terms of participant characteristics (S1 Table). In the univariable Cox proportional hazard model, breast cancer (HR, 0.176; 95% CI, 0.113 to 0.274; p < 0.001) (S2 Table), hyperlipidemia, and age were significant determinant of the risk for developing dementia. In the multivariable analysis of matched cohort, breast cancer was a significant predictor of a lower incidence of dementia (HR, 0.167; 95% CI, 0.107 to 0.259; p < 0.001) (S2 Table).