In this national wide, retrospective cohort study with propensity score matched (PSM), the data was extracted from the National Health Insurance Research Database (NHIRD), which is a database constructed by 95% of residents in Taiwan since 1995 through the National health care insurance (NHI) system. Disease profiles are based on the International Classification of Diseases, 9th Revision, and Clinical Modification (ICD-9-CM) systems; the NHIRD provides information including demographics, outpatient visits, and hospitalizations with dates, prescriptions codes, diagnostic codes, laboratory tests and interventional procedure codes, and medical costs. De-identification was done for the protection of personal privacy.

We also extract the major illness registry data, also called the catastrophic illness registry, which is a certification for patients who were diagnosed with some severe, chronic, or fatal disease including cancer, diabetes mellitus, major injury, and SLE. The certification provides a discount on admission and medical charge. The Longitudinal Health Insurance Database 2000 (LHID 2000) was also used in the comparator group without HT of this study; LHID 2000 has collected a population of 1 million which was randomly sampled from the beneficiaries’ registration files within the year 2000. The representative of gender and age distribution has been statistically confirmed between the LHID 2000 and the origin NHIRD data.

We identified our population as patients who have had at least three outpatient visits or one hospital admission for autoimmune thyroiditis (Hashimoto’s thyroiditis) (ICD9 code: 245.2) between 2003 and 2012. The index date for the corresponding matching was defined as the date of the first diagnosis of autoimmune thyroiditis, outpatient, or admission.

We also constructed a comparator group without HT sampled from the LHID 2000 data, which includes patients ever visited outpatient departments between 2005 and 2012. The index date of the comparator group without HT was defined as the first visit to the outpatient department each year. Those who ever had at least one outpatient visit plus one hospitalization under the diagnosis of disorders of the thyroid gland (ICD9 codes 240–246: 240 for simple and unspecified goiter, 241 for nontoxic nodular goiter, 242 for thyrotoxicosis with or without goiter, 243 for congenital hypothyroidism, 244 for acquired hypothyroidism, 245 for thyroiditis, 246 for other disorders of the thyroid) between 1997 and 2013 were excluded.

We excluded the data of index date which were unmatched among the study and comparator group without HT (2003–2004), SLE (ICD9 code: 710.0) diagnosed before the index date and death during follow-up.

Overall, we extracted a total population of 17,978 cases with HT between the years 2005 and 2012 as our study group; and the comparator group without HT contained 783,345 patients without any disorders of the thyroid gland. To reduce the confounding bias, propensity score matching (PSM) was used, which was estimated by logistic regression modeling. Predictors involved index date, gender, and selected co-morbidities. The 1:2 matched comparator shares the same propensity score as the exposure group.

The primary outcome of the study was SLE occurrence, which was defined as patients who were diagnosed with SLE (ICD9 code: 710.0) and were identified as having “major illness” according to the NHI document for ensuring only correctly diagnosed patients were included. The follow-up started on the respective index date for different individuals until SLE was diagnosed or withdrawn from NHI due to any cause such as death, leaving, loss of data, or end of the study (December 2013), whichever occurred first. Relevant data of background variation including gender, age, urbanization, low income, length of hospital stays, times of outpatient department visits, medication control, and co-morbidities were also extracted and listed in Table 1.

To minimize surveillance bias, patients who were diagnosed with SLE during the period of 2 years before the index date were excluded. Besides, due to the chronic and latent nature of SLE, patients who were diagnosed with SLE whose follow-up time was less than 3 months and 6 months were excluded, respectively, in 2 scenarios, which were conducted to increase the accuracy, and in the background variations, groups of 3 months before the index date and 3 and 6 months after the index date were also corrected with length of hospital stays and times of outpatient department visits (Table 1).

Comorbidities were captured by tracing all the ambulatory care and admission records in the NHI database within 1 previous year of the index date and have had at least three outpatient visits or one hospital admission. We analyzed autoimmune disorders including rheumatoid arthritis (RA, ICD9 code: 714.0), Sjögren’s syndrome (SS, ICD9 code: 710.2), systemic sclerosis (SSc, ICD9 code: 710.1), and vasculitis (ICD9 code: 433.0) that not rarely occur with SLE. Other common comorbidities such as hypertension (ICD9 codes: 401–405), diabetes mellitus (ICD9 code: 250), hyperlipidemia (ICD9 codes: 272.0–272.4), coronary artery disease (ICD9 codes: 410–414), osteoporosis (ICD9 code: 733), cerebral vascular accident (ICD9 codes: 430–438), chronic obstructive pulmonary disease (COPD)/asthma (ICD9 codes: 490–496), chronic kidney disease (CKD, ICD9 code: 585), chronic liver diseases (ICD9 codes: 571, 573), pancreatitis (ICD9 codes: 577.0, 577.1), affective psychosis (ICD9 code: 296), ankylosing spondylitis (ICD9 code: 720.0), inflammatory bowel disease (ICD9 codes: 555–556), HIV infection (ICD9 codes: 042–044, V08), autoimmune hemolytic anemia (AIHA) (ICD9 code: 283.0), and idiopathic thrombocytopenic purpura (ITP) (ICD9 code: 287.3) were also included in the study. Baseline treatment of HT including (1) no drug admiration, (2) anti-thyroid medication (carbimazole, propylthiouracil, methimazole)/eltroxin only, and (3) HCQ/corticosteroid+/− anti-thyroid medication/eltroxin was also analyzed, and all treatments were given within 6 months after diagnosis. Besides, hyperthyroidism (ICD9 code: 242) and hypothyroidism (ICD9 codes: 243, 244) diagnosed before the index date were separately analyzed by multivariable statistical analysis, which was listed in Tables 5, 6, and 7.

To compare and increase the similarities between our exposure group of HT and the comparator group without HT, the chi-square (χ2) tests and the two-tailed T test was used for the baseline demographic characteristics such as gender, age, urbanization, income level, admission duration, and comorbidities. Time-to-event analysis was conducted based on the index date defined as the fixed time point (January 2005) for every participation. All participants were followed up from their respective index date until the occurrence of SLE, until withdrawal, or until the end of 2013, whichever occurred first.

We also constructed a multivariable Cox proportional hazard model to estimate the hazard ratios (HRs) and 95% confidence intervals (CIs) for the SLE incidence. In the 1:20 age- and gender-matched population, 3 models were conducted. The first would be the model of HT alone, which also analyzed other thyroid disorders. For the second model, hyperthyroid and hypothyroid disorders were excluded. Model 3 contains HT with demographic variables, medical utilization, and comorbidities, and for the 1:2 PSM population, models A and B were constructed by controlling variables such as HT, hyperthyroid, and hypothyroid disorders. All the data and statistics were processed and analyzed by the Statistics Analysis System (SAS) software version 9.3 (SAS Institute, Inc., Cary, NC), and a p-value less than 0.05 was considered to indicate statistical significance.

To test the reliability of our study results, we established 4 sensitivity analysis scenarios including SLE medication treatment and the exclusion of autoimmune thyroiditis with other autoimmune diseases. The SLE treatment was identified as systemic corticosteroids or disease-modifying anti-rheumatic drugs (DMARDs) (including hydroxychloroquine (HCQ) or azathioprine) within 6 months after the first diagnosis of SLE. In scenarios 1–3, adjusted hazard ratio (aHR) was analyzed based on a different definition of SLE event. The main finding without medication treatment analysis would be scenario 1, systemic corticosteroids or DMARDs treatment were brought into scenario 2, and systemic corticosteroids were excluded in scenario 3. Scenario 4 modified the exclusion criteria and exclusion of the patients with rheumatic arthritis (RA), Sjögren’s syndrome (SS), systematic sclerosis (SSc), vasculitis, ankylosing spondylitis (AS), and inflammatory bowel disease (IBD) at baseline; hence, the autoimmune thyroiditis accompanied with other autoimmune diseases could be ruled out. The sensitivity analysis scenarios were listed in Tables 8, 9, and 10.