Data source

Taiwan’s single-payer National Health Insurance (NHI) program was established in 1995. This study used the Longitudinal Health Insurance Database, which contains the deidentified and encrypted medical claims data (for both inpatient and outpatient appointments) of 1 million beneficiaries randomly selected from the National Health Insurance Research Database (NHIRD). All diagnoses were made and recorded according to International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) codes.

Data and materials

The dataset used in this study is managed by Taiwan’s Ministry of Health and Welfare (MOHW). The MOHW approved our application to access these data. Any researcher interested in accessing this dataset must submit the relevant application to the MOHW. Please contact MOHW personnel (email: stcarolwu@mohw.gov.tw) for further assistance. MOHW address: No. 488, Sec. 6, Zhongxiao E. Rd., Nangang Dist., Taipei City 115, Taiwan (R.O.C.). Phone: +886-2-8590-6848. All relevant data are provided in this paper.

Study population

The ICD-9-CM codes on the basis of which ILD-PF was diagnosed are provided in S2 Appendix. Patients having ILD-PF with infection, such as viral pneumonia (code 480), influenza (codes 487 and 488), and viral infection (079.0, 079.1, 079.2, 079.3, 079.4, 079.5, 079.6, and 079.8), or ILD-PF with respiratory failure (518.81, 518.82, 518.83, and 518.84) between 2000 and 2012 were enrolled [7,8,17–20]. The date of the first diagnosis of ILD-PF was set as the index date [18,20]. We excluded patients aged younger than 18 years and patients with a history of CAD or stroke before the index date, the ICD-9-CM codes of which are as follows: acute myocardial infarction (AMI), 410; other acute and subacute forms of ischemic heart disease, 411; old myocardial infarction, 412; angina pectoris, 413; other forms of chronic ischemic disease, 414; hemorrhagic stroke, 430–432; and ischemic stroke, 433–437. To assess the statin-associated risk of CAD and stroke, patients with ILD-PF were assigned to statin and nonstatin groups. The follow-up period ended on the date of a CAD or stroke event, date of death, date of withdrawal from the NHI, or December 31, 2013, whichever occurred first.

Comorbidities and medications

Effects of comorbidities and medication were analyzed. Comorbidities correlated with the study endpoints comprised sleep disorders (codes 307.4 and 780.5), diabetes mellitus (code 250), hypertension (code 401–405), hyperlipidemia (code 272), alcohol-related illnesses (codes 291, 303, 305, 571.0, 571.1, 571.2, 571.3, and 790.3), chronic kidney disease (codes 585, 586, 588.8, and 588.9), gout (code 274), and cancer (codes 140–208). Furthermore, the effects of using inhaler corticosteroids (ICSs; e.g., budesonide/formoterol, fluticasone, and budesonide) were analyzed, as were those of using oral steroids (OSs; e.g., betamethasone, dexamethasone, methylprednisolone, triamcinolone, prednisone, prednisolone, hydrocortisone, and cortisone).

Atherosclerotic CV disease (ASCVD) included acute coronary syndromes, myocardial infarction, stable or unstable angina, arterial revascularization, stroke/transient ischemic stroke, and peripheral arterial diseases. However, a ≥20% 10-year ASCVD risk for a composite 3-point major atherosclerotic CV event of nonfatal myocardial infarction, nonfatal stroke, or CV death can serve as an arbitrary definition of patients with very high risk [21]. Thus, following the practice of the Taiwanese Secondary Prevention for Patients with Atherosclerotic Disease Registry [22,23], we replaced ASCVD with the following ICD-9-CM codes: AMI, 410; other acute and subacute forms of ischemic heart disease, 411; old myocardial infarction, 412; angina pectoris, 413; other forms of chronic ischemic disease, 414; hemorrhagic stroke, 430–432; ischemic stroke, 433–437; peripheral artery disease, 440.0, 440.2, 440.3, 440.8, 440.9, 443, 444.0, 444.22, 444.8, 447.8, and 447.9; arterial revascularization, 36.0–36.3; and operations on heart vessels, 36.9 [23,24]. The major atherosclerotic CV event was replaced with ICD-9-CM codes for all-cause mortality as follows: CV death, 390–459; nonfatal stroke, 430–438; nonfatal heart failure, 428; and AMI, 410. CV death was defined as a primary diagnosis of CV disease within 90 days of death.

Statistical analysis

Descriptive statistical differences in demographic characteristics, comorbidities, and medication use between the statin and nonstatin groups were analyzed. The between-group differences in baseline distributions were examined through t tests for continuous variables and chi-square tests for categorical variables. The Kaplan–Meier method was employed to estimate the cumulative CAD-free and ischemic stroke–free survival rates, and the corresponding survival curves of both groups were plotted using a Cox model in which age, sex, comorbidities, and medication use were controlled for. Between-group differences were determined through a likelihood ratio test. Incidence rates of CAD and stroke (including ischemic and hemorrhagic stroke) were calculated for each group as the number of observed events divided by the total person-years of experience. Use of statins, namely atorvastatin, fluvastatin, lovastatin, pravastatin, rosuvastatin, and simvastatin, for a 6-month period or longer was quantified as a binary variable [25]. On account of the frequency of variations in statin use among patients with ILD during the study period, statins were considered time-dependent covariates in the Cox proportional hazards models. The Cox regression results were used to obtain the hazard ratios and 95% confidence intervals (CIs) of CAD and ischemic stroke in the statin and nonstatin groups. All adjusted hazard ratios (aHRs) and their 95% CIs were measured after the following were controlled for: age, sex, ICS use, OS use, and comorbidities of sleep disorders, diabetes, hypertension, hyperlipidemia, mental disorders, alcohol-related illnesses, chronic kidney disease, gout, and cancer. Analyses were performed using SAS software, Version 9.4 of the SAS System for Unix (SAS Institute Inc., Cary, NC, USA), and the figures were plotted in R software. Statistical significance was indicated if P < 0.05.

Validation of interstitial lung disease, pulmonary fibrosis, stroke, and CAD

Studies outside Taiwan using administrative and claims data have applied validated methods for identifying ILD-PF; one study, for example, determined that ICD-9-CM codes 515 and 516.3 indicated idiopathic pulmonary fibrosis–pulmonary fibrosis [18,26]. Regardless of the diagnostic codes used for ILD-PF identification, such algorithms can be improved using procedural codes, such as those for lung biopsies and relevant imaging techniques (e.g., high-resolution computed tomography) [19]. Studies in Taiwan have diagnosed ILD-PF on the basis of these criteria [20,26,27].

A study in Taiwan used the Patients with Catastrophic Illnesses or Rare Diseases database to enroll patients with autoimmune diseases, such as rheumatoid lung disease [28]; systemic lupus erythematosus; multiple sclerosis; Sjögren’s syndrome; cancer, occupational lung diseases; and postinflammatory pulmonary fibrosis, alveolar pneumonopathy, or idiopathic pulmonary fibrosis [8,18]. Data on viral pneumonia, influenza, and viral chlamydia infection combined with ILD-PF are included in this database, which is a subsection of the NHIRD [8,29–31]. All registrants with ILD-PF or respiratory failure are given catastrophic illness certificates on the basis of clinical and laboratory diagnoses by rheumatologists or pulmonologists [31]. The NHI program waives copayments for ILD-related treatments for these registrants. The records of all patients identified as having ILD-PF according to our classification criteria were reviewed by a pulmonologist and a rheumatologist [32]. In Taiwan, protocols concerning the diagnosis of CAD and stroke are well established. In this context, CAD encompasses angina, coronary intervention, coronary surgery, and death from myocardial infarction. CAD and stroke were diagnosed according to clinical data, imaging results (e.g., angiography), or surgical intervention [23,33]. These risks have been validated in previous studies [23,32]. The validation of CAD and stroke in the NHIRD revealed that their positive predictive values were high—up to 88% and 88.4% for CAD and ischemic stroke, respectively [23,33].

Sensitivity analysis

We stratified the study cohort on the basis of sex, age (<50 and ≥50 years), statin (ICS/OS) use, and statin use without ICS or OS use. Statin users in all subgroups, except for the younger subgroup, had a lower risk of ischemic stroke than did statin nonusers. Moreover, statin users in all subgroups had a lower risk of CAD and ischemic stroke, except in those with combined statin and ICS use. These findings are consistent with the primary outcome. The two populations differed in size and baseline characteristics; thus, we conducted time-dependent and subgroup analyses [34]. For example, the older subgroup (statin use: statin nonuse, n = 8413:10,949) had lower risks of CAD and ischemic stroke. Moreover, after baseline cofounders were adjusted, independent risk factors between the two groups were identified, revealing that the statin cohort had a lower risk of ASCVD than the nonstatin cohort.

Strengths

The strength of this study is that our data source was a large population-based database under a universal health-care system. Furthermore, we employed validated algorithms for the determination of baseline comorbidities and confirmation of ILD-PF diagnoses.

Considering that adherence to statin treatment was suboptimal in practice, we performed a time-dependent analysis to prevent bias related to the confounding factors of nonadherence from affecting the results. We replaced hypertension, hyperlipidemia, and diabetes with obesity; alcohol-related disease with lifestyle; ICS and OS use with chronic obstructive pulmonary disease; and smoking, air pollution, and mental and sleep disorders with adherence. Furthermore, we stratified the patients by age and sex for sensitivity analysis, the results of which accord with our principal findings.

Limitations

Our study has several limitations. First, as with all studies that use administrative data, unmeasured confounders may be present. These confounders include the use of over-the-counter medications (e.g., nonsteroidal anti-inflammatory drugs), intercurrent illness, and the underlying risk of CAD and stroke. Second, the NHIRD does not contain data on blood pressure, blood glucose, cholesterol, renal function, or liver function, which are useful for assessing CAD and stroke status. However, the inclusion of such data is unlikely to change our key findings, although it may reduce our findings’ statistical power due to effect modification by baseline CAD and stroke status. Notably, low estimated glomerular filtration rate does not greatly affect blood statin levels because these medications work through hepatic metabolism rather than renal excretion. Third, as is the case in most pharmacoepidemiological studies, we examined medication use according to data on the dispensation of medication from pharmacies. Some participants may have not taken the medications on the indicated dates, if at all. However, statins were considered time-dependent covariates in the present Cox proportional hazards models. Despite these limitations, the findings contribute to the management of ILD-PF for the prevention of CV diseases.

Conclusion

Statin use, whether alone or in combination with OS use, plays an auxiliary role in the management of CAD and ischemic stroke in patients with ILD-PF. Large-scale follow-up studies on the use of steroids and statins in ILD-PF are required.