Risk of ischemic stroke during the initiation period of α-blocker therapy among older men

Data source

The National Health Insurance Research claims database has been described previously.10^,12 To comply with Taiwanese privacy regulations, all personal identifiers are encrypted and data can be analyzed only anonymously. The protocol for this study was approved by the Institutional Review Board of the National Taiwan University Hospital, which waived requirement for informed consent.

Study design and cohort definition

We used claims data from the National Health Insurance Research claims database for the period 2006–2009 and applied a self-controlled case series design.10^,13^,14 We identified all men aged 50 years or more as of Jan. 1, 2007, who had continuous coverage under the National Health Insurance program for at least 12 months. The study period for each case was defined as the period from Jan. 1, 2007, to Dec. 31, 2009. We selected men who were incident users of α-blockers and had a diagnosis of ischemic stroke within the 3-year study period. We excluded those who had any diagnosis concerning ischemic stroke or any prescription of α-blocker in the baseline period (Jan. 1, 2006, to Dec. 31, 2006). Only the first event of ischemic stroke was included in the analysis.14 Because of the different clinical implications between use of antiplatelet or anticoagulant agents before (primary prevention) and after (secondary prevention) the incidence of ischemic stroke, we excluded patients who were prescribed any antiplatelet or anticoagulant agent before the incidence of ischemic stroke. This exclusion also increased the accuracy of identifying incident ischemic strokes with our study design.

Study timeframe

In the database’s outpatient records, each drug dispensed at outpatient clinics or pharmacies is included, with the corresponding date of dispensing. All drugs prescribed during hospital stays, however, are recorded in the database’s inpatient records, with only the dates of hospital admission and discharge. Therefore, we defined the index date either as the dispensing date of α-blocker therapy identified from outpatient records or as the date of hospital admission if the first dose of α-blocker was prescribed during the hospital stay. Because all patients with ischemic stroke were identified from the inpatient records, we presumed that the dates of hospital admission were the dates of occurrence of ischemic stroke.

We operationally defined the early initiation period (≤ 21 d after the index date) as post-exposure risk period 1 and the 22–60 days after the index date as post-exposure risk period 2.17 To determine whether cases of exposure were event-dependent, we also defined the 21 days before the index date as pre-exposure risk period 1 and the 22–60 days before the index date as pre-exposure risk period 2.14^,18 The times before and after the pre- and post-exposure risk periods in the 3-year study period were defined as the unexposed periods (Figure 1). For analysis, we grouped the 2 unexposed periods into 1 unexposed period because they were far away from the index date, they conveyed no acute hypotensive effect of α-blockers, and the crude incidence rates of ischemic stroke in the 2 unexposed periods were similar (see Table S1 in Appendix 1, available at www.cmaj.ca/lookup/suppl/doi:10.1503/cmaj.150624/-/DC1). Because we could not determine the true temporal relationship between initial exposure and the event for cases occurring on the same day, we excluded the index date during partition of the risk periods and included it in the unexposed period in the primary analysis for all the participants to get more conservative estimates.13^,18

• Download figure
• Open in new tab
• Download powerpoint

Figure 1:

Pictorial representation of the study design.

Statistical analysis

Using the conditional Poisson regression model, we compared the incidence rates of ischemic stroke in various risk periods with the rates in the unexposed period and presented the results as incidence rate ratios (IRRs).10^,13^,14^,17^,18 The key advantage of a self-controlled case series design is that time-invariant confounders, such as sex, location, genetics and underlying state of health, are controlled for implicitly.14 In the conditional Poisson regression model, we adjusted only for age at the index date and development of atrial flutter or fibrillation (ICD-9-CM code 427.3), which was included as a time-varying term.

Because the acute hypotension-related adverse events associated with α-blockers may be modified by the status of concomitant antihypertensive treatment,9^,10 we repeated the primary analysis with stratification according to whether patients were taking concomitant antihypertensive therapy on the index date. We also performed a stratification analysis according to different α-blockers administered.

We performed 2 sensitivity analyses to test the robustness of our results. In the first analysis, we included the index date in pre-exposure risk period 1, on the assumption that all patients whose index date and stroke event occurred on the same day had their α-blocker therapy prescribed during the hospital stay after a preceding ischemic stroke. In the second analysis, we excluded all patients who began their α-blocker therapy during a hospital admission (regardless of whether it was the same admission for stroke).

We repeated the same design as the primary analysis for patients newly prescribed diuretics and calcium-channel blockers, to explore whether the observed results for α-blockers represented a possible group effect of antihypertensive agents.

We performed all analyses using SAS software, version 9.2 (SAS Institute Inc.).