Sleep-Disordered Breathing Is Associated With Recurrent Ischemic Stroke

Brown, Devin L. MD; Shafie-Khorassani, Fatema MPH; Kim, Sehee PhD; Chervin, Ronald D. MD, MS; Case, Erin BA; Morgenstern, Lewis B. MD; Yadollahi, Azadeh PhD; Tower, Susan MD; Lisabeth, Lynda D. PhD


Stroke. 2019;50(3):571-576. 

In This Article



The data that support the findings of this study are not available to other investigators as this process is not covered by the informed consent document. Methods of the BASIC project (Brain Attack Surveillance in Corpus Christ), an ongoing population-based stroke surveillance study, have been previously described in detail.[12,14] The study takes place in Nueces County, a geographically isolated location in south Texas without an academic medical center. This facilitates complete case capture for stroke and avoids tertiary care bias. Both active and passive surveillance are used to identify all strokes in the 7 acute care hospitals in the county. Active surveillance includes review of admission logs for validated stroke symptom terms. Study staff also canvas hospital floors and intensive care units to search for stroke patients. Passive surveillance supplements this process and involves systematic review of all hospital discharge codes for stroke-related codes. For this analysis, only ischemic strokes identified from 2010 to 2015, validated by stroke fellowship-trained physicians using source documentation, inclusive of brain imaging reports, and based on a standard clinical definition,[15,16] were included. Ischemic stroke was defined as a rapidly developed focal neurological deficit referable to a vascular distribution without documented resolution within 24 hours (unless treated with thrombolytics) and not explained by a nonvascular cause. Brain imaging reports were used to differentiate ischemic stroke and intracerebral hemorrhage. Use of the clinical definition, as opposed to a requirement for imaging evidence of acute infarction, remains preferable for studies, such as BASIC, that examine temporal trends over many years including those when magnetic resonance imaging was not as widely used.[17]

Enrollment into BASIC was offered to those eligible. Eligibility entailed being age 45 or older and residency in Nueces County. Participation included a baseline interview shortly after stroke and follow-up interviews. Prestroke cognitive function (the Informant Questionnaire for Cognitive Decline in the Elderly) and prestroke function (modified Rankin Scale) were collected by way of interview of a close relative or friend and the subject or proxy, respectively. Medical records were abstracted to obtain information about stroke risk factors, demographics, and key clinical variables. Eligible subjects were also offered SDB assessment with a home sleep apnea test performed as soon as possible after stroke symptom onset, often during the stroke hospitalization. Subjects were eligible for this objective SDB assessment if they were identified through active surveillance within 30 days of stroke onset, or through passive surveillance within 45 days. Exclusion criteria included current pregnancy or current use of oxygen or positive pressure ventilation. Institutional Review Board approval was obtained from the University of Michigan and the 2 Corpus Christi hospital systems. Written informed consent was obtained from subjects or their surrogates.

Primary Exposure

SDB assessment was performed with the ApneaLink Plus. This is a well-validated home sleep apnea test device that monitors nasal pressure, respiratory effort, pulse, and oxygen saturation.[18–21] A trained study coordinator applied the device in the hospital or subject's home and then downloaded the raw data from the device for processing by the ApneaLink software. The software provided automated scoring, based on default settings used in published validation studies, complemented by a registered polysomnographic technologist's edits of the start and stop times and artifacts.[18–20] Detailed methods and definitions of apneas and hypopneas have been published previously.[22] An REI score, the sum of apneas plus hypopneas per hour of recording (similar to an apnea-hypopnea index but from a home sleep apnea test device that does not provide information about sleep duration), was calculated as a quantitative measure of SDB. Presence of SDB was determined by an REI score ≥10.

Outcome Identification

Recurrent ischemic strokes, the primary outcome of interest, were identified through the same surveillance procedures as described above for index strokes. All-cause mortality, the secondary outcome, was identified through the Texas Department of State Health Services records. Links to the BASIC data were made based on first and last name, sex, race/ethnicity, date of birth, and patient address.

Statistical Analysis

Baseline characteristics were compared by SDB status with χ2 and Kruskal-Wallis tests as appropriate. The analytic sample for modeling was limited to only MAs and NHWs given the small number of individuals of other races/ethnicities. Subjects were followed from the date of index stroke presentation until the first recurrent ischemic stroke, death, or the last follow-up date (December 31, 2015), whichever came first. The primary event of interest was ischemic stroke recurrence; however, some patients died without having a recurrence. Hence, death before a recurrence was considered a competing risk event. Competing risks methodology[23] may provide further insights into the effect of SDB on recurrent stroke and poststroke mortality, separately. Cases were censored at the end of the follow-up period if they had not yet had either event. Cause-specific proportional hazards models were used to assess the associations of continuous REI (modeled linearly) with the rate of recurrent ischemic stroke (the event of interest) and the rate of death (the competing risk event), unadjusted and adjusted for potential confounders.[24] For ease of interpretation, this fully adjusted model was used to examine the association between a 20-point increase in REI and recurrent stroke. Fully adjusted models included demographics, hypertension, diabetes mellitus, body mass index, and initial stroke severity. The proportional hazards assumption was tested for each variable using the cumulative sums of martingale residuals test.[24] Linearity was checked for all continuous variables (REI, REI by ethnicity interaction, age, and initial National Institutes of Health Stroke Scale) using the cumulative sums of martingale residuals. In separate models, an interaction term between REI and ethnicity was added to the fully adjusted models for each outcome. Finally, the association between MA ethnicity and each outcome was assessed with and without inclusion of the marginal REI effects in the fully adjusted models. A 10% change was considered the threshold for confounding of the MA-outcome association by REI.