Antiplatelet Therapy After Spontaneous Intracerebral Hemorrhage and Functional Outcomes

Santosh B. Murthy, MD, MPH; Alessandro Biffi, MD; Guido J. Falcone, MD, ScD, MPH; Lauren H. Sansing, MD, MS; Victor Torres Lopez, BS; Babak B. Navi, MD, MS; David J. Roh, MD; Pitchaiah Mandava, MD, PhD, MSEE; Daniel F. Hanley, MD; Wendy C. Ziai, MD, MPH; Hooman Kamel, MD; Jonathan Rosand, MD; Kevin N. Sheth, MD


Stroke. 2019;50(11):3057-3063. 

In This Article


Study Design and Patients

We conducted a cohort study using data from the following 3 participating studies: (1) the single-center longitudinal study of ICH conducted at Massachusetts General Hospital (MGH) in Boston, MA from 1994 onward and currently ongoing;[7] (2) the Virtual International Stroke Trials Archive-ICH (VISTA-ICH) established in 2010, which includes anonymized, individual patient-level data from completed ICH trials;[13] and (3) the single-center longitudinal ICH database of Yale University School of Medicine from 2013 and currently ongoing. Inclusion criteria in each study were (1) diagnosis of acute primary (nontraumatic) ICH confirmed on computed tomography scan, (2) age 18 years or older at time of ICH, (3) no history of prior ICH, and (4) follow-up data available at 90 days. We excluded patients with secondary causes of ICH such as trauma, vascular malformation, tumor or other mass lesions, cerebral venous sinus thrombosis, and hemorrhagic transformation of ischemic infarct. In addition, patients on prior anticoagulant medications, ascertained on chart review, were also excluded from our analysis. Because of institutional data sharing restrictions, we analyzed the 3 cohorts separately. The study was approved by the institutional review board of Weill Cornell Medicine, Yale School of Medicine, and MGH.

Data Availability Statement. The VISTA-ICH dataset used in this analysis may be obtained after approval of a written proposal by the VISTA-ICH steering committee. The Yale stroke registry data may be shared with investigators after institutional review board approval. Given data sharing restrictions for the MGH data instituted by the institutional review board, this dataset cannot be shared with investigators. However, results of statistical analyses from the MGH ICH registry may be shared with other investigators.

Exposures and Outcomes

Our exposure was APT after ICH. Our primary outcomes were all-cause mortality and a composite of death or major disability, defined as modified Rankin Scale (mRS) scores of 4 to 6 assessed at 90 days. Follow-up information was obtained via a combination of telephone interviews, clinic visits, chart reviews, and social security death index database interrogations to identify unreported mortality events. All follow-up and outcome adjudications were conducted centrally by the coordinating center for each cohort.

Statistical Analysis

Overall Analysis Plan. Because of institutional data sharing restrictions, we analyzed the 3 cohorts separately. We first performed Cox regression analyses in each cohort and subsequently meta-analyzed the hazard ratio (HR) obtained from the 3 studies to calculate a pooled HR for each outcome.

Analysis of the Individual Cohorts. We defined APT as a time varying covariate. Specifically, they contributed person-time to the unexposed group (controls) from the time of the study entry until the start of APT and then contributed to the exposed group from the time of stating APT to the end of follow-up. Our rationale for using this methodology was to mitigate the confounding effect of early deaths and variable timing of APT. Univariate and multivariate analyses were performed as detailed below within each individual study dataset. We used Cox regression models in the multivariate analysis assessing the relationship between APT and outcome (mortality and favorable outcome). To ensure uniformity in the statistical analyses across the unpublished cohorts, the Cox models were adjusted for prespecified covariates that included age, sex, admission Glasgow Coma Scale score, admission hematoma volumes, presence of intraventricular hemorrhage, hematoma location, and prior APT, regardless of significance on the univariate analysis. We tested the proportional hazard assumption separately within each study by means of graphical checks and Schoenfeld residual-based tests. Multicollinearity was assessed by computing variance inflation factors for all variables.

Meta-analysis Plan. We performed a meta-analysis using results from the 3 cohorts to assess the association between APT and ICH outcomes using the pooled HR as the effect parameter for primary outcomes and pooled OR for the secondary outcome. We used the adjusted HR (OR for mRS shift) and CIs from each study to generate random-effects (DerSimonian-Laird) models to calculate the pooled HR or pooled OR and generated forest plots.[14,15] The rationale of using the more conservative random-effects model was to account for the variability in effect sizes, design, and follow-up between the individual studies. We assessed heterogeneity using the Cochran Q test. Statistical analyses were performed using Stata (version 14.0, College Station, TX). All tests were 2-tailed and P<0.05 (CIs not including the null value of 1) were considered significant.

Additional Analyses. We performed additional analyses to study the relationship between APT and primary outcomes, stratified by hematoma location (lobar and deep). Similar to the primary analysis, we first calculated the HR in the 3 cohorts for each outcome, and then meta-analyzed the HRs to generate a pooled HR.