Oct 13, 2015 - From Duke Clinical Research Institute, Duke University Medical Center, ... Downloaded from .... among sta
Health Services and Outcomes Research Clinical Effectiveness of Statin Therapy After Ischemic Stroke: Primary Results From the Statin Therapeutic Area of the Patient-Centered Research Into Outcomes Stroke Patients Prefer and Effectiveness Research (PROSPER) Study Emily C. O’Brien, PhD; Melissa A. Greiner, MS; Ying Xian, MD, PhD; Gregg C. Fonarow, MD; DaiWai M. Olson, PhD, RN; Lee H. Schwamm, MD; Deepak L. Bhatt, MD, MPH; Eric E. Smith, MD; Lesley Maisch, BA; Deidre Hannah, MSN, RN; Brianna Lindholm, BA; Eric D. Peterson, MD, MPH; Michael J. Pencina, PhD; Adrian F. Hernandez, MD, MHS Downloaded from http://circ.ahajournals.org/ by guest on October 31, 2017
Background—In patients with ischemic stroke, data on the real-world effectiveness of statin therapy for clinical and patientcentered outcomes are needed to better inform shared decision making. Methods and Results—Patient-Centered Research Into Outcomes Stroke Patients Prefer and Effectiveness Research (PROSPER) is a Patient-Centered Outcomes Research Institute–funded research program designed with stroke survivors to evaluate the effectiveness of poststroke therapies. We linked data on patients ≥65 years of age enrolled in the Get With The Guidelines–Stroke Registry to Medicare claims. Two-year to postdischarge outcomes of those discharged on a statin versus not on a statin were adjusted through inverse probability weighting. Our coprimary outcomes were major adverse cardiovascular events and home time (days alive and out of a hospital or skilled nursing facility). Secondary outcomes included all-cause mortality, all-cause readmission, cardiovascular readmission, and hemorrhagic stroke. From 2007 to 2011, 77 468 patients who were not taking statins at the time of admission were hospitalized with ischemic stroke; of these, 71% were discharged on statin therapy. After adjustment, statin therapy at discharge was associated with a lower hazard of major adverse cardiovascular events (hazard ratio, 0.91; 95% confidence interval, 0.87–0.94), 28 more hometime days after discharge (P25% of their patients, as well as patients missing all medical history data. For our secondary cohort of patients discharged on statin therapy, we included patients discharged between 2008 (when data collection on statin intensity began) and 2011 and required registry documentation of statin intensity for all patients.
Drug Exposure The primary exposure variable was any statin prescription at the index stroke hospitalization discharge as recorded in GWTG-Stroke. The secondary exposure variable was high- versus low/moderate-intensity discharge statin use in the subset of patients discharged on any statin therapy for whom the intensity level was recorded. High-intensity statin therapy was ascertained on the basis of either a registry flag indicating intensive statin or a combination of statin agent and dose of any of the following: atorvastatin ≥40 mg, rosuvastatin ≥20 mg, simvastatin 80 mg, or simvastatin/ezetimibe 80/10 mg. Low/moderateintensity statin therapy was defined as all other statin agents/doses.
Patient and Hospital Characteristics From the registry data, we obtained patient demographic characteristics, medical history, results of admission laboratory tests and examinations, in-hospital treatments, and discharge medications. Data on hospital-level characteristics (ie, number of beds, academic teaching status, annual volume of stroke discharges, rural location, and geographical region) were obtained from the American Hospital Association annual survey.16 Data on primary stroke center status were obtained from The Joint Commission Web site. For variables with low rates of missingness (ie, 5% missing, we treated the missing values as a separate category.18
Clinical Outcomes Outcomes of interest were major adverse cardiac events (MACEs), home-time days, all-cause mortality, all-cause readmission, ischemic stroke readmission, hemorrhagic stroke readmission, and cardiovascular readmission. Follow-up for all events of interest was censored at 2 years after the index discharge. We identified all-cause mortality on the basis of death dates recorded in the Medicare denominator files, and we identified all-cause readmission on the basis of a subsequent inpatient claim except transfers to or from another hospital and admissions for rehabilitation. Ischemic stroke readmission was based on a primary International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis code of 433.x1 or 434.x1 on the inpatient claim.19 Hemorrhagic stroke readmission was based on a primary International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis code of 430.x or 431.x on the inpatient claim. We defined cardiovascular readmission broadly as any readmission with a primary diagnosis code from the Agency for Healthcare Research and Quality’s Clinical Classifications Software circulatory diagnosis categories 96 through 118.20 MACE was defined as the composite end point of death or any cardiovascular readmission. For
1406 Circulation October 13, 2015 mortality, MACE, and readmission survival analyses, we censored data for patients if they enrolled in Medicare managed care. For readmission outcomes, we censored data for patients at the time of death. We calculated home time as total days alive and not in a hospital or skilled nursing facility in the 2 years after discharge. For all hometime analyses, we excluded patients who enrolled in Medicare managed care within 1 year of the index stroke hospitalization.
Statistical Analysis
Downloaded from http://circ.ahajournals.org/ by guest on October 31, 2017
The a priori statistical analysis plan is provided in the online-only Data Supplement. We present baseline characteristics of the study population, using frequencies with percentages for categorical variables and means with standard deviations or medians with interquartile ranges for continuous variables. We compare the respective distributions of baseline characteristics among patients receiving any versus no statin and high-intensity versus low/moderate-intensity statin. Variance between groups was evaluated with standardized differences for all variables (with differences of >10% considered meaningful), χ2 tests for categorical variables, and Kruskal-Wallis tests for continuous variables. We calculated Kaplan–Meier estimates for all-cause mortality and MACE at 2 years after the index hospitalization and used logrank tests to examine differences between statin treatment groups (any versus none and high versus low intensity). For all readmission outcomes, we calculated incidence at 2 years using estimates from the cumulative incidence function to account for the competing risk of mortality, and we used Gray tests to test for differences between groups. For home-time days, we calculated the mean (standard deviation) weighted by the proportion of 2-year follow-up and compared differences between groups using Kruskal-Wallis tests. To assess differences in outcomes among treatment groups (any statin versus no statin), we used inverse probability-weighted (IPW) estimates based on the probability of a patient receiving a treatment conditional on observed covariates.21 We used logistic regression to fit the propensity model with treatment as the dependent variable and age, sex, race, comorbidities, emergency medical services transport, NIHSS, low-density lipoprotein cholesterol (LDL-C), hospital characteristics, and year of the index hospitalization as independent variables. We used standardized differences, weighted χ2 tests for categorical variables, and weighted ANOVA for continuous variables to compare baseline characteristics between treatment groups after weighting.22 First, we estimated the unadjusted relationship between treatment and each outcome. For the home-time outcome, we used a negative binomial model with an offset for log of the proportion of follow-up time within 2 years after index discharge. For all other outcomes, we used Cox proportional hazards models. Next, we estimated the adjusted relationship between treatment and home-time days using weighted negative binomial models and between treatment and all other outcomes using weighted proportional hazards regression models. Finally, because patients who are prescribed discharge statins may also be more likely to receive other evidencebased therapies, we controlled for other discharge medications (antihypertensive, antithrombotic) in addition to the treatment indicator. To account for the clustering of patients within hospitals, we used random effects for negative binomial models and robust sandwich estimators for Cox models. We repeated this process to examine the unadjusted and adjusted relationships between high-intensity statin treatment and outcomes. Because the effect of statin use on outcomes may vary by patient characteristics, we examined treatment effects in the following prespecified subgroups: age (65–80 versus >80 years), sex (male versus female), race (nonwhite versus white), LDL-C among patients recorded at admission (
