Flow diagram of enrollment, follow-up, and analysis for Adherence Evaluation After Ischemic Stroke–Longitudinal (AVAIL) subjects. GWTG indicates Get With The Guidelines program.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Bushnell CD, Zimmer LO, Pan W, et al. Persistence With Stroke Prevention Medications 3 Months After Hospitalization. Arch Neurol. 2010;67(12):1456–1463. doi:10.1001/archneurol.2010.190
In the United States, there are an estimated 180 000 recurrent strokes each year.1 Reducing the rate of recurrent stroke requires detecting and treating modifiable risk factors in the early poststroke period and developing strategies to improve patient persistence with medication regimens. Although early initiation of prevention strategies has shown sustained adherence and good outcomes in stroke patients,2,3 other evidence suggests that use of these secondary prevention therapies may not persist long-term.4-6 In patients with coronary heart disease, nonpersistence with secondary prevention therapies was associated with a 2-fold increase in cardiovascular disease events, including stroke.7 However, in stroke patients—a group at high risk for potential recurrent cardiovascular events, comorbidities, and stroke-related disability—there is limited information on longitudinal use of prevention medications.
The objective of the Adherence Evaluation After Ischemic Stroke–Longitudinal (AVAIL) Registry was to measure secondary prevention medication “regimen” persistence from hospital discharge to 3 months in a nationwide sample of patients from hospitals participating in the American Heart Association/American Stroke Association–administered Get With The Guidelines–Stroke program (GWTG-Stroke). We hypothesized that a combination of patient- and stroke-related factors, patient-provider communication, and system factors would be associated with regimen persistence.
Comprehensive detail on the purpose, scope, and methods of the AVAIL Registry program has been published previously.8 Briefly, 3068 patients were recruited during the acute stroke hospitalization from 106 hospitals participating in the GWTG-Stroke program. Outcome Sciences Inc serves as the data collection and coordination center for GWTG-Stroke. The Duke Clinical Research Institute serves as the data analysis center and has an agreement to analyze the aggregate deidentified data for research purposes. Each participating site obtained institutional review board approval. Study inclusion criteria were age 18 years or older; hospitalization for a primary diagnosis of acute ischemic stroke or transient ischemic attack; direct hospital admission based on physician evaluation or arrival through the emergency department; patient or legally authorized representative consent to participate; and patient inclusion in the GWTG-Stroke program.
Subjects were contacted by research personnel at the coordinating center 3 months after hospital discharge. Interviewers used standardized scripts to ascertain stroke prevention medication persistence. Subjects were asked to gather all prescription and nonprescription (such as aspirin) medication bottles for reference during the telephone call. Medication persistence was determined by comparing the complete discharge medication list with the current medications actually used by the subject. Subjects who reported discontinuing a medication were asked whether they chose to stop the medication or were instructed to do so by a physician. If discontinuation was self-initiated, then subjects were asked to select from a list the response that most closely represented their reason for discontinuation: adverse effects, cost, medication not helping, or other.
For subjects who could not respond because of illness severity, speech or language deficits, or death, interviewers spoke with an informed proxy, such as a family member or caregiver. A subject was classified as “lost to follow-up” only after multiple contact attempts were unsuccessful and the time from discharge was greater than 137 days or the subject/proxy refused.
Proxy responders (eg, caregiver) were not asked about the subject's understanding of why or how medications were prescribed or the adverse effects, how to refill medications, the subject's stroke symptoms or perceived recovery, depression, or quality of life.
The primary outcome for AVAIL was persistence, defined herein as continuing a therapy or class of therapy from discharge to the 3-month follow-up. Subjects prescribed an individual medication at discharge but who were not taking that medication at 3 months were defined as “nonpersistent.” Persistence for the specified medication classes (ie, antiplatelet, warfarin, antihypertensive, lipid-lowering, or diabetic agent) was defined in the same way; however, subjects were considered persistent if there was a switch to another medication within the same class. Both regimen and composite persistence summary scores were calculated. Regimen persistence refers to subjects who still took all discharge medication classes at the 3-month follow-up (regimen persistence = 1); regimen persistence = 0 for subjects taking less than the total number of medication classes prescribed at discharge. Composite persistence was defined as the percentage (0%-100%) of discharge medication classes that subjects were still taking at 3 months.
Subject and hospital characteristics were summarized with frequency distributions for categorical variables. Medians and interquartile ranges were reported for continuous variables. To compare subject and hospital characteristics among those with or without regimen persistence, χ2 tests were used for categorical variables and Wilcoxon rank sum tests were used for continuous variables. The comparisons were performed for all patients and repeated after excluding proxy responses.
To test our hypothesis that a combination of patient/caregiver–, provider-, and system-level factors are associated with regimen persistence, a multivariable logistic model was fitted. Prespecified covariates included proxy vs subject response, subject demographics (including age, sex, marital status, living situation, working status, medical history, and discharge medications), and socioeconomic factors that included income adequacy (yes or no), monthly medical cost (≥$100), financial hardship, and insurance status. Provider-level factors (having a primary care provider and the type of provider seen during follow-up); receipt of medication instructions at discharge; understanding how medications are taken, why medications are prescribed, and how to refill medications; methods to track medication use; monthly medical cost; income adequacy (yes or no); financial hardship on a scale between 1 (none) and 5 (severe); insurance status; stroke symptom type and subjective recovery; stroke vs transient ischemic attack; modified Rankin Scale score categorized by less than 3 or 3 or more9; EuroQol-5D score10; and Patient Health Questionnaire 8 score were also considered for the model.11,12 In addition, hospital-level factors were considered: hospital type (academic vs other), number of beds, number of annual stroke discharges, stroke center certification, and geographic region. The covariate missing rate was low (<2%). Missing data were imputed by the most popular category in the regression modeling analysis. National Institutes of Health Stroke Scale scores were missing for about 25% of subjects; therefore, this scale was not used in the analysis.
A backward variable selection procedure was applied to obtain a parsimonious model and higher statistical power. The final reduced model only contained predictors with a P value of less than .10. Because of self-responders and proxy responders reporting different persistence rates, the multivariate model of persistence was analyzed with and without proxy responders. Since the factors associated with persistence may differ based on medication class, the same modeling approaches were reconducted for each class separately.
All P values are 2-sided, with less than .05 considered as statistically significant. P values are reported without adjustment for multiple comparisons. All analyses were performed using SAS software (version 9.2; SAS Institute Inc, Cary, North Carolina).
A total of 3068 stroke subjects from 106 sites were assessed for eligibility for the AVAIL Registry. Of these, 180 (5.9%) were ineligible (Figure). Additional subjects were excluded from the analysis, leaving a final sample size of 2598 subjects and an overall lost to follow-up rate of 3.5%. The baseline characteristics of these subjects are listed in Table 1.
The median time between hospital discharge and 3-month follow-up was 123 days. Table 2 displays the percentage of subjects prescribed medications by class and by type at discharge, 3-month persistence, and discontinuation rates (physician's recommendation or self-discontinued). Persistence at 3 months was greatest for antihypertensive and antiplatelet medications. Overall, 0.6% of subjects were discharged taking none of the classes of medication; 7.2% were discharged taking 1; 23.4%, taking 2; 43.8%, taking 3; 23.4%, taking 4; and 1.6% taking all 5 classes of interest. Of those treated, 75.5% were persistent with all the medications prescribed by their physicians at discharge. Composite persistence analysis showed 5.8% were taking 75% to 99% of baseline discharge medications; 13.3% were taking 50% to 74%; 1.9% were taking 25% to 49%; and 3.5% were taking none of the baseline medications at 3 months.
Subject demographic characteristics, including age, sex, race, education level, and working status, were not associated with persistence in the univariate analysis, although older age was independently associated with persistence in the multivariable model. The variables associated with persistence with and without the proxy responders are shown in Table 3. The multivariable models for these same groups are shown in Table 4.
Patient-level factors independently associated with persistence included increasing age, absence of atrial fibrillation, and the presence of hypertension, diabetes mellitus, dyslipidemia, and coronary artery disease/prior myocardial infarction. These factors were significant in both multivariable models (Table 4). In evaluating the models separately by class, atrial fibrillation was positively associated with warfarin persistence (adjusted odds ratio [OR], 2.70; 95% confidence interval [CI], 1.52-4.80). However, atrial fibrillation was negatively associated with persistence in the other 4 models, which explains the overall negative association. Proxy responders (n = 421) were primarily a spouse or child (80%), and 270 (64%) reported regimen persistence for the subject at 3 months, which was significantly lower than regimen persistence by subject self-report (78%; P < .001).
Having some type of medication insurance was reported by 87.6% of subjects and was independently associated with persistence in the model combining subject and proxy responders. In contrast, in the model with self-responders only, working status (home not by choice vs home by choice or working) was associated with regimen persistence (Table 4). For both models, greater financial hardship was independently associated with persistence.
Questions related to specific provider specialties were collected from self-responders. There was no relationship between persistence and self-reported satisfaction with provider communication, including physician listening skills, clear medication explanations, physician use of understandable language, and subject involvement in treatment decisions. However, persistence was independently associated with subject-reported understanding of the reasons why prescribed medications should be taken (OR, 1.81; 95% CI, 1.19-2.76; P = .006) and with how to obtain medication refills (OR, 1.64; 95% CI, 1.04-2.58; P = .03) in the multivariable model with self-responders only. In addition, lower number of prevention medication classes prescribed at discharge was independently associated with regimen persistence in both multivariable models (Table 4). The type of provider was not associated with persistence.
Lower disability was associated with higher persistence when modified Rankin Scale scores were dichotomized as a good or bad outcome in the subject plus proxy model (OR, 1.54; 95% CI, 1.24-1.90; P < .001) but less significant in the subject responder only model (OR, 1.28; 95% CI, 0.97-1.69; P = .08) (Table 4). Quality of life measured with the EuroQol-5D (score between 0 and 1) was higher in persistent subjects (median, 0.83; interquartile range, 0.76-1.0) vs nonpersistent subjects (median 0.81, interquartile range, 0.69-1.00; P = .001) and independently associated with persistence in the self-responder only model (Table 4).
Regimen persistence of stroke prevention medications was fair at 76%, although considering the setting, ie, hospitals motivated to provide the best stroke care, this may be the “best-case scenario.” Multiple factors were associated with persistence at 3 months, including the presence of cardiovascular disease and risk factors prior to the stroke; having insurance; fewer discharge medication classes; and patient understanding why these prevention medications were prescribed and how to refill prescriptions. In addition, increasing age, lesser stroke disability, and increased financial hardship were associated with persistence. The assessment of and reasons for nonpersistence at 3 months poststroke are important because the risk of recurrent stroke is greatest during this period.
In AVAIL, nonpersistence included medication discontinuation by the patient or provider. In fact, subjects reported that the majority of discontinuation was because of the provider. However, the small proportion of self-discontinuation for each medication would not allow for detailed analyses of this group. Although self-reported medication persistence is fairly accurate,13,14 factors associated with nonpersistence likely encompassed both patient- and provider-related factors. For example, it is unknown whether subjects prompted the provider to stop a medication because of adverse effects or cost or whether the physician did not believe that the specific medication was indicated. These subtle aspects of medication-taking behavior and medication-prescribing behavior need to be explored further in the stroke population.
Medication class persistence rates for AVAIL are lower than in the PROTECT program, a single-center study of medication and lifestyle adherence in stroke patients.15 The PROTECT program used performance improvement tools/algorithms including written patient materials, primary care physician hand-off letters, and telephone contact 2 to 4 weeks postdischarge. PROTECT reported 3-month persistence rates of 100% for antithrombotics, 99% for statins, 92% for angiotensin-converting enzyme inhibitors/angiotensin receptor blockers, and 80% for thiazide diuretics.2 AVAIL sites were participating in the GWTG-Stroke program, and therefore, it was expected that they were performing stroke education at discharge. In contrast, persistence was much lower in stroke patients in a South London, England, inner-city population, with 76% persistence for antiplatelet therapy, 41% for anticoagulants, and 70% for blood pressure–lowering therapy.16
A recent study measured medication persistence by linking registry subjects to a pharmacy database with prescription refills for up to 2 years following the acute stroke hospital discharge.17 Medication persistence at 3 months ranged from 89% for warfarin in patients with atrial fibrillation to 96% for any antiplatelet drug, which is in the range of the AVAIL data. However, they also found that medication persistence declined steadily over the 2 years of follow-up. The factors that were independently associated with persistence for most drugs included advanced age, comorbidity, good self-perceived health, absence of low mood, acute treatment in a stroke unit, and institutional living at follow-up.17 However, this study did not include stroke disability, patient-provider communication, or patient knowledge about medications in their analyses.
Mild vs moderate or severe disability was associated with 54% higher persistence (OR, 1.54; 95% CI, 1.24-1.90) in the combined model in AVAIL, a result that was independent of proxy response (Table 4). This could reflect withdrawal of some medications in patients with severe strokes who were referred to palliative care. Alternatively, these stroke patients may have more difficulty with access to medications, or a more complex regimen, similar to those who required a proxy response. Other studies that assessed persistence of prevention medications associated with stroke severity or disability have reported conflicting results. The South London inner-city study reported lower persistence in patients with more severe strokes assessed at discharge.16 A study conducted in Germany, however, found that more severe strokes on admission and cardioembolic stroke types were associated with better long-term persistence.18 The German study included the use of proxy respondents for the assessment of persistence, but there was no analysis of this group separate from subject responders. The South London study obtained information from the subject or the treating physician. Based on the current AVAIL analysis, at least some of the inconsistencies among studies may be related to the use or nonuse of proxy respondents.
Lower number of discharge medications, understanding why the medications were prescribed, and knowing how to refill prescriptions were important indicators of persistence. Several new medications may be prescribed as a result of the stroke, as well as medications for risk factors newly diagnosed during the hospitalization. These data indicate that patients and their caregivers managing complicated regimens or who had new medications prescribed during the hospitalization may benefit from streamlined medication regimens at discharge to avoid placing the burden on the provider to perform this function at follow-up visits. In addition, the education process at discharge may need to include more specific information about why medications are being prescribed and how they are refilled.
We found that higher scores on the EuroQol-5D10 were associated with increased persistence. This makes intuitive sense that patients who report a better quality of life may have recovered more, or adjusted better, to the stroke and are, therefore, ready to take the necessary steps to prevent future strokes.
Subjects and proxies who reported having any insurance were 31% more likely to be persistent at 3 months. This finding is not surprising because the discharge medication regimen may include new and expensive medications that can only be acquired or maintained with the help of insurance coverage. In addition, we found that perceived hardship score greater than 2 (on a scale of increasing hardship from 1-5) was independently associated with a higher rate of persistence compared with a score of 2 or less. One possible explanation is that patients who were consciously making efforts to pay for their medications may have felt financial strain.
The AVAIL Registry has several strengths: it is one of the largest registries outside of a clinical trial of medication-taking behavior specific to stroke patients. The AVAIL Registry includes comprehensive patient-, provider-, and system-level data and is also an extension of ongoing quality improvement efforts in nationally represented hospitals.
Study limitations include the use of only GWTG-Stroke sites; thus, the patient sample may not be generalizeable to other settings, such as hospitals that are not interested in stroke center certification. Second, persistence was obtained by self-report. Nevertheless, for registries of this size, such as the PREMIER study of patients after myocardial infarction,19 self-report is clinically useful, agrees with pharmacy claims data,13 and is the best method to ascertain reasons for nonpersistence.14 Lastly, information regarding the number of provider visits was not obtained at 3 months.
The AVAIL Registry showed that medication persistence is multifactorial. Understanding the complex patient, provider, and caregiver characteristics related to optimal medication-taking behavior in stroke patients is important. Using the insights from AVAIL, we can begin to develop and evaluate strategies to improve appropriate use of evidence-based therapies and reduce the risk of recurrent stroke.
Correspondence: Cheryl D. Bushnell, MD, MHS, Wake Forest University Health Sciences, Medical Center Boulevard, Winston-Salem, NC 27157 (email@example.com).
Accepted for Publication: May 9, 2010.
Published Online: August 9, 2010. doi:10.1001/archneurol.2010.190
The AVAIL Publications Committee: Mark Alberts, MD, Northwestern University; Bruce Coull, MD, University of Arizona; Pamela Duncan, PT, PhD, Duke University; Susan Fagan, PhD, PharmD, University of Georgia; Michael Frankel, MD, Emory University; Larry Goldstein, MD, Duke University; Philip Gorelick, MD, MPH, University of Illinois at Chicago; S. Claiborne Johnston, MD, PhD, University of California, San Francisco; Chelsea Kidwell, MD, Georgetown University; Ken LaBresh, MD, Research Triangle International; Pamela Mitchell, RN, PhD, University of Washington; Frederick Munschauer, MD, Buffalo General Hospital; Bruce Ovbiagele, MD, University of California, Los Angeles; Ralph Sacco, MD, MS, University of Miami; Lee Schwamm, MD, Massachusetts General Hospital; Linda Williams, MD, Roudebush VA Medical Center and Indiana University; Richard Zorowitz, MD, Johns Hopkins Bayview Medical Center.
Author Contributions: The authors are solely responsible, with approval by the project executive committee, for the design and conduct of this study, all study analyses, and the drafting and editing of the manuscript and its final contents. Study concept and design: Bushnell, Olson, Schwamm, Williams, and LaBresh. Acquisition of data: Zimmer, Olson, Meteleva, Ovbiagele, and Peterson. Analysis and interpretation of data: Bushnell, Zimmer, Pan, Olson, Zhao, Ovbiagele, Williams, and Peterson. Drafting of the manuscript: Bushnell, Zimmer, Olson, and Meteleva. Critical revision of the manuscript for important intellectual content: Zimmer, Pan, Zhao, Schwamm, Ovbiagele, Williams, LaBresh, and Peterson. Statistical analysis: Pan and Zhao. Obtained funding: Bushnell and Peterson. Administrative, technical, and material support: Zimmer, Olson, Meteleva, Schwamm, and LaBresh. Study supervision: Bushnell, Zimmer, and Olson.
Financial Disclosure: Drs Bushnell and Peterson both received research salary support from the Bristol-Myers Squibb/Sanofi Pharmaceuticals Partnership for their roles as co–principal investigators on the AVAIL registry. Dr Schwamm is chair of the national steering committee for GWTG. This research project was supported by unrestricted funds from the Bristol-Myers Squibb/Sanofi Pharmaceuticals Partnership and conducted through collaboration with the GWTG-Stroke program.
Funding/Support: The study was conceived and designed by the AVAIL team, researchers at the Duke Clinical Research Institute, the project executive committee, and an American Heart Association representative. The AVAIL analyses were also supported in part by Agency for Healthcare Research and Quality grant U18HS016964.
Role of the Sponsor: Duke Clinical Research Institute developed the protocol, owns the data, and is responsible for study oversight, materials development and data collection, site communications, and all regulatory and clinical questions related to the AVAIL Registry study.
Disclaimer: The content does not necessarily represent the official views of the Agency for Healthcare Research and Quality.
Additional Contributions: We thank Laura Drew, RN, BSN, and Judith A. Stafford, MS, from the AVAIL coordinating team, as well as the site investigators and coordinators, for their work on the study. We also acknowledge Andrea Davis, MSPH, Leslie Wilson, BA, and Vicky Pena, BS, for expertise in patient interviewing and data collection.
Create a personal account or sign in to: