Representativeness of the Get With The Guidelines–Stroke Registry
Comparison of Patient and Hospital Characteristics Among Medicare Beneficiaries Hospitalized With Ischemic Stroke
Background and Purpose—Get With The Guidelines (GWTG)—Stroke is a large quality improvement-based registry of acute stroke; however, its generalizability is unclear. We used fee-for-service Medicare claims to ascertain the representativeness of ischemic stroke admissions in GWTG-Stroke.
Methods—All 228 815 ischemic stroke admissions aged ≥65 years enrolled in GWTG-Stroke between April 2003 and December 2007 were linked to 926 756 unique fee-for-service Medicare patients with ischemic stroke (primary International Classification of Diseases, 9th Revision discharge code 434 or 436) from the same period. Patient characteristics and in-hospital outcomes were compared between the linked GWTG-Stroke Medicare cohort and the remaining unlinked Medicare cohort. Characteristics of GWTG-Stroke hospitals were compared with non-GWTG-Stroke hospitals.
Results—A total of 144 344 of the 228,815 GWTG-Stroke admissions (63.1%) were successfully linked to the 926 756 Medicare ischemic stroke beneficiaries, leaving 782 412 unlinked Medicare patients. Differences in patient characteristics, including age, race, gender, and comorbidities, between the linked and unlinked Medicare cohorts were minimal. Length of stay and rate of discharge home were almost identical between the linked and unlinked groups; however, in-hospital mortality was slightly lower in the linked Medicare cohort (6.3%) compared with the unlinked cohort (7.0%). There were large differences in hospital characteristics between GWTG-Stroke and non-GWTG-Stroke hospitals; GWTG-Stroke hospitals tended to be larger, urban, teaching centers.
Conclusions—Despite substantial differences between GWTG-Stroke and non-GWTG-Stroke hospitals, Medicare beneficiaries with acute ischemic stroke entered in the GWTG-Stroke program were similar to other Medicare beneficiaries. These data suggest that the Medicare-aged GWTG-Stroke ischemic stroke admissions are generally representative of the national fee-for-service Medicare ischemic stroke population.
By focusing on large “real-world” patient populations, clinical registries are serving an increasingly important role in measuring the delivery, effectiveness, and efficiency of health care.1 However, the usefulness of any registry depends largely on the reliability of its data and the degree to which it accurately represents the underlying patient population targeted by the registry. The Get With The Guidelines (GWTG)–Stroke Program was developed by the American Heart Association/American Stroke Association as a national stroke registry and quality improvement program.2 Because of its substantial size, broad geographic scope, and prospective data collection, the GWTG-Stroke registry provides an important source of data concerning the characteristics, treatments, quality indicators, and in-hospital outcomes for a broad cohort of patients with acute stroke hospitalized throughout the United States.3,4 However, like with all registries because of the potential for selection bias both at the hospital and patient level, questions remain about its representativeness.5–8 Using national fee-for-service Medicare claims data of acute ischemic stroke hospitalizations, we examined the representativeness of the GWTG-Stroke registry by: (1) comparing patient characteristics and in-hospital outcomes between Medicare beneficiaries who were linked to the GWTG-Stroke registry and those that remained unlinked; and (2) comparing hospital characteristics between GWTG-Stroke and non-GWTG-Stroke hospitals.
GWTG-Stroke Registry Data
GWTG-Stroke is a voluntary, hospital-based, quality improvement registry that collects a wide range of clinical data on hospitalized patients with stroke. Details of the oversight, design, inclusion criteria, and data collection methods of the registry have been reported previously.3,4 For this analysis, data were limited to admissions aged ≥65 years with a clinical diagnosis of ischemic stroke that were entered into the registry between April 2003 and December 2007. All participating institutions were required to comply with local regulatory and privacy guidelines.
Medicare Claims Data
We obtained all fee-for-service Medicare inpatient claims files for acute stroke and transient ischemic attack (any International Classification of Diseases, 9th Revision discharge code 430–436) along with corresponding denominator files for the period April 2003 to December 2007. The inpatient files contain institutional claims for facility costs and provide dates of service, hospital identifiers, and diagnosis and procedure codes. The denominator files contain beneficiary demographic data, including age, sex, and birth and death dates. These files are research-identifiable, meaning that unique patient identifiers have been assigned to each unique individual beneficiary. Subjects in Medicare managed care plans (15% to 25% of the population) are excluded from fee-for-service claims.
Details of the data linkage methodology are provided online (Supplemental Table I; http://stroke.ahajournals.org) and are summarized in the Figure. Briefly, all stroke and transient ischemic attack admissions entered in the GWTG-Stroke registry between April 2003 and December 2007 were linked to all Medicare acute stroke and transient ischemic attack discharges from the same time period. Using previously described methods,9 individual admissions in the registry were matched to the Medicare claims by identifying unique combinations of data fields, including hospital ID code, admission and discharge dates, date of birth, and sex. This approach takes advantage of the hospital clustering observed in each data set to identify unique patient records without the need for direct patient identifiers.9 After this initial linkage was completed, the data were restructured to retain only the first index ischemic stroke admission for each unique Medicare beneficiary (see Supplemental Table I). After completion of the data linkage, the following 2 groups had been created: a linked GWTG-Stroke Medicare cohort and the remaining unlinked Medicare cohort (Figure).
Within-Medicare Comparisons of Linked and Unlinked Beneficiaries
We compared patient characteristics between the linked GWTG-Stroke Medicare cohort and unlinked Medicare cohort using the following data from the Medicare files: demographics, comorbidities, hospital length of stay, discharge destination, and in-hospital mortality. Comorbid conditions were defined using Hierarchical Condition Categories; all inpatient claims generated in the 365 days before the stroke discharge date were searched for evidence of relevant comorbid conditions.10
Within-GWTG-Stroke Comparisons of Linked and Unlinked Admissions
To identify differences between those GWTG-Stroke ischemic stroke admissions that were linked to the Medicare claims files and those that remained unlinked, we compared patient characteristics (demographics, medical history, in-hospital outcomes) between the 2 groups using data collected in the GWTG-Stroke registry (Figure).
Comparisons of GWTG-Stroke and Non-GWTG-Stroke Hospitals
Using the Medicare inpatient claims files, hospitals were defined as participating in GWTG-Stroke if they had at least 1 admission linked to the registry. Using information from the 2007 American Heart Association database,11 we compared the characteristics of GWTG-Stroke hospitals (ie, geographic region, rural/urban, number of beds, annual stroke volume, and teaching status) with non-GWTG-Stroke hospitals (defined as all other hospitals with Medicare claims for acute ischemic stroke hospitalizations).
We used χ2 and Wilcoxon rank sum statistics, respectively, to evaluate differences in categorical and continuous variables. Because of the extremely large size of the Medicare claims data, all descriptive comparisons no matter how trivial were statistically significant (< 0.001); thus, probability values are not reported. Instead we show absolute differences between the groups, defining potentially clinically meaningful differences as >2.0%.
To test for differences in patient outcomes between the linked GWTG-Stroke Medicare cohort and the unlinked Medicare cohort, we used a combination of multivariable analyses, including logistic regression models for binary outcomes (discharge home, in-hospital mortality) and linear models for length of stay. All models used general estimating equation methods to account for within-hospital clustering. All multivariable models controlled for patient characteristics (age, sex, race/ethnicity, medical history of stroke, coronary heart disease, myocardial infarction, carotid stenosis, chronic obstructive pulmonary disease, dementia, diabetes, peripheral vascular disease, pneumonia, renal dysfunction, and hypertension) and hospital-level variables (region, rural/urban, number of beds, annual stroke volume, and teaching status).
A total of 228 815 admissions age ≥65 years with a clinical diagnosis of ischemic stroke were enrolled at 1087 GWTG-Stroke registry hospitals during the study period; 66.4% (n=151 964) were successfully linked to the Medicare fee-for-service claims data. In the Medicare claims files during the same period, there were 926 756 unique beneficiaries with acute ischemic stroke from 4234 hospitals. Of these 144 344 (15.6%) were successfully linked to the GWTG-Stroke registry, leaving 782 412 unlinked Medicare beneficiaries. Of the 4234 hospitals in the Medicare claims, 1022 (24.1%) had ≥1 linked patients indicating that they had participated in the GWTG-Stroke program. These 1022 hospitals represented 94% of the 1087 hospitals in the GWTG-Stroke registry during this time period.
Table 1 compares patient characteristics between the linked GWTG-Stroke Medicare cohort (n=144 344) and the unlinked Medicare cohort (n=782 412). Overall, there were only relatively minor absolute differences (ie, approximately ≤1.0%) between the linked and unlinked cohorts. The mean age was almost identical (79.4 versus 79.7 years in the linked and unlinked cohorts, respectively). The linked GWTG-Stroke patients were slightly more likely to be male (42.1% versus 39.9%), white (86.0% versus 83.1%), and less likely to be black (10.2% versus 12.3%). There were a few differences in comorbid conditions between the 2 groups; GWTG-Stroke beneficiaries were more likely to have a history of carotid stenosis (16.3% versus 8.3%), prior stroke (10.9% versus 8.7%), and renal disease (12.8% versus 9.8%).
Comparisons of hospital characteristics between the 1022 GWTG-Stroke hospitals and 3212 non-GWTG-Stroke hospitals are shown in Table 2. There were large geographical differences in the location of the 2 groups of hospitals; substantially more GWTG-Stroke hospitals were from the Northeast and South and fewer from the Midwest. GWTG-Stroke hospitals were much more likely to be larger, teaching hospitals from urban areas with higher annual ischemic stroke volumes (median 102 for GWTG-Stroke hospitals versus 26 for non-GWTG-Stroke hospitals).
Data on length of stay, discharge to home, and in-hospital mortality for the linked GWTG-Stroke Medicare and unlinked Medicare cohorts are shown in Table 3. Length of stay was identical between the 2 groups (ie, mean, 6.5 days; median, 5 days; interquartile range, 4–8 days), whereas the proportion discharged home was similar (absolute difference 0.1%). The in-hospital mortality rate was slightly lower in the linked GWTG-Stroke cohort compared with the unlinked cohort (absolute difference 0.7%), and this difference remained after adjustment for patient- and hospital-level variables (Table 3).
To identify differences between GWTG-Stroke admissions that were successfully linked to the Medicare claims files (n=151 964 [66.4%]) and those that were not (n=76 851 [33.6%]), we compared patient characteristics between the 2 groups (Supplemental Table II). Linked admissions were slightly older (median, 80 years versus 78 years), more likely to be white (82% versus 74%), and less likely to be black, Asian, or Hispanic. There were minimal differences in medical history and length of stay. Linked admissions were slightly less likely to be discharged home (37% versus 40%) but had slightly lower in-hospital mortality (6.6% versus 7.5%) compared with unlinked admissions.
The major finding of this analysis was that Medicare beneficiaries with acute ischemic stroke entered in the GWTG-Stroke program were similar in terms of demographics, comorbidities, and in-hospital outcomes to the remaining acute ischemic stroke Medicare beneficiaries who were not linked to the GWTG-Stroke program. The similarity in these 2 patient groups was despite the fact that the GWTG-Stroke program has been more widely adopted by larger, academic, and urban hospitals. Overall, these results provide evidence to support the premise that the GWTG-Stroke program is representative of elderly patients with stroke and helps justify the generalization of the registry's findings to the national level with respect to patient characteristics, in-hospital care, and in-hospital outcomes. This is the first study to address the issue of representativeness of the GWTG-Stroke program and as such is responsive to the call for more studies that establish the representativeness of cardiovascular disease registries.1 The demonstration of representativeness of GWTG-Stroke is an important milestone given the increasing importance given to studying “real-world” populations and, in particular, to determining the presence disparities in healthcare delivery among patient subgroups (ie, racial and ethnic minorities, women, the elderly, and socioeconomically disadvantaged individuals) that are often underrepresented in scientific studies.1
We are not aware of previous studies that have examined the representativeness of other hospital-based stroke registries using claims-based data as we have done here. The few previous studies that have examined the representativeness of hospital-based stroke registries have used a comparable population-based stroke registry as a frame of reference.5,6,8 All of these studies have reported systematic differences suggesting some degree of patient-selection bias in the hospital-based registries. A study from Dijon, France, found that patients with stroke entered into a hospital-based registry were younger but had more severe stroke and higher mortality when compared with the population-based Dijon Stroke Registry.6 A quality assessment study of the Riks-Stroke national quality registry in Sweden found evidence that registry participants had lower case fatality and were more likely to be treated in a stroke unit and to receive rehabilitation care when compared with a population-based registry.5
Within the GWTG-Stroke registry itself, we did find some differences between the admissions that were linked to the Medicare data and those that remained unlinked, the most notable being the lower proportion of minorities in the linked group. The reasons for these differences remain unclear, but lower linkage of minorities has also been reported by other registry-based studies.12 The most important reason for why ischemic stroke admissions entered into GWTG-Stroke program did not link to the Medicare data are likely the fact that they had Health Maintenance Organization insurance and so were not included in the fee-for-service claims. Other potential reasons include data errors in the GWTG-Stroke registry for the fields used in the matching process (eg, admission date, discharge date, date of birth) or discrepancies in the Medicare data, especially if the final primary discharge International Classification of Diseases, 9th Revision code was not in the range of 430 to 436.
There are some important limitations to this study. First, by definition, the study was limited to Medicare-aged subjects and so findings may not generalize to the >30% of acute ischemic stroke hospitalizations that occur in subjects <65 years of age.13 Second, as mentioned previously, approximately one third of the GWTG-Stroke admissions were not linked to the Medicare data, and some of these admissions remain undetected in the unlinked Medicare cohort. Third, although measures of stroke severity such as the National Institutes of Health Stroke Scale are critical determinants of stroke outcomes, we were unable to assess the comparability of stroke severity between the linked and unlinked cohorts because these data are not available in Medicare claims. However, it could be argued that given the almost identical distributions for many of the patient characteristics and in-hospital outcomes between the linked and unlinked cohorts, that substantial differences in the distribution of National Institutes of Health Stroke Scale are unlikely.
In summary, although the characteristics of hospitals that participate in GWTG-Stroke are considerably different from hospitals that do not participate, Medicare beneficiaries with acute ischemic stroke treated at GWTG-Stroke hospitals were very similar to other Medicare beneficiaries with acute ischemic stroke with respect to demographics, comorbidities, and in-hospital outcomes. Overall, our findings suggest that Medicare-aged GWTG-Stroke ischemic stroke admissions are generally representative of the national fee-for-service Medicare ischemic stroke population.
Sources of Funding
The Get With The Guidelines–Stroke (GWTG-Stroke) program is provided by the American Heart Association/American Stroke Association. The GWTG-Stroke program is currently supported in part by a charitable contribution from Ortho-McNeil. GWTG-Stroke has been funded in the past through support from Boehringer-Ingelheim and Merck, Bristol-Myers Squib/Sanofi Pharmaceutical Partnership, and the American Heart Association Pharmaceutical Roundtable.
M.J.R. receives salary support from the Michigan Stroke Registry and serves on several American Heart Association Get With The Guidelines (AHA GWTG) Committees. G.C.F. serves as a member of the AHA GWTG Steering Committee; receives research support from the National Institutes of Health (NIH; significant) and the Agency for Healthcare Research and Quality (significant); and serves as a consultant to Novartis (significant). E.E.S. serves as a member of the GWTG Subcommittee and receives research support from the NIH and the Canadian Stroke Network and salary support from the Heart and Stroke Foundation of Canada and the Canadian Institute for Health Research. W.P. is a member of the Duke Clinical Research Institute (DCRI), which serves as the AHA GWTG data coordinating center. D.O. is a member of the DCRI, which serves as the AHA GWTG data coordinating center. A.F.H. is a member of the DCRI, which serves as the AHA GWTG data coordinating center. A.F.H. is a recipient of an AHA Pharmaceutical Roundtable grant. E.D.P. has received research grants from Bristol-Myers Squibb, Sanofi-Aventis, Merck, and Ortho McNeil Pharmaceuticals. E.D.P. serves as Principal Investigator of the Data Analytic Center for AHA GWTG. L.H.S. serves as chair of the AHA GWTG Steering Committee and serves as a consultant to the Massachusetts Department of Public Health.
Stephen M. Davis, MD, was the Guest Editor for this paper.
- Received May 23, 2011.
- Revision received July 28, 2011.
- Accepted August 5, 2011.
- © 2012 American Heart Association, Inc.
- Bufalino VJ,
- Masoudi FA,
- Stranne SK,
- Horton K,
- Albert NM,
- Beam C,
- et al
American Stroke Association. Get With The Guidelines–Stroke. Available at: www.heart.org/HEARTORG/HealthcareProfessional/GetWithTheGuidelinesHFStroke/GetWithTheGuidelinesStrokeHomePage/Get-With-The-Guidelines-Stroke-Home-Page_UCM_306098_SubHomePage.jsp. Accessed July 7, 2011.
- Schwamm LH,
- Fonarow GC,
- Reeves MJ,
- Pan W,
- Frankel MR,
- Smith EE,
- et al
- Fonarow GC,
- Reeves MJ,
- Smith EE,
- Saver JL,
- Zhao X,
- Olson DW,
- et al
- Gliklich RE,
- Dreyer NA
- Curtis LH,
- Greiner MA,
- Hammill BG,
- DiMartino LD,
- Shea AM,
- Hernandez AF,
- Fonarow GC
American Hospital Association. AHA Hospital Statistics 2007. Chicago, IL: AHA; 2007.
- Kociol RD,
- Hammill BG,
- Fonarow GC,
- Klaskala W,
- Mills RM,
- Hernandez AF,
- Curtis LH
- DeFrances CJ,
- Lucas CA,
- Buie VC,
- Golosinskiy A