Data quality in the American Heart Association Get With The Guidelines-Stroke (GWTG-Stroke): Results from a National Data Validation Audit

Transcription

1 Data quality in the American Heart Association Get With The Guidelines-Stroke (GWTG-Stroke): Results from a National Data Validation Audit Ying Xian, MD, PhD, a Gregg C. Fonarow, MD, b Mathew J. Reeves, PhD, c Laura E. Webb, CCRP, a Jason Blevins, MPH, a Vladimir S. Demyanenko, MS, a Xin Zhao, MS, a DaiWai M. Olson, PhD, RN, a Adrian F. Hernandez, MD, MHS, a Eric D. Peterson, MD, MPH, a Lee H. Schwamm, MD, d and Eric E. Smith, MD, MPH e Durham, NC; Los Angeles, CA; East Lansing, MI; Boston, MA; and Alberta, Canada Background Get With The Guidelines (GWTG)-Stroke is a national stroke registry and quality improvement program. We examined the accuracy and reliability of data entered in GWTG-Stroke. Methods Data entered by sites in the GWTG-Stroke database were compared with that abstracted from de-identified medical records by trained auditors. Accuracy for each individual data element and a composite accuracy measure were calculated. Reliability was assessed using kappa (κ) statistics for categorical variables and intraclass correlation (ICC) for continuous variables. Results A random selection of 438 medical records from 147 GWTG-Stroke hospitals was obtained. Overall accuracy was above 90% for all variables abstracted except for weight (84.9%), serum creatinine (88.1%), deep venous thrombosis prophylaxis (79.0%), and date/time last known well (85.3%). Intermediate to good (κ or ICC ) or excellent agreement (κ or ICC 0.75) was observed for nearly all audited variables, including time-related performance measures such as arrival within 2 hours of symptom onset (κ = 0.90) and door-to-needle time 60 minutes (κ = 0.72). The overall composite accuracy rate was 96.1%. The composite measure varied slightly by region and hospital academic status, but there were no significant differences in composite accuracy by bed size, ischemic stroke volume, primary stroke center certification, or Coverdell Registry participation. Conclusions This audit establishes the reliability of GWTG-Stroke registry data. Individual data elements with suboptimal accuracy should be targeted for further data quality improvement. (Am Heart J 2012;163: e1.) Stroke is a leading cause of serious long-term disability and the fourth leading cause of mortality in the United States. 1 Despite the wide dissemination of evidence-based guidelines, 2,3 many patients with stroke do not receive recommended therapies. 4,5 The Get With The Guidelines (GWTG)-Stroke program is a national stroke registry and performance improvement program developed by the American Heart Association From the a Duke Clinical Research Institute, Durham, NC, b Division of Cardiology, University of California, Los Angeles, CA, c Department of Epidemiology, Michigan State University, East Lansing, MI, d Department of Neurology, Massachusetts General Hospital, Boston, MA, and e Department of Clinical Neurosciences, Hotchkiss Brain Institute, University of Calgary, Alberta, Canada. John S. Rumsfeld, MD, PhD served as guest editor for this article. Submitted September 15, 2011; accepted December 20, Reprint requests: Ying Xian, MD, PhD, Duke Clinical Research Institute, 2400 Pratt St, Durham, NC ying.xian@dm.duke.edu /$ - see front matter 2012, Mosby, Inc. All rights reserved. doi: /j.ahj and the American Stroke Association (AHA/ASA) to promote stroke quality improvement efforts, reduce disparities in stroke care, and improve clinical outcomes through increased adherence to guideline recommendations. 6-8 During the last decade, dramatic improvement in multiple processes of care has been documented in the GWTG-Stroke program. 7 Nevertheless, participation in GWTG-Stroke is voluntary. Hospitals participate based on their level of interest in stroke quality improvement and their capacity to fulfill the requirements. Similar to the Paul Coverdell National Acute Stroke Registry and the Joint Commission's Primary Stroke Center Certification Program, the GWTG-Stroke registry relies on user-reported data. Collectively, more than 2 million stroke admissions have been captured in these registries. However, published data on the accuracy and reliability of these user-submitted data are limited. 9 Therefore, it is important to evaluate the quality of data submitted by participating hospitals and judge the reliability of the

2 American Heart Journal Volume 163, Number 3 Xian et al 393 information obtained. Moreover, GWTG-Stroke data are recorded from a wide variety of hospitals including a mix of academic and nonacademic hospitals, Joint Commission certified stroke centers, Paul Coverdell Registry hospitals, and small- and large-sized hospitals. It remains unclear as to what degree data quality is consistent across centers. To assess the validity of the GWTG-Stroke data, we performed an audit on a random sample of records from a random sample of hospitals in the GWTG-Stroke registry. Specifically, we sought to (1) assess the data accuracy and reliability based on the submitted chart documentation source, (2) evaluate the variation in data accuracy across hospitals; (3) determine the degree to which hospital performance measures differed due to disagreement between audit and site data abstraction, and (4) identify areas for future data quality improvement. Methods GWTG-Stroke program Details of the GWTG-Stroke program have been previously described. 6-8,10,11 In brief, the AHA/ASA developed the GWTG-Stroke program focused on the redesign of hospital systems of care to improve the quality of care for patients with stroke. The GWTG-Stroke program includes learning sessions, didactic sessions, best-practice sharing, interactive workshops, postmeeting follow-up, and an Internet-based Patient Management Tool (Outcome, Sciences Inc, Cambridge, MA). This Internet-based tool provides the opportunity for concurrent data collection, ongoing real-time feedback, and clinical decision support to enable rapid cycle improvement. Trained hospital personnel are instructed to use the Patient Management Tool to collect data on consecutive patients admitted to the hospital with a principal diagnosis code of stroke. The eligibility of each admission is confirmed through chart review. Data abstracted include demographics, medical history, brain imaging, in-hospital treatment and events, discharge treatment and counseling, mortality, and discharge destination. The data abstraction tool includes predefined logic features and user alerts to identify potentially invalid data formats or values. Required fields are structured so that valid data must be entered before the data can be saved as a complete record and entered into the database. Edit checks are used to identify inconsistent or outof-range data and prompt the user to correct or review data entries that are outside a predefined range. Feedback on completeness and quality of the data are provided to participating hospitals on a regular basis. Training in the use of the tool is provided online and via telephone for all users. Outcome Sciences, Inc, serves as the data collection and coordination center for GWTG. The Duke Clinical Research Institute serves as the data analysis center and has an agreement to analyze the aggregate de-identified data for research purposes. GTWG-Stroke chart data validation audit This study is part of the AHA GWTG ongoing data quality assurance efforts. A random sample of 223 GWTG-Stroke hospitals with a minimum of 10 records submitted during the 12-month period from October 2008 to September 2009 was generated for the audit. Among 1,342 GWTG-Stroke hospitals during the time period, only 202 sites did not meet the minimum for eligibility for audit. The random list of 223 hospitals was reviewed by the GWTG Executive Committee to assure representation of participating hospitals, with consideration given to hospital region, bed size, available services, academic vs nonacademic status, and stroke volume. A randomization algorithm was developed to generate a list of 3 stroke admissions to be requested from each qualifying hospital. All stroke admissions including ischemic stroke, transient ischemic attack, intracerebral hemorrhage, and subarachnoid hemorrhage were eligible for the auditing. Among those 3 selected, at least 1 admission was of a tissue plasminogen activator (tpa) eligible patient with ischemic stroke. The resulting list of hospitals and admissions was provided to Outcome Sciences, Inc, to request de-identified copies of the medical records from each hospital. To minimize the burden on audited sites, the information requested was limited to specific documents in the chart that were expected to contain source data for the fields of interest; this included the emergency department records; admission history and physical, computed tomography (CT), and neurology reports; laboratory reports; and discharge summaries and instructions. Sites were blinded to the elements that would be abstracted centrally. Once received by Outcome Sciences, Inc, the hospital and patient identifiers were removed. These medical records were scanned and provided to Duke Clinical Research Institute for data validation audit. Trained auditors were instructed to review the data contained in the GWTG- Stroke database against medical record files using a standardized data auditing instrument specifically designed for this study. Using the instrument, data auditors reabstracted the data from chart review and determined whether each audited data element was correct, incorrect, or not able to be verified based on the contents of the source medical records received. Data assessed included patient demographics, symptom onset and arrival timeline, medical history, diagnosis and evaluation, thrombolytic use and treatment timeline, medications, laboratory results, in-hospital and discharge care, and discharge status. Statistical analysis Using audit-abstracted values as the reference standard of information on the specific measures of interests, we report the accuracy of data submitted by the hospital to the GWTG-Stroke database. An accurate data element was defined as present when data abstracted from medical chart review by auditor and site were in exact agreement or both agreed that the element was missing. Records where the data value entered in the GWTG-Stroke database could not be identified in the available source documents were excluded from the accuracy calculations. Overall, this affected only 4% of data values. These records were excluded to avoid potential bias because the data value reported to the GWTG-Stroke database might have been derived from other parts of the medical record not available for central review. For continuous variables such as blood pressure, data were considered correct if it was within ±10% of the value recorded in the GWTG database by the hospital. We defined an accurate time as being within 15 minutes based on prior work that had proposed, a specific time that could be

3 394 Xian et al American Heart Journal March 2012 identified with certainty within (±) 15 minutes. 12 In addition, an opportunity-based composite accuracy measure, defined as the total number of accurate data elements divided by the total number of possible data elements among eligible records, was calculated. This composite measure was used to summarize the accuracy rates of all data elements into a single variable. 7-8 To assess whether data quality varies by hospital characteristics, we determined whether composite accuracy was associated with hospital academic status, bed size, annual ischemic stroke discharge volume, the Joint Commission primary stroke center certification, and Paul Coverdell Stroke Registry participation. We calculated the interrater reliability using kappa (κ) statistics for categorical variables and intraclass correlations (ICCs) for continuous variables. 9,13 A κ greater than 0.75 represents excellent agreement beyond chance, a κ below 0.40 represents poor agreement, and a κ of 0.40 to 0.75 represent intermediate to good agreement. The interpretation of an ICC is similar to κ statistics. In addition, we determined the degree to which hospital performance measures were altered due to inconsistency in time-sensitive variables using exact time intervals (rather than accurate within 15 minutes) for the following measures: patients arriving within 2 hours after symptom onset, patients arriving within 2 hours after symptom onset and receiving treatment with intravenous (IV) tpa within 3 hours of symptom onset, door to CT 25 minutes, and door to needle 60 minutes. All tests were evaluated at a 2-sided significance level of P b.05. The analyses were performed at the Duke Clinical Research Institute using SAS 9.2 software (SAS Institute, Cary, NC). The authors are solely responsible for the design and conduct of this study, all study analyses, the drafting and editing of the paper, and its final contents. Results Of the 223 hospitals originally selected (Figure 1), 38 (17.0%) did not respond, 20 (9.0%) declined to participate, and 18 (8.1%) who initially agreed to participate never sent any records. This left 147 (65.9%) hospitals who participated in this voluntary audit effort, representing 11.0% (147/1,342) of GWTG-Stroke hospitals. From these centers, a total of 441 medical records were requested and 438 (99%) records were received and included in the audit, accounting for 0.15% (438/294,670) of total cases in the time period. The overall patient and hospital characteristics and performance measures were similar among audited sites (n = 147), sites selected for data validation (n = 223), and overall GWTG-Stroke hospitals (n = 1,342), indicating that the available data were likely representative of the larger registry (see online Appendix for pairwise comparisons). Because of the large sample size, some differences may be statistically significant but not clinically meaningful. Overall, less than 4% of the audited data values were unable to be verified from the source documents. The data elements that were most frequently not obtainable Figure 1 All hospitals approached (n = 223) Hospitals included (n = 147) Medical records received (n = 438) Hospitals excluded (n = 76) Declined: 20 No response: 38 Responded but did not send records: 18 Study diagram. were Hispanic ethnicity (46.3%; 203/438), body weight (20.8%; 91/438), and date/time last known to be well (19.4%; 85/438). Across all audited data elements, the range of accuracy rates was from 79.0% to 99.5% (Table I). Except for deep venous thrombosis (DVT) prophylaxis (79.0%), date/time last known well (85.3%), and serum creatinine (88.1%), the accuracy of audited variables was above 90%. The direction and magnitude of the differences between values appeared to be random for date/time last known well and serum creatinine (data not shown). In the case of disagreement on DVT prophylaxis, hospitals were significantly more likely to inaccurately report that DVT prophylaxis had not been given when, in fact, it had (71/76 cases) than to report that it had been given when in fact it was not (5/76 cases, P b.001). Measures of interrater reliability using κ statistics and ICC found excellent agreement (κ or ICC 0.75) or intermediate to good agreement (κ = ) for all audited variables. The only exception was serum creatinine (ICC 0.04). However, a close examination of medical records found 2 probable data entry errors for serum creatinine (63 and 67 mg/dl, probably mean to represent 0.63 and 0.67 mg/dl, respectively). After excluding these 2 entries outside the physiologic range for serum creatinine likely caused by data entry error, the ICC of serum creatinine was 0.72 ( ), indicating that the 2 erroneous entries were influential outliers that heavily leveraged the ICC toward 0. The overall composite accuracy rate, a summary score of all data elements, was 96.1%. There was a slight variation in composite accuracy measure by regions (Table II). Academic hospitals had slightly higher accuracy than nonacademic hospitals. However, there were no differences by bed size, annual ischemic stroke

4 American Heart Journal Volume 163, Number 3 Xian et al 395 Table I. Accuracy and reliability for audited variables in 438 medical records Variable Accuracy (%) Interrater reliability (95% CI) Demographics Age Gender ( ) Race ( ) Hispanic ethnicity ( ) Insurance ( ) Medical history ( ) Diagnosis and evaluation Final clinical diagnosis ( ) Initial NIHSS ( ) Medication before admission Antiplatelet ( ) Anticoagulation ( ) Antihypertensive ( ) Cholesterol reducer ( ) Diabetic medication ( ) Symptom timeline Date/time patient last known 85.3 to be well Arrival date/time 93.6 Brain imaging Brain imaging completed at ( ) your hospital for this episode of care? Date/time brain imaging 90.4 Completed interpretation of ( ) first brain image after symptom onset, done at any facility IV thrombolytic therapy IV tpa initiated at this hospital ( ) IV tpa at an outside hospital ( ) Date/time IV tpa initiated 99.5 DVT prophylaxis Was DVT prophylaxis ( ) initiated by the end of hospital day 2? Measurement LDL ( ) Serum creatinine ( ) Systolic blood pressure ( ) Weight ( ) Discharge information Discharge destination ( ) Comfort measures at the ( ) time of discharge Discharge treatment Antithrombotic medication ( ) Persistent or paroxysmal ( ) atrial fibrillation/flutter If atrial fibrillation/flutter, ( ) was discharged on anticoagulation? Antihypertensive ( ) treatment Cholesterol-reducing ( ) treatment Diabetic treatment ( ) Antismoking treatment ( ) Stroke rehabilitation Table I. (continued) Variable Accuracy (%) Patient was assessed for or received rehabilitation service? Interrater reliability (95% CI) ( ) Accuracy was defined as the data value abstracted in the audit and the data value recorded in the GWTG database being in exact agreement. NIHSS indicates National Institutes of Health stroke scale; LDL, low-density lipoprotein. κ Statistics for categorical variables and ICCs for continuous variables. Medical history includes atrial fibrillation/flutter, prosthetic heart valve, coronary artery disease/prior myocardial infarction, carotid stenosis, diabetes mellitus, peripheral vascular disease, hypertension, smoker, dyslipidemia, heart failure, sickle cell disease, current pregnancy (or up to 6 weeks postpartum), previous stroke, and previous transient ischemic attack. Accurate to within 15 minutes. Intraclass correlation 0.72 ( ) after excluding 2 probable data entry errors by hospitals that were outside the physiologic range for creatinine (63 and 67 mg/dl). When these 2 erroneous entries are included, the ICC is 0.04 (95% CI 0.06 to 0.14). Table II. Overall data accuracy among GWTG-Stroke hospitals Hospital type Level Composite accuracy Overall 96.1 No. of ischemic stroke discharges No. of beds Region West 96.1 b.001 South 95.5 Midwest 96.6 Northeast 96.5 Teaching status Academic Nonacademic 95.8 Primary stroke center Yes No 96.1 Paul Coverdell hospital Yes No 96.1 discharges volume, primary stroke center certification, or Coverdell Registry participation. We evaluated the degree to which time-sensitive performance measures varied between the GWTG-Stroke database and the data abstracted by the audit (Table III). Overall, compliance with specific performance measures was changed for a total of 38 records when an exact match in the time variable was required. These changes in timeframe were distributed in such a way that 21 records would have been beyond the targeted timeline, and another 17 records would have been fallen within the targeted timeline (P =.63). Except for door to IV tpa 60 P

5 396 Xian et al American Heart Journal March 2012 Table III. Reliability of time-sensitive performance measures Measure Interrater reliability (95% CI) Arrival by 2 h (yes/no) 0.90 ( ) Arrival by 2 h, IV tpa by 3 h (yes/no) 1.00 ( ) Door to CT 25 min (yes/no) 0.80 ( ) Door to IV tpa 60 min (yes/no) 0.72 ( ) minutes (κ = 0.72), the κ statistics indicated excellent agreement (κ 0.75) on time-sensitive performance measures based on audited records. Discussion Clinical registries have been increasingly used as a scientific resource to foster innovative clinical research as well as a tool to monitor and improve the quality of care. 7,8 A major advantage of a registry is the ability to collect detailed clinical information similar to a clinical trial while including a wider range of patients seen in real world settings. However, the use of a clinical registry largely depends on the quality of its data. This data validation audit, using randomly selected hospitals and admissions from the GWTG-Stroke registry, confirms the generally high quality of data collected in the GWTG- Stroke database. Our study supports 2 key observations. First, the overall accuracy and reliability of the GWTG-Stroke database data elements are generally high. Except for a few variables, the accuracy of all data elements was above 90%, with excellent agreement between data collected by hospitals and medical records review. These findings appear comparable with state-level quality audits of GWTG-Stroke and Paul Coverdell National Stroke Registry, in which generally good agreement was found between data entered into the GWTG-Stroke database and data abstracted from medical record audit review. 9,14 Despite slight variations in data accuracy by region and hospital academic status, we found that a high level of accuracy can be achieved in hospitals regardless of bed size, volume, or participation in other stroke care initiatives. Second, there has been concern that submission of incorrect data without external chart review may cause patients to be included or excluded erroneously from quality or performance measures, thus benefiting hospitals through falsely elevated performance rates. However, this concern is mitigated by our observation of high-level agreement among timesensitive performance measures. Importantly, even among the 38 records with time-sensitive measures affected by incorrect data abstraction, the net effect was random with respect to performance (P =.63),in that based on audit abstraction, 55% (n = 21) of the cases would have shifted from success to failure to reach the targeted treatment timeline, and 45% (n = 17) would have shifted from failure to success at reaching the timeline. In the case of DVT prophylaxis, hospitals were more likely to underreport adherence to the DVT performance measure than overreport adherence, suggesting that the quality improvement award incentives incorporated into GWTG-Stroke were not biasing the data collection. In summary then, data collected in the GWTG-Stroke registry appear to be valid and reliable as a resource for clinical research, surveillance, and quality improvement needs. Despite these findings, we did identify some areas where data quality was suboptimal. Weight was one of the variables that were most frequently missing or unverifiable at the chart review. Date and time last known to be well and brain imaging date/time were also a challenge because there was a relatively high percent of cases with unverifiable values. The missing rates for date and time last known well may be even higher in the overall GWTG-Stroke database than in the audited sample because the audit selection strategy required submission of at least 1 tpa-eligible patient, and these tpa-eligible patients are probably more likely to have complete information on symptom timelines. Even when documented, there may be differing times or time ranges recorded in different locations in the medical record. Therefore, values may vary or appear out of sync within the audited documentation. For example, if the original source was an estimate, such as between 10 and 10:30 AM, there could be differing opinions as to how best to abstract a single value. These results indicate the challenge in documentation of stroke time related events. 9,12 More work, however, is required to define the ideal set of data definitions and data collection templates for these critical timerelated variables. We excluded records where the data value entered by sites could not be identified from the available source documents for 2 reasons. First, the specific sections of the medical records received from the hospitals may not have been the ideal source for a particular variable, and had the entire medical record been available for review, a higher level of accuracy might have resulted. For example, if registration data were not received and race/ethnicity were only captured in this location, they would not have been present in the submitted source documents. Second, in some cases, nonprotected health information such as insurance status and race was inadvertently blinded, making data audit impossible for that variable. Although only affecting less than 4% of data values, this approach might have introduced potential bias for the analysis, especially among variables with a higher percentage of missing data. However, it is reassuring that even among those variables with the most

6 American Heart Journal Volume 163, Number 3 Xian et al 397 inaccurate or missing fields, the overall reliability was good or even excellent. Systematic data auditing is an ongoing effort of GWTG- Stroke data quality assurance. This is in addition to ongoing training through local and national meetings and webinars. The GWTG data validation audit is designed to explore the quality of the GWTG-Stroke registry, identify areas for improvement, understand any potential limitations in current data collection processes, and provide hospitals feedback for further improvement. Many data quality assurance procedures have been developed in medical research, including standardization of data collection procedures across participating hospitals, training of data collectors, application of automatic error checks at the point data entry, internal or external data audits of randomly selected centers/admissions, provision of data quality reports and feedback, and implementation of recommended changes to resolve causes of data errors. 15,16 Many of these procedures have already been adapted by the GWTG-Stroke program and contribute to the high level of data quality found in this audit. However, given the time-consuming nature of data abstraction and the large number of medical records, the emerging use of chart abstraction from electronic medical records may provide substantial efficiency and quality benefits in future quality improvement efforts. Our study has additional limitations. First, although our data validation audit provides a unique opportunity to monitor data quality, we were limited to a relatively small sample of participating hospitals and records. To balance this limitation, a randomization algorithm was used during the site selection and case selection stage. In addition, eligible hospitals were reviewed by both the GWTG Executive Committee and the Duke Clinical Research Institute. A final list of 147 sites was available for data audit, representing more than 10% of GWTG- Stroke hospitals. All these efforts were aimed to assemble a representative study sample. Second, approximately 30% of the hospitals that were initially selected for data validation did not submit medical records in response to the auditing request. It is possible that there is nonresponse (selection) bias in that there may be higher quality data collection in hospitals that chose to participate in the audit. Nevertheless, we maintained reasonably sufficient response rates (N60%) for a survey study. 17 More importantly, as shown in the online Appendix, patient and hospital characteristics were similar between the audited and nonresponding sites, which provides additional evidence that our study sample is representative of GWTG-Stroke hospitals in general. Third, we only audited selected variables contained in the registry, so the reliability of other variables in the registry is not confirmed. Nevertheless, the audited variables were selected to represent the information that is most relevant to hospital quality and performance measures. Fourth, the accuracy of our central abstraction was dependent on the source documentation requested and provided. Although a variety of sources were requested, it is possible that the data were documented in other parts of the medical record in some cases. Accordingly, we excluded b4% of records where the data value entered in the GWTG-Stroke database could not be identified in the available source documents. Fifth, our central audit identified accuracy in the sense of the most consistent and plausible value from the source documents without independent verification; this value may not represent the true value if the source documentation itself was in error. Finally, our quality audit methodology did not allow us to assess the accuracy of case identification or if consecutive case ascertainment was used, as has been done in other studies. 18 In conclusion, the present study demonstrates overall good reliability of data abstraction in GWTG-Stroke registry. Only a limited number of fields were identified where the quality of data collection could be improved. The data collected by the GWTG-Stroke hospitals contribute to a valid, reliable, and accurate database that serves as a powerful tool to monitor and improve evidence-based stroke care in GWTG-Stroke hospitals and to support innovative clinical research in quality of care and outcomes in stroke. Financial disclosures Ying Xian, Gregg C. Fonarow, Mathew J. Reeves, Laura E. Webb, Jason Blevins, Vladimir S. Demyanenko, Xin Zhao, DaiWai M. Olson, Adrian F. Hernandez, and Eric E. Smith have no financial disclosures to state. Eric D. Peterson is a principal investigator of AHA GWTG Data Analysis Center. Lee H. Schwamm is the chair of the AHA national steering committee for GWTG (unpaid). Disclosures Sources of funding: The 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 Janssen Pharmaceutical Companies of Johnson & Johnson. The GWTG-Stroke program has been funded in the past through support from Boeringher-Ingelheim, Merck, Bristol-Myers Squib/Sanofi Pharmaceutical Partnership, and the American Heart Association Pharmaceutical Roundtable. References 1. Minino AM, Xu J, Kochanek KD. National Vital Statistics Reports. CDC; Vol. 59, Number Sacco RL, Adams R, Albers G, et al. Guidelines for prevention of stroke in patients with ischemic stroke or transient ischemic attack: a

7 398 Xian et al American Heart Journal March 2012 statement for healthcare professionals from the AHA/ASA council on stroke. Stroke 2006;37: Adams Jr HP, del Zoppo G, Alberts MJ, et al. Guidelines for the early management of adults with ischemic stroke: a guideline from the AHA/ASA stroke council, clinical cardiology council, cardiovascular radiology and intervention council, and the atherosclerotic peripheral vascular disease and quality of care outcomes in research interdisciplinary working groups. Stroke 2007;38: Reeves MJ, Arora S, Broderick JP, et al. Acute stroke care in the US: results from 4 pilot prototypes of the Paul Coverdell National Acute Stroke Registry. Stroke 2005;36: Hinchey JA, Shephard T, Tonn ST, et al. Benchmarks and determinants of adherence to stroke performance measures. Stroke 2008;39: LaBresh KA, Reeves MJ, Frankel MR, et al. Hospital treatment of patients with ischemic stroke or transient ischemic attack using the Get With The Guidelines program. Arch Intern Med 2008; 168: Schwamm LH, Fonarow GC, Reeves MJ, et al. Get With The Guidelines-Stroke is associated with sustained improvement in care for patients hospitalized with acute stroke or transient ischemic attack. Circulation 2009;119: Fonarow GC, Reeves MJ, Smith EE, et al. Characteristics, performance measures, and in-hospital outcomes of the first one million stroke and transient ischemic attack admissions in Get With The Guidelines-Stroke. Circ Cardiovasc Qual Outcomes 2010;3: Reeves MJ, Mullard A, Wehner S. Inter-rater reliability of data elements from a prototype of the Paul Coverdell National Acute Stroke Registry. BMC Neurol 2008;8: Smith EE, Liang L, Hernandez A, et al. Influence of stroke subtype on quality of care in the Get With The Guidelines-Stroke program. Neurology 2009;73: Reeves MJ, Smith E, Fonarow G, et al. Off-hour admission and inhospital stroke case fatality in the Get With The Guidelines-Stroke program. Stroke 2009;40: Rosamond WD, Reeves MJ, Johnson A, et al. Documentation of stroke onset time: challenges and recommendations. Am J Prev Med 2006; 31:S McGraw KO, Wong SP. Forming inferences about some intraclass correlation coefficients. Psychol Methods 1996;1: Yoon SS, George MG, Myers S, et al. Analysis of data-collection methods for an acute stroke care registry. Am J Prev Med 2006;31: S Arts DGT, de Keizer NF, Scheffer GJ. Defining and improving data quality in medical registries: a literature review, case study, and generic framework. J Am Med Inform Assoc 2002;9: Willis CD, Jolley DJ, McNeil JJ, et al. Identifying and improving unreliable items in registries through data auditing. Int J Qual Health Care Response rates an overview. American Association for Public Opinion Research (AAPOR). Available at: Response_Rates_An_Overview.htm. 18. Reeves MJ, Wehner S, Organek N, et al. Accuracy of identifying acute stroke admissions in a Michigan Stroke Registry. Prev Chronic Dis 2011;8:A62.

8 American Heart Journal Volume 163, Number 3 Xian et al 398.e1 Appendix Appendix Table. Patient characteristics, hospital performances, and characteristics among audited, selected, and overall GWTG-Stroke hospitals Variable Level Audited GWTG-Stroke hospitals (n = 147), % or value Selected GWTG-Stroke hospitals (n = 223), % or value P Overall GWTG-Stroke hospitals (n = 1342), % or value P Patient characteristics Age (y) Median b th-75th Sex Female Race/Ethnicity White b b.001 Black Asian Hispanic Diagnosis Ischemic stroke b b.001 Transient ischemic attack Subarachnoid hemorrhage Intracerebral hemorrhage Stroke not otherwise specified No stroke-related diagnosis Discharge status Home b b.001 Skilled nursing facility Rehabilitation Transfer to acute care facility Left against medical advice Hospice Died Length of stay (d) Median b th-75th Acute performance measures IV tpa 2 h Yes Early antithrombotic Yes DVT prophylaxis Yes Discharge performance measures Antithrombotics Yes Anticoagulation for Yes atrial fibrillation/flutter Smoking cessation Yes Lipid-lowering drug for LDL 100 mg/dl Yes Summary of performance measures Composite measure Mean b Defect-free measure Mean b.001 Additional quality measures Door to CT 25 min Yes b b.001 Dysphagia screening Yes Stroke education Yes Rehabilitation assessment Yes Hospital characteristics No. of ischemic stroke discharges No. of beds Median Region West South Midwest Northeast Hospital type Academic LDL indicates Low-density lipoprotein. May not add up to 1 because of missing values. P value for comparison with the audited hospitals. Comparisons are by χ 2 test for tests of differences in proportions or t test for comparisons of continuous variable means.