A User's Guide to the NINDS rt-PA Stroke Trial Database

Robert Dachs and colleagues describe the process for researchers to acquire and decipher the dataset from the original NINDS rt-PA trial.

I n December 1995, results from the National Institute for Neurological Disorders (NINDS) Recombinant Tissue Plasminogen Activator (rt-PA) Stroke Trial were published [1]. The implications of this trial have been profound, affecting the emergency management of stroke patients as well as any patient with apparent symptoms of an ischemic stroke.
Based on the NINDS fi ndings, the United States Food and Drug Administration approved the use of rt-PA for the treatment of acute ischemic stroke. Subsequent to this approval, both the American Heart Association and the American Academy of Neurology endorsed the use of rt-PA for selected stroke patients [2,3].
The approval and endorsement of rt-PA for stroke therapy has been controversial. Since the publication of the NINDS rt-PA trial, concerns regarding the study's design, results, and ramifi cations have been published and disseminated. Specifi c criticisms of the NINDS trial have included: (1) lack of early improvement (within 24 hours) in patients treated with rt-PA; (2) a trial enrollment policy that required half of the participants to be treated within 0-90 minutes of presentation and the remainder at 91-180 minutes-thereby not refl ecting a "real world" setting; (3) discrepancies in baseline characteristics of patients favoring those treated with rt-PA; (4) a small sample size with resulting large confi dence intervals; and (5) a reported benefi t of therapy, a fi nding inconsistent with previous thrombolytic trials in stroke [4][5][6][7][8].
Additional concerns have been expressed regarding the generalization of the NINDS results to the community hospital setting, the reorientation of emergency medical systems, and an unacceptably high risk of harm with limited benefi t to patients. The NINDS authors have attempted to address many of these concerns [9], and supporters have defended and encouraged the increased use of rt-PA in patients with acute stroke [10,11].
In hopes of clarifying many of the issues raised, researchers who were not involved in the NINDS study have made requests for access to the NINDS rt-PA trial data. Access to this data was initially denied, even after petition to the Food and Drug Administration under the Freedom of Information Act, heightening concern and suspicions regarding the trial fi ndings [12]. During this period, the NINDS authors published no less than 27 reports using various permutations of the NINDS dataset [9,.
In October 2003, nearly eight years after the NINDS publication, the original data from the NINDS rt-PA Stroke Trial were made available to the public. Since this time, we are aware of only three reports by non-NINDS researchers that have used the original dataset [39][40][41].
We document our efforts to access and decipher the NINDS rt-PA Stroke Trial dataset. By documenting the steps necessary to obtain and organize the data, along with a description of each data point, we hope to assist future researchers and clinicians who desire to further analyze this landmark trial.

Obtaining the NINDS rt-PA Database
A CD-ROM containing the NINDS rt-PA database can be purchased from the National Technical Information Service (http://www.ntis.gov/; 1-888-584-8332). The cost is US$79.00 plus US$5.00 for shipping and handling, and delivery can be expected within seven to ten business days. However, the data on the CD-ROM are not in a readable format. Multiple conversions are required, which can only be accomplished * To whom correspondence should be addressed. E-mail: dachsmd@aol.com The Guidelines and Guidance section contains advice on conducting and reporting medical research.

Summary Points
• In 1995, the NINDS rt-PA study documented an improvement in neurologic outcomes at three months for those patients given rt-PA within 3 hours of onset of an acute ischemic stroke.
• This trial has become the sentinel study supporting the use of rt-PA in acute ischemic stroke.
• Questions raised regarding this trial have been diffi cult to answer, since the raw data have been unavailable to non-NINDS researchers.
• In October 2003, the data from this trial became available for purchase.
• In this report, we describe the process of acquiring and deciphering the dataset from the original trial, provide a detailed description of the dataset, and discuss the implications for potential future research.
with technical assistance from the software company (SAS Institute,; http://www.sas.com/). Once these conversions are completed and the data are placed in an Excel spreadsheet format, no key accompanies the dataset to decode the numeric designations given to the variables. The steps necessary to obtain the NINDS database, software requirements, necessary conversions, additional instructions, incomplete/missing data, and methods used to decode data variables are documented in Text S1.

Patient Results in the Database
There are 100 separate variables recorded for the 624 patients enrolled in the NINDS study. Text S2 documents each individual variable in the order in which it is displayed in the Excel format.
Patient results can be divided into nine natural groupings: A, baseline characteristics; B, patient demographics; C, baseline vital signs and laboratory values; D, head computerized tomography results; E, treatment randomization and timing of drug delivery; F, deleterious effects (death and hemorrhage); G, neurological outcomes; H, dosing of rt-PA and use of heparin; and I, stroke type and clinical outcomes. Text S3 summarizes the data for each of the above groups and details the location of missing data, inconsistencies with previous reported data, and protocol violations not previously reported. Texts S4 through S12 present the results of every variable in the groups noted above.

Discussion
This report is intended to demonstrate the process for establishing access and analyzing the NINDS rt-PA Stroke Trial dataset. While the purchase of the database is simple, the process of gaining access to the information on the CD-ROM is convoluted and diffi cult. We have summarized these steps and provided access to the entire database for researchers desiring to analyze the data from this landmark trial for independent research queries.
Some may suggest that a reanalysis of the NINDS rt-PA database is not necessary and that those who review the data are immersed in a futile attempt to topple or tarnish the study's impact. However, the NINDS rt-PA trial has guided the worldwide care of acute ischemic stroke victims since its publication. The study has been referenced over 1,900 times [42] and stands alone as the largest randomized control trial to demonstrate rt-PA's benefi t in patients with acute ischemic stroke.
The data from this trial have been incorporated into infl uential pooled analyses [31] and reviews [43], with the salutary results widely promoted by groups such as the American Heart Association and American Stroke Association as the basis of the "Brain Attack" campaign and the creation of stroke centers. Until the results of future prospective trials such as IST-3 (The Third International Stroke Trial), ECASS-3 (The European Cooperative Acute Stroke Study), and EPITHET (Echoplanar Imaging Thrombolysis Evaluation Trial) become available, the NINDS rt-PA study will likely continue to provide the primary evidence for the use of thrombolytic therapy in the management of acute ischemic stroke.
Accessing the database as described in this report can lead to future investigation by interested researchers. There have been few studies by non-NINDS researchers using the original NINDS data. Published reports include: (1) an analysis of potential baseline imbalances in the NINDS trial, and (2) a review of severity-adjusted end points [39,40]. Both of these studies support the use of rt-PA in acute ischemic stroke. In a more recent reanalysis of the NINDS data, Hoffman and Schriger found similar responses to treatment in the placebo and rt-PA treatment groups [41]. Contradictory opinions and analyses of the original NINDS data such as these have fueled the controversy regarding the risks and benefi ts of rt-PA therapy in acute ischemic stroke. Our report provides researchers with a method to easily access the NINDS database for independent analysis in the hope that further review of this investigation will yield future insights into the risks and/or benefi ts of this intervention.
Analysis of patients in this trial with elevated blood pressure [15], elevated glucose levels [16], transient ischemic attacks [26], and minor strokes [36] have been performed by the NINDS researchers. Other subgroups of patients, such as those who received aspirin, those with congestive heart failure, those with prior strokes, and many others can now be analyzed, with the understanding that such subgroup analysis should not guide therapeutic decisions but may allow the creation of new hypotheses for future studies.
While the purchased dataset provides a large amount of data, some data appear to be missing. Documentation of inclusion and exclusion criteria in the dataset is incomplete • Patients requiring aggressive treatment to reduce their blood pressure to specifi ed limits (Box 1). This will limit any attempt to analyze protocol violations in this study. This is unfortunate given that protocol violations have been noted to be problematic in other studies, particularly in community hospital settings [44,45]. Other baseline data points that were not recorded (for example, patient's baseline living situation) can also potentially hinder future research queries. In summary, this report describes the process for researchers to acquire and decipher the dataset from the original NINDS rt-PA trial. A detailed description of the dataset has been provided to assist researchers interested in further evaluating this pivotal trial and for those researchers involved in future trials assessing the role of thrombolytic therapy for acute ischemic stroke.

Supporting Information
Text S1. Methods