Hakan Ay, MD, Assistant Professor in the Departments of Neurology and Radiology, Massachusetts General Hospital, Harvard Medical School, discusses his paper "The Causative Classification of Stroke system: an international reliability and optimization study" that was recently published in Neurology (2010; 75:1277-1284). He spoke with José Merino, MD, Science Editor of AAN.com.
AAN.com: Please describe the Causative Classification of Stroke (CCS) system.
Ay: The CCS is an evidence-based computerized algorithm that can be used to identify the most likely etiology of a stroke—taking into account the information gathered as part of a typical "stroke work-up". It was developed for use in stroke research, and designed to maximize inter-rater reliability.
Patients can be categorized into five or eight major subtypes. When the strength of the evidence to support a specific subtype is considered, strokes may be classified into 16 subtypes.
The 5-subtype CCS classifies ischemic strokes as due to large artery atherosclerosis, cardiac embolism, small artery occlusion, other uncommon cause, and undetermined cause.
The 8-subtype CCS divides the strokes due to an undetermined cause into four groups: cryptogenic-embolism, unknown-cryptogenic, unknown-incomplete evaluation, and unclassified.
The 16-subtype CCS takes into account the strength of the available evidence and classifies the major subtypes as evident, probable, or possible.
AAN.com: How does the system differ from the TOAST criteria?
Ay: The TOAST classification is a simple, logical, and useful system developed in the early 1990s to standardize the classification of stroke subtypes. While it has been used extensively, it has several shortcomings. It lacks criteria for subtype classification when there are several potential etiologies or the diagnostic work-up was not complete. In addition, some subtype definitions are open to the users' opinion or interpretation. As a result, the classification scheme has poor to moderate inter-rater reliability. When inter-rater reliability is poor, it is difficult to apply classification results to patient care or to compare studies from different investigators.
The motivation for the CCS project was the desire to create a system with higher reliability than the TOAST criteria. Evidence from the three studies we have conducted so far shows that the CCS assigns stroke subtypes with good to excellent intra-rater and inter-rater reliability.
The CCS provides an approximately 0.30 absolute increase in kappa value over the TOAST. Higher kappa means less variance in stroke research. This means that multicenter stroke studies that use the CCS instead of TOAST as an outcome measure would require a 40% smaller sample to have similar power.
The CCS is a more complex algorithm than TOAST. Etiologic stroke classification requires careful consideration of diagnostic test findings, clinical stroke features, the stroke risk of potential of etiologic factors, and the quality and completeness of diagnostic investigations.
AAN.com: What data is needed to use the computerized algorithm modules for the Causative Classification of Stroke System etiologic classification of ischemic stroke?
Ay: The computerized CCS algorithm provides a well-referenced, well-defined, and evidence-based etiologic subtype classification. The software has five modules: a log-in module to maintain confidentiality of the data; an introduction page to summarize the key features of the software and important CCS criteria for subtype classification; an interactive 10-patient training module; a certification module to measure the users' ability to understand the classification rules, identify critical data elements in case report forms, and perform etiologic stroke classification; and a one-page classification form to enter the data using checkboxes.
The CCS relies on five sources of data: clinical evaluation; brain imaging, vascular imaging, cardiac evaluation, and work-up for uncommon causes of stroke. Once the data are entered, the CCS software provides phenotypic and causative stroke subtypes.
The CCS has several user-friendly features including automatic error checking and feedback, tooltips to explain the terms and definitions used in the classification form, and automatic checking of dependent elements. In addition, user accounts may be set up to allow data storage and retrieval. The system is flexible and can generate customized versions for use in internal web-based systems in stroke networks and trials.
AAN.com: How does the current version (version 2.0) differ from version 1.0?
Ay: Version 2.0 features phenotypic subtypes in addition to causative subtypes. Additionally, in version 2.0, we have made several revisions and clarifications to the terms and definitions used in the CCS based on feedback from members of the International Stroke Genetics Consortium (ISGC) and other external users of the prior version. We believe this new version of CCS will reduce the variance from subjective interpretation of clinical data and thus provide a satisfactory basis for communication among different investigators.
AAN.com: How was the CCS computerized algorithm created?
Ay: The CCS project was launched seven years ago by a group of MGH physicians interested in developing an evidence-based etiologic classification scheme for acute ischemic stroke. The CCS is a rule-based system that can incorporate advances in the field as they emerge. In order to program the CCS algorithm, we have generated a script that defines stroke subtypes for different combinations of clinical and diagnostic stroke features.
The current version of the script includes most of the clinically relevant combinations and assigns a logical subtype in the majority of stroke patients. Nevertheless, there is a high variance in diagnostic test findings in stroke, and there might be occasional patients, particularly those with complex diagnostic findings, in whom the CCS assignment differs from expert opinion. We estimate that disagreement between expert judgment and CCS assignment occurs in approximately 2%-4% of stroke patients. We encourage all users of the CCS to provide feedback to us on their experience with the software.
AAN.com: Briefly describe the methodology and findings of your study.
Ay: We previously showed that the CCS classifies etiologic stroke subtypes with excellent inter-rater reliability in internal and external settings with limited number of raters. In the study that was just published in Neurology we wanted to find out if CCS could be used in a multicenter setting where several investigators from diverse backgrounds collect and enter the data.
Twenty members of the International Stroke Genetics Consortium from 13 centers in eight countries participated to this study. We provided each study participant a package consisting of abstracted case summaries of 50 consecutive MGH patients. The raters completed the classification form in the CCS website for each case vignette. They also stated how they would classify each stroke. Overall, we found good to excellent inter-rater reliability for both phenotypic and causative CCS subtypes. The agreement between expert opinion and CCS-assigned subtype was 96%.
AAN.com: How do you select which CCS should be used in a study (5-, 8-, or 16-subtype version)?
Ay: The 5- and 8-subtype CCS are particularly well suited for small studies testing biological hypotheses. The 16-subtype CCS is the choice for large-scale research projects where the confidence level for subtype diagnosis is important. For instance, a study testing the hypothesis that lacunar infarcts benefit from a particular treatment would profit from an initial testing in patients whose infarcts are clearly attributable to small artery occlusion.
If successful in this evident population, the intervention could be moved forward and tested in patients with probable (there is a lacunar infarct but there is also another evident etiology such as large artery atherosclerosis or a major cardiac emboli source) and possible (there is a lacunar infarct but other alternative causes are not investigated) small artery occlusion.
AAN.com: In the CCS, what is the difference between the causative and etiologic classifications?
Ay: The CCS provides both "processed" causative subtypes and "unprocessed" phenotypic subtypes.
Causative subtyping is a process that requires integration of multiple aspects of ischemic stroke evaluation including symptom characteristics, vascular risk factors, and diagnostic test findings. Phenotypic subtyping, on the other hand, designates positive test finding organized in major etiologic groups. It does not require any judgment on the part of the investigator.
Each subtyping method serves a different goal. Causative subtypes can be used to generate etiology-specific stroke management algorithms. They can also be used in research projects with limited sample size that require assessment of etiologic stroke subtypes in simultaneous context with other covariates of interest. Phenotypic subtypes are well suited for studying of interactions among etiologic factors, patient selection in large-scale epidemiologic and genetic studies, as well as for coding for administrative purposes.
AAN.com: What are the uses of the CCS in research? Is this classification system currently being used in a research study?
Ay: The CCS is particularly well suited for collaborative studies involving multiple investigators, but can be used in all stroke research studies that require etiologic subtyping. Several national and international stroke registries are switching to or have already started using the CCS. NIH-funded multicenter studies have also begun to use the CCS.
AAN.com: Can the CCS be used clinically to identify potential secondary prevention interventions (i.e., please discuss any clinical applications)?
Ay: The CCS is primarily a research tool. Its high reliability, ease of use, evidence-based design, and the unique structure that avoids contamination of etiologic subtypes with each other suggest utility as a classification tool in clinical stroke research including therapeutic intervention trials.
AAN.com: Where can researchers access the CCS?
Ay: CCS is free for academic use at http://ccs.mgh.harvard.edu.
Dr. Ay has received research support from NINDS and is an editorial board member for the journal Stroke.
Dr. Merino performed a one-time consultation with staff from Bell, Falla and Associates.
He is a member of the Stroke Publishing Technology Committee for the journal Stroke and was a member of the editorial board of Stroke from 2008-2010.
Dr. Merino has received research support from the cost reimbursement contract between NIH/NINDS Intramural Program and Suburban Hospital to support the clinical, administrative, and technical activities of the NIH Stroke Program at Suburban Hospital. He is also stroke adjudicator for the Women's Health Initiative at NIH.
Disclaimer: The opinions expressed in this posting are those of the author only and do not represent the views of the American Academy of Neurology or any of its affiliated subsidiaries.
Please login to view and submit comments.