CDR Blog

The initial derivation study (Johnston et al., 2007) for the ABCD² study was retrospective, involving cohorts from Oxford and California, attempting to validate a risk stratification tool for stroke following TIA. For the Oxford cohort, patients were assessed shortly after TIA by a stroke neurologist, whereas the California cohort had information obtained via review of records from the original treating doctor, inpatient, and outpatient records. 1916 patients were included in this study and the majority of patients were from emergency departments (EDs), with remaining patients from family practices. This study found the ABCD^2­ score to be more accurate and existing “California” and “ABCD” scores for predicting short-term risk of stroke after initial TIA. This study was validated in 2893 patients from EDs, primary care clinics, family practices, and hospital TIA clinics. C-statistics varied from 0.62-0.83 for four groups, and both 7- and 90-day stroke risk rose similarly to 2-day stroke as ABCD² score increased which allowed “low”, “medium”, and “high” risk groups to be identified.

However, a subsequent study in 2011 by Perry and colleagues (2011) found that in a cohort of 2056 patients enrolled from the ED, a high-risk cut-off score (≥5) had low sensitivity (31.6%) of detecting future stroke in patients and could not differentiate between patients at high or low risk for subsequent stroke. A low risk score (<2) also had low specificity (12.5%) for excluding future stroke. This large-sample size, prospective, ED-based trial raised concerns regarding the validity of the ABCD² score.

This has led to multiple subsequent trials and systematic reviews evaluating its effectiveness in predicting risk of future stroke. None of these studies found the ABCD² score to be sufficient for stroke-clinic triage or predicting future risk of stroke. A meta-analysis conducted by Wardlaw et al. (2014) laid out significant flaws in that:

1) The score was derived in a population with TIA diagnosed by a neurologist/specialist, therefore applying it to ED populations presenting with TIA or TIA mimic presentations in the ED is detrimental.

2) Wide variability in cut-off points were found between studies, and up to 2/3 of patients were found to be high risk (>4) across all time points, indicating a flaw in the cut-off point or a biased study population.

3) The score does not account for significant carotid stenosis, a validated independent risk factor for stroke, which could lead these patients to miss rapid assessment and endarterectomy.

In our appraisal of the derivation methodology, we applied the criteria per McGinn et al. (2000):

A) Were all important predictors included in the derivation process? The study included 19 variables that were known clinical factors independently predictive of stroke risk. These variables are also in the proposed prognostic scores for short-term risk of stroke after stroke: the California Score and ABCD score. Carotid stenosis is a significant independent risk factor that was not included. It is important to note that the treatment variables were not considered in creating the scores.

B) Were all important predictors included in a significant proportion of the study population? The derivation groups had data missing in four patients for duration of TIA symptoms and three patients for blood pressure for n=1916.

C) Were all outcome events and predictors clearly defined? Yes, the predictors are clear and left little room for inter-rater variability and interpretation. These predictors were in the categories of demographics, medical history, symptoms, blood pressure, and management at initial presentation. The outcome of interest was stroke. “Stroke was defined as a rapidly developed clinical symptom of focal (or occasionally global) disturbance of cerebral function, lasting more than 24 h or until death, with no apparent non-vascular cause, that was clearly distinguishable from the event leading to the initial diagnosis of TIA.”

D) Were those assessing outcome events blinded to the presence of predictors & vice versa? It is unclear if the study neurologists who diagnosed the outcome of stroke were blinded to the presence of predictors. It is likely they were not as stroke was diagnosed from reviewing medical and imaging records. Most data was collected prospectively. There was some data retrospectively collected but this was obtained without knowledge of the outcome.

E) Was sample size adequate to account for all potential outcomes? The derivation sample size (n=1916) appears to be adequate for the outcomes. 5%, 6%, 11% of patients had a stroke within 2, 7, 90 days respectively.

F) Does the rule make clinical sense? The ABCD2 score was developed to better predict stroke risk within 2 days after a TIA. 2-day risk is relevant when a patient presents at an emergency department and a doctor is deciding on the need for admission for urgent evaluation. This score was derived by validating and combining the California and ABCD score with the goal of better predicting 2-day risk, and makes inherent clinical sense

Based on the application of these criteria combined with the recent evidence, we determined the CDR level of the ABCD² tool to be Level 4 – Not Useful for Clinical Practice. Given it’s lack of inclusion of all important predictive factors, and multiple studies finding it to be ineffective for the prediction of high and low groups at risk of stroke following TIA, it is not recommended that this tool be used as the sole tool for risk stratification following TIA. Further use of this tool should be reconsidered in light of the evidence provided. This tool is also overshadowed by the emergence of the ABCD³I score (Kiyohara et al., 2013) which includes “dual-TIA” and imaging findings as predictors, which in some studies has shown improvement in the prediction of stroke following TIA, however, further assessment is required.

1. Dutta, D., & Bailey, S. (2016). Validation of ABCD² scores ascertained by referring clinicians: A retrospective transient ischaemic attack clinic cohort study. Emergency Medicine Journal, 33(8), 543-547. doi:10.1136/emermed-2015-205519

2. Johnston, S. C., Rothwell, P. M., Nguyen-Huynh, M. N., Giles, M. F., Elkins, J. S., Bernstein, A. L., & Sidney, S. (2007). Validation and refinement of scores to predict very early stroke risk after transient ischaemic attack. The Lancet, 369(9558), 283-292. doi:10.1016/s0140-6736(07)60150-0

3. Lee, J., & Shah, K. (2013). In Patients Presenting With Transient Ischemic Attack, Does the ABCD² Clinical Prediction Rule Provide Adequate Risk Stratification for Clinical Decision making in the Emergency Department? Annals of Emergency Medicine, 62(1), 14-15. doi:10.1016/j.annemergmed.2013.02.007

4. McGinn, T. G., Guyatt, G. H., Wyer, P. C., Naylor, C. D., Stiell, I. G., & Richardson, W. S. (2000). Users’ guides to the medical literature XXII: How to use articles about clinical decision rules. Journal of the American Medical Association, 284(1), 79–84.

5. Perry, J. J., Sharma, M., Sivilotti, M. L., Sutherland, J., Symington, C., Worster, A., . . . Stiell, I. G. (2011). Prospective validation of the ABCD² score for patients in the emergency department with transient ischemic attack. Canadian Medical Association Journal, 183(10), 1137-1145. doi:10.1503/cmaj.101668

6. Wardlaw, J., Brazzelli, M., Miranda, H., Chappell, F., Mcnamee, P., Scotland, G., Quayyum, Z., Martin, D., Shuler, K., Sandercock, P., & Dennis, M. (2014). An assessment of the cost-effectiveness of magnetic resonance, including diffusion-weighted imaging, in patients with transient Ischaemic attack and minor stroke: A systematic review, meta-analysis and economic evaluation. Health Technology Assessment, 18(27), 1–368. https://doi.org/10.3310/hta18270