Response to Letter by Sanders et al
We thank Sanders and colleagues for their interest in the recent systematic review and pooled analysis of validations of the ABCD and ABCD2 scores.1 Various points were raised about the interpretation and application of the area under the curve (AUC) statistic, the search methods, and the comparison of the ABCD and ABCD2 scores, which we will attempt to address in order.
The ABCD system is designed for use as a triage tool between primary or emergency care and specialist care settings and is therefore based on clinical information that is available at the early stage of the patient pathway. We would argue that an AUC of 0.72 does indicate “good predictive value” given that it is based on elements obtained after a history and examination and compares favorably with other predictive tools.2 We agree that the AUC statistic does have drawbacks in fully describing the clinical usefulness of a prognostic score, and other measures are relevant. However, for the purposes of a systematic review and meta-analysis of validation studies, the AUC is the standard statistic3 and it is therefore the statistic that we have included in our review.
We would concede that there is scope for further refinement of the score with the incorporation of additional information available after evaluation in secondary care such as brain or vascular imaging results and these studies are ongoing. We have shown that, in principle, the performance of the score can be improved in this way.4
The application of predictive tools in decision-making in clinical practice cannot only be based on measures of discrimination alone (ie, AUC), but also of calibration. For the ABCD2 score, this has been provided in the review according to healthcare setting (population-based studies, emergency department, rapid access, neurovascular unit, and outpatient department). The use of the score in informing decisions about admission to a stroke unit should be made according to both measures of the score’s performance when applied to an individual in a particular healthcare service and after specialist assessment.
Although large artery disease is an important etiology to identify in the evaluation of transient ischemic attack, it causes a relatively small proportion of all transient ischemic attacks and the minority of subsequent strokes.5 The ABCD2 score is not designed as a tool to triage patients for investigation for large artery disease, and it would be inappropriate to use it in this way. Interestingly, however, the score has been shown to be predictive of subsequent stroke in a cohort of patients with transient ischemic attack of large artery disease etiology.6
Sanders and colleagues questioned how thoroughly conference abstracts had been searched and pointed out the abstract of Chandratheva et al. This article from our research group showed that the ABCD2 score was not only predictive of stroke, but also that lower scores were predictive of recurrent transient ischemic attack. The cohort on which these observations were based was included in the Oxford Vascular Study (OXVASC) and Effect of urgent treatment of transient ischemic attack and minor stroke on early recurrent stroke (EXPRESS) cohorts, references 2 and 28 in the article.1
In the cohort published by Asimos and colleagues,7 the AUC of 0.59 was based on an outcome of stroke after transient ischemic attack defined by tissue-based criteria as opposed to time-based criteria. Our analysis was based on outcomes defined by time-based criteria, and this statistic was therefore not used.
Overall, the performance of the ABCD2 score was superior to the ABCD score, although this was not a statistically significant difference. The diabetes element of the score can be incorporated very easily and we would therefore recommend the use of the ABCD2 score.
Giles MF, Rothwell PM. Systematic review and pooled analysis of published and unpublished validations of the ABCD and ABCD2 transient ischemic attack risk scores. Stroke. 41: 667–673.
Altman DG, Vergouwe Y, Royston P, Moons KG. Prognosis and prognostic research: validating a prognostic model. BMJ. 2009; 338: b605.
Giles MF, Albers GW, Amarenco P, Arsava EM, Asimos AW, Ay H, Calvet D, Coutts SB, Cucchiara BL, Demchuk AM, Johnston SC, Kelly PJ, Kim AS, Labreuche J, Lavallee PC, Mas J-L, Merwick A, Purroy F, Rosamond WD, Sciolla R, Rothwell PM. Addition of brain infarction to the ABCD2 score (ABCD2I): a collaborative analysis of unpublished data on 4574 patients. Stroke. 2010; 41: 1907–1913.
Lovett JK, Coull AJ, Rothwell PM. Early risk of recurrence by subtype of ischemic stroke in population-based incidence studies. Neurology. 2004; 62: 569–573.
Asimos AW, Johnson AM, Rosamond WD, Price MF, Rose KM, Catellier D, Murphy CV, Singh S, Tegeler CH, Felix A. A multicenter evaluation of the ABCD(2) score’s accuracy for predicting early ischemic stroke in admitted patients with transient ischemic attack. Ann Emerg Med. 2010; 55: 201–210. e5.