Background and Purpose—Some patients diagnosed with transient ischemic attack (TIA) in the emergency department may actually have alternative diagnoses such as seizure, migraine, or other nonvascular spells. The ABCD² score has been shown to predict subsequent risk of stroke in patients with TIA diagnosed by emergency physicians, but perhaps high ABCD² scores simply separate those patients with true TIA from those with alternative diagnoses. We investigated this hypothesis in a cohort of patients with TIA identified in the emergency department whose records were reviewed by an expert neurologist.

Methods—Among patients diagnosed by emergency physicians with TIA in 16 hospitals in the Kaiser-Permanente Medical Care Plan over a 1-year period ending February 1998 (before publication of prediction rules), an expert neurologist reviewed all records for those in which the diagnosis of TIA was considered questionable by a medical records analyst and determined whether the spell was likely to represent a true TIA. Subsequent strokes within 90 days were identified. ABCD² scores were calculated for all patients and 2-sided Cochrane-Armitage trend tests were used to assess subsequent risk of stroke.

Results—Of the 713 patients reviewed by the expert neurologist, 642 (90%) were judged to likely have experienced a true TIA. Ninety-day stroke risk was 24% (95% CI, 20% to 27%) in the group judged to have experienced a true TIA and 1.4% (0% to 7.6%) in the group judged to not have a true TIA (P<0.0001). ABCD² scores were higher in those judged to have a true TIA compared with others (P=0.0001). In the group judged to have a true TIA, 90-day stroke risk increased as ABCD² score increased (P<0.0001); there was no relationship between ABCD² score and stroke risk in those judged unlikely to have had a TIA (P=0.73).

Conclusions—Among patients diagnosed by emergency department physicians with TIA, higher ABCD² score was associated with a greater likelihood that the diagnosis was confirmed on expert review. The predictive power of the ABCD² model is therefore partially explained by identification of those patients likely to have experienced a true TIA, an important aspect of the score when used by nonneurologists. However, higher ABCD² scores still remained predictive of 90-day stroke rate in the group of patients judged to have a true TIA by an expert neurologist.