Background and purpose: In patients with a recent TIA or minor stroke, prediction of long-term risk of major vascular events is important, but difficult. We aimed to study the external validity of currently available prediction models.
Methods: We validated predictions from 3 population-based models (Framingham, SCORE, and INDIANA project) and 4 stroke cohort-based models (Stroke Prognosis Instrument II, Oxford TIA, Dutch TIA study, and the ABCD(2) study) in an independent cohort of patients with a recent TIA or minor stroke. The validation cohort consisted of 592 patients with TIA or minor stroke, with a mean follow-up of 2 years. The primary outcome was the 2-year risk of the composite outcome event of nonfatal stroke, myocardial infarction, or vascular death. We used calibration graphs and c-statistics to evaluate the 7 models.
Results: The 2-year risk of the primary outcome event was 12%. Calibration was adequate for stroke population-based studies. After adjustment for baseline risk and for prevalence of risk factors, calibration was adequate for the Dutch TIA, the ABCD(2), and Stroke Prognosis Instrument II models. Discrimination ranged from 0.61 to 0.68.
Conclusions: Discrimination was poor for all currently available risk prediction models for patients with a recent TIA or minor stroke, indicating the need for stronger predictors. Clinical usefulness may be best for the ABCD(2) model, which had a limited number of easily obtainable variables, a reasonable c-statistic (0.64), and good calibration.