Given that functional outcome is one of the most commonly used parameters in studying acute stroke treatment, developing accurate prognostication scores would greatly facilitate treatment decisions and improve communicating expectations to patients and families. Cooray et al. sought to validate the two most recently developed scores designed to predict functional outcome at three months, one studied in unselected acute stroke patients (ASTRAL) and the other in acute stroke patients treated with iv-tPA (DRAGON) using the SITS-International Stroke Thrombolysis Register (ISTR), a global stroke thrombolysis database. Outcomes were dichotomized into modified Rankin Scale (mRS) 0-2 and 3-6 as were done in both of the initial studies, and the area under the curve (AUC) of the receiver operating characteristic (ROC) was used in both scores to assess the overall predictive and discriminative performance.
The DRAGON score was developed in a single center cohort of acute ischemic stroke patients treated with iv-tPA using similar statistical design to the ASTRAL score. It is a 10 point scale and the included parameters are hyperdense cerebral artery sign (1 point) and early infarct signs (1 point) on baseline CT, pre-stroke mRS score >1 (1 point), age (<65 years = 0 points, 65-79 years = 1 point, >80 years ≥ 2 points), acute blood glucose >8 mmol/L (1 point), time from symptom onset to treatment >90 min (1 point) and NIHSS score (0-4 = 0 points, 5-9 = 1 point, 10-15 = 2 points and >15 = 3 points). A total of 33,716 iv-tPA treated patients with complete data for the DRAGON score were registered in the SITS-ISTR database. The main differences between the SITS and DRAGON cohorts were higher median baseline stroke severity (NIHSS 12 vs 9), lower proportion of early infarct signs (16.5% vs 30.6%) and higher onset-to-treatment time in the SITS cohort. The AUC-ROC value for functionally dependent outcome on the DRAGON score using the SITS-ISTR cohort was 0.77 (95% CI 0.769-0.779). The largest discrepancy between observed and predicted outcome was close to 17%.

Prediction models won't get any better until objective points are used. 3d representation of the dead and penumbra or bleed areas and location of the epicenter. That way you would at least have a decent starting point so clinical trials can be compared.