Clinical prediction algorithm (BRAIN) to determine risk of hematoma growth in acute intracerebral hemorrhage

Abstract

Background and Purpose—We developed and validated a simple algorithm to predict the risk of hematoma growth in acute spontaneous intracerebral hemorrhage (ICH) to better inform clinicians and researchers in their efforts to improve outcomes for patients.

Methods—We analyzed data from the computed tomography substudies of the pilot and main phases of the Intensive Blood Pressure Reduction in Acute Cerebral Hemorrhage Trials (INTERACT1 and 2, respectively). The study group was divided into a derivation cohort (INTERACT2, n=964) and a validation cohort (INTERACT1, n=346). Multivariable logistic regression was used to identify factors associated with clinically significant (≥6 mL) increase in hematoma volume at 24 hours after symptom onset. A parsimonious risk score was developed on the basis of regression coefficients derived from the logistic model.

Results—A 24-point BRAIN score was derived from INTERACT2 (C-statistic, 0.73) based on baseline ICH volume (mL per score, ≤10=0, 10–20=5, >20=7), recurrent ICH (yes=4), anticoagulation with warfarin at symptom onset (yes=6), intraventricular extension (yes=2), and number of hours to baseline computed tomography from symptom onset (≤1=5, 1–2=4, 2–3=3, 3–4=2, 4–5=1, >5=0) predicted the probability of ICH growth (ranging from 3.4% for 0 point to 85.8% for 24 points) with good discrimination (C-statistic, 0.73) and calibration (Hosmer–Lemeshow P=0.82) in INTERACT1.

Conclusions—The simple BRAIN score predicts the probability of hematoma growth in ICH. This could be used to improve risk stratification for research and clinical practice.