It has been suggested that subtle signs of early cerebral infarction on CT
are important indicators of outcome and of the effect of thrombolytic treat
ment in acute ischaemic stroke. We studied these signs prospectively, in 26
0 patients with an anterior circulation stroke from a European-Australian r
andomised trial of lubeluzole in acute ischaemic stroke. Interobserver reli
ability was assessed by means of the x, statistic. The validity of the earl
y signs was assessed by comparing the assessments of the first CT with anot
her CT at 1 week after the onset of stroke, and with stroke outcome at 12 w
eeks. Each initial CT study was assessed by two of a group of five reviewer
s, who were blinded to each other's assessments and to the findings on the
follow-up CT. The images were assessed twice, once without clinical informa
tion and again after disclosure of the side (left or right hemisphere) of t
he lesion. All reviewers were experienced clinicians with a special interes
t and training in vascular neurology and CT. The median time between stroke
onset and the first CT was 3.2 h; 59 % of the patients were imaged within
3 h and 77 % within 6 h. More than half of the patients (52 %) had a large
middle cerebral artery territory (MCA) infarct on follow-up CT. Chance-adju
sted interobserver agreement (x) for any early infarct was 0.27 (95 % confi
dence interval (CI): 0.15 to 0.39), Agreement ( x) on the extent of a middl
e cerebral artery (MCA) infarct and on the indication for treatment with re
combinant tissue plasminogen activator (rt-PA) was fair: 0.37 and 0.35, res
pectively. Patients with early signs of an infarct of more than 1/3 of the
MCA territory were more likely to have a large MCA infarct on follow-up CT
(odds ratio 5.7, 95 % confidence interval 2.8-11.5); the positive and negat
ive predictive value of these signs was 81 % and 57 %, respectively. Chance
-adjusted interobserver agreement on early, subtle signs of a large MCA ter
ritory infarct on CT by neurologists was thus no more than fair, and the ac
curacy of prediction of actual infarct size on the basis of these signs onl
y moderate, under circumstances which resemble everyday clinical practice.