Lb. Goldstein et V. Chilukuri, RETROSPECTIVE ASSESSMENT OF INITIAL STROKE SEVERITY WITH THE CANADIANNEUROLOGICAL SCALE, Stroke, 28(6), 1997, pp. 1181-1184
Background and Purpose The severity of the initial neurological defici
t is a critical determinant of outcome after acute stroke. Retrospecti
ve outcome studies are generally limited by a lack of quantitative dat
a relating to this initial stroke severity. We evaluated the validity
and reliability of measuring initial stroke severity retrospectively w
ith the Canadian Neurological Scale (CNS). Methods The CNS was used to
prospectively score the initial neurological deficit in a series of p
atients with acute ischemic stroke (n=24). An algorithm was devised fo
r applying the CNS retrospectively on the basis of information in the
patient's hospital discharge summary. Those dictating the discharge su
mmaries were not aware of the study, and the retrospective scoring was
performed without reference to other scores. The level of agree ment
between the prospective and retrospective scores (validity) and both i
ntraobserver and interobserver reliability for the retrospective score
s were determined. Results Agreement was high between retrospective an
d prospective scores (r=.84, R-2=.71, P<.0001), between two sets of re
trospective scores obtained by one rater (r=.95, R-2=.91, P<.0001), an
d between retrospective scores obtained by different raters (r=.91, R-
2=.82, P<.0001). Weighted kappa statistics (kappa(W)) for prospectivel
y versus retrospectively scored items varied from almost perfect (kapp
a(W)>0.81 for level of consciousness and orientation) to substantial (
kappa(W)=0.68 for speech) and moderate (kappa(W)=0.41 to 0.60 for faci
al weakness, proximal arm, distal arm, proximal leg, and distal leg st
rength). Using the retrospective algorithm, there was almost perfect i
ntraobserver and interobserver reliability for each of the individual
CNS items (kappa(W)=0.81 to 1.00). Conclusions These data show that re
trospective scoring of initial stroke severity using an algorithm base
d on the CNS is valid and can be reliably performed using information
available in hospital discharge summaries.