Scientific Report 2007

Molecular and Experimental Medicine

Division of Biomathematics

Outcome Scales in Stroke

J.A. Koziol, G.J. del Zoppo

The development of stroke scales, such as the National Institutes of Health Stroke Scale (NIHSS), the Scandinavian Stroke Scale, or the Canadian Neurological Score, was motivated by the desire to describe the outcomes of stroke in terms of what clinicians understand most readily, the physical examination. Still, the issue arises as to how to handle the composite score represented by the NIHSS and similar scoring instruments. In particular, could some component of the NIHSS examination, or a neurologic examination in general, be a more sensitive indicator of outcome?

Operationally, the NIHSS score represents a reduction of all the data elements to a single score. Use of a summary score has benefits, including clinical validity (so long as each of the individual components that make up the score is a clinically important disease manifestation that has face validity), avoidance of multiplicity, and improved sensitivity because of expected reduction in measurement noise. Nevertheless, the fundamental question is whether or not the reduction of information from a large number of elements to an overall single quantity or score provides an adequate representation of the information available in the original clinical assessments. In particular, it is not at all obvious how to determine the optimal weights or, more generally, what constitutes the best method of combining the individual measures into a single index. One way to determine the appropriate set of weights is to use principal components analysis. An important initial step in this process is to examine the individual components or elements to ensure their reproducibility and validity; subsequent analyses should accommodate the ordinal and not numerical nature of the clinical scale.

The basic principles, methods, and terminology used in the evaluation of scales for clinical research are well established. And the use of composite scores for neurologic assessment is pervasive. For example, in one rating method for multiple sclerosis, the results of neurologic examination were converted into a weighted ordinal impairment scale. The various items that made up the impairment scale were not expected to be homogeneous, so it was not at all surprising that various components of the scale appeared more responsive than did the overall summary measure, and others, less responsive, to a treatment. Analogously, in clinical trials of treatment of stroke, certain components of the neurologic examination might be more sensitive to arterial recanalization than the summary NIHSS score and other components less sensitive. This possibility appears to be the case.

In a prospective study on dosages of recombinant tissue plasminogen activator (rt-PA), 93 of 104 patients with symptomatic documented cerebral arterial occlusions, infusion of intravenous rt-PA was completed within 5.4 ± 1.7 hours after the onset of signs or symptoms. Each patient was scored by the same neurologist at baseline and at 24, 48, and 72 hours and subsequently according to a neurologic examination based on the Harvard Stroke Registry. After rt-PA infusion, 4 patients had complete recanalization, 31 had partial recanalization, and 58 had no recanalization as indicated by angiography at the end of the infusion.

In this trial, we found no significant difference between the 35 patients who had complete or partial recanalization and the 58 patients who had no recanalization for 3 of the categories of the neurologic examination: general characteristics, behavioral examination, and motor examination. The 2 groups differed significantly in the sensory examination. In order to investigate this difference further, the proportions of patients with abnormal findings for the hand and face sensation components of the sensory examination at each time point were observed (Fig. 1). Compared with patients who did not have recanalization, those who had complete or partial recanalization improved dramatically during the first 24 hours after rt-PA infusion. Thereafter, the sensory responses seemed stable within each group.

Fig. 1. Time changes in the hand and face sensation components of the individual sensorimotor evaluations for patients who had partial or complete recanalization (n = 35) vs no recanalization (n = 58) after treatment with rt-PA. Proportions of normal vs abnormal findings are depicted, where abnormal status comprises subjective decrease, partial loss of function, or severe loss of function.

In this particular trial then, sensation seemed to be an early (24 hour) indicator of recanalization status, more so than the other components of the neurologic examination. Such fine distinctions may be obscured in a summary measure, particularly if little weight is given to the sensation domain in the summary.

In general, clinical changes associated with recanalization might be expected to be greatest in the first 24 hours after onset of stroke (especially if rt-PA is given within 3 hours of onset). Hence, a plausible scenario is that the recanalization group improves and the no-recanalization group does not in the first 24 hours, and the differences are then maintained over the next 2 days. Although the data tend to support this notion, we caution that only few patients achieve complete recanalization with intravenous infusion of rt-PA, and combining patients with complete and partial recanalization outcomes may dilute real effects of improved blood flow. More to the point, one might question the underlying premise that recanalization of a major artery should be associated with improved brain function. In the end, it would be useful to know which elements of stroke scales are most sensitive to recanalization.

Publications

Koziol, J.A., Feng, A.C. On the analysis and interpretation of outcome measures in stroke clinical trials: lessons from the SAINT I study of NXY-059 for acute ischemic stroke. Stroke 37:2644, 2006.