Statistical methods traditionally used in the analysis of change (e.g., repeated measures ANOVA) may be inadequate for the investigation of cognitive decline if a study's effect size is small, the variance within groups is heterogeneous, or the statistical power is low. To examine an alternative approach to the determination of clinically meaningful cognitive decline and investigate whether such decline occurs during the first year after stroke, we administered a neuropsychological test battery to 172 patients (age = 70.3 ± 7.6years;education= 10.3 ± 4.7years) 3 and 12 months after stroke and 199 nondemented stroke-free control subjects (age= 71.1 ± 6.4years;education= 12.8 ± 4.2years) on two occasions 12 months apart. Two neuropsychologists classified each subject's test performance as having declined, improved, or remained stable based solely on clinical judgment. Reliability of the rating of decline versus the pooled rating of improvement/stability was excellent (kappa= 0.79). The two rating groups differed significantly and in the appropriate directions in change on most tests. While a MANOVA comparing the stroke and control groups on change in test scores was not significant, logistic regression analysis determined that a rating of clinically meaningful cognitive decline was associated with stroke status (Odds Ratio= 1.8, 95%Confidence Interval= 1.0to3.2), while adjusting for demographic factors. We propose that this alternative approach to the analysis of cognitive change can facilitate the recognition of decline in subgroups of subjects. It would be valuable as an adjunct to studies of the incidence of dementia, for example, in which the recognition of cognitive decline might be difficult in highly educated patients whose baseline level of performance is far above the cutoffs operationalized for the diagnosis of dementia.

You do not currently have access to this article.