Background: Vascular cognitive impairment (VCI) is considered to be the most common pattern of cognitive impairment. We aimed to devise a diagnostic algorithm for VCI, and evaluate the reliability and validity of our proposed criteria.
Methods: We based our new algorithm on previous literature, a Delphi consensus method, and preliminary testing. First, successive 100 patients with cerebrovascular disease (CVD) in hospital underwent a structured medical examination. Twenty-five case vignettes fulfilled the proposed criteria of diagnosis for probable or possible VCI were divided into three subtype categories: vascular cognitive impairment, no dementia (VCIND), vascular dementia (VaD) or mixed VCI/Alzheimer's disease (AD). Inter-raters reliability was assessed using a Fleiss kappa analysis. Convergent validity was also evaluated by correlation coefficients (r) between the proposed key points for each subtype and the currently accepted criteria. Forty-five patients with probable VCI were examined to determine the accuracy of identification for each subtype.
Results: The proposed criteria showed clinical diagnostic validity for VCI, and were able to define probable, possible and definite VCI, three VCI subtypes, and vascular causes. There was good consensus between experts (Cronbach's alpha = 0.96 for both rounds). Significant moderate to good items-total correlations were found for two questionnaires (50-r range, 0.40 - 0.97 and 0.41 - 0.99, respectively). Significant slight and moderate inter-raters reliability were obtained for VCI (k = 0.13) and three VCI subtypes (k = 0.45). Furthermore, good convergent validity was observed in a comparison of significant correlations between criteria: good (4-r range, 0.75 - 0.92) to perfect (3-r = 1.00) validity for the VCIND subtype, and moderate to good validity for the VaD subtype (1-r = 0.46; 5-r range, 0.76 - 0.92) and for the mixed VCI/AD subtype (r = 0.92 and 1.00; 4-r range, 0.47 - 0.70). Importantly, the area under receiver operating characteristic (ROC) curves for the subtypes of VCIND, VaD and mixed VCI/AD were 0.85, 0.67 and 0.93, respectively.
Conclusion: Our results suggest that the new VCI diagnostic algorithm might be a suitable clinical approach for assessing stroke patients.