The development of "clinically sensible" tools to screen for cognitive impairment in community-dwelling elderly persons. Bridging the gap between research and clinical practice by balancing discriminant ability vs. practicality.

Abstract

Background. Despite its prevalence and clinical relevance, cognitive impairment typically remains undetected in 50% of cases. Objective. To develop clinically sensible (quick, simple, acceptable), accurate and readily recalled screens for cognitive impairment based on easily reproducible analytic strategies. Methods. The Canadian Study of Health and Aging (CSHA-1) served as the derivation data set. 3MS cognitive screening questions which were judged as most likely to be employed by busy clinicians and which were significantly associated (via chi2 analysis) with cognitive impairment were selected as independent variables for multivariate analysis. The screening tests derived from logistic regression and recursive partitioning analyses which most closely approximated the sensitivity and specificity of the entire 3MS were externally validated. Results. Two logistic regression based scales and two recursive partitioning algorithms demonstrated sensitivities and specificities approaching those of the complete 3MS (approximately 80% and 60% respectively). The sensitivity was superior to that of the MMSE. Conclusion. Readily reproducible multivariate analysis based strategies can be developed which generate practical screening tests with psychometric properties approaching those of the 3MS. Given the existence of verification bias, these screens as well as screens with higher sensitivity and lower specificity must be validated prospectively before they can be clinically employed.