An item response theory approach to clinical diagnosis of MCI and dementia: Illustrations from ADNI and NACC data

Abstract

AbstractBackgroundDiagnostic labels, such as cognitively normal, mild cognitive impairment, and dementia rely on application of several different criteria and are often based at least in part on subjective clinical judgments. These factors may lead to heterogeneity in diagnostic practices across datasets. A modern psychometric technique known as item response theory (IRT) can be used to take information from many different markers of an underlying spectrum (e.g., cognitive impairment) and evaluate how diagnostic labels perform across datasets when shared items are used.MethodIn this poster, we illustrate how IRT might be used to evaluate how diagnostic labels function across two major databases of older adults that include cognitive data and diagnostic labels, the National Alzheimer’s Coordinating Center Uniform Data Set (n = 6,042) and the Alzheimer’s Disease Neuroimaging Initiative (n = 513).ResultTaking an IRT driven Cochran‐Mantel‐Haenzel approach revealed no significant differential item functioning for the diagnostic labels among individuals from the Alzheimer’s Disease Neuroimaging Initiative and National Alzheimer’s Coordinating Center databases (Figure 2; X2 =11.79, p < .001, Class AA item characterization).ConclusionThe current study demonstrated how IRT scoring approaches outline similarities in cognitive functioning across a group of measures, despite being diagnosed under different criteria across studies. Findings illustrate a manner of evaluating if diagnostic labels are applied correctly across participants. The utility of this approach will be tempered by the ability to identify underlying unidimensional constructs across studies, although multi‐dimensional IRT approaches have been developed.