Year: 2019
Type: article
Abstract: Defining neuropsychological test criteria that optimally identify mild cognitive impairment (MCI) across diverse cohorts remains an open question. Our goal was to determine the accuracy of various algorithmic screening procedures for identification of MCI compared to gold standard of consensus diagnosis. A total of 470 individuals [mean age=72; range=55-95; White=321(68%); African-American=144(30%)] enrolled in Wake Forest ADRC affiliated cohorts and underwent clinical assessments between 2016-2018. Assessments included the UDSv3 battery and the Rey Auditory Verbal Learning Test (RAVLT). All participants underwent cognitive adjudication by consensus conference and were classified as having either normal cognition or cognitive impairment (MCI or dementia). We compared algorithms to identify cognitive impairment using three criteria previously described in the literature: “Broadly Inclusive” - any score <1.0 SD below the mean based on normative data; “Original MCI” - any score <1.5 SD below the mean; and “Revised MCI” - scores <1.0 SD for any two memory tests, or 1 memory test and 1 test in another domain. In sensitivity analysis, we performed classification by Random Forests (RF) using all scores. Model performance was evaluated by generating estimates of accuracy, sensitivity and specificity. Of 470 participants, 321(67%) were independently adjudicated by consensus conference as cognitively normal, 121(26%) had MCI, and 34(7%) had dementia. The ‘Broadly Inclusive’ algorithm identified 98% of impaired participants, with two false-negative MCI cases. The ‘Original MCI’ algorithm identified 97% of impaired participants, with four false-negative MCI cases. ‘Revised MCI’ identified 80% of impaired participants, with 21% false-negatives. When RAVLT was added to the UDSv3 battery, each algorithm increased in sensitivity and decreased in specificity and misclassified fewer MCI participants (Table). Each method performed similarly well in cohort subgroups defined by race/gender (p-interaction>0.80). RF generated accuracy = 86.4%, sensitivity = 88.5% and specificity = 85.4%.
Source: Alzheimer s & Dementia
Authors: Bonnie C. Sachs, Timothy M. Hughes, Ramon Casanova, Mark A. Espeland, Eric Fischer, Kathleen M. Hayden, Stephen R. Rapp, Alexis N. Webb, Suzanne Craft
Language: English
Subfield: Psychiatry and Mental health
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W2981266149