LAUSR

Differentiating between a clinical diagnosis of Mild Cognitive Impairment (MCI) and dementia is difficult due to expansive data needs in concert with ambiguity of clinical criteria. Novel artificial intelligence (AI) and machine learning algorithms provide potential avenues for efficiently analyzing data sets and informing clinical judgment in distinguishing MCI from dementia. To date no formal meta-analysis of extant studies has been conducted to compare the efficacy of such procedures. A meta-analysis was conducted to synthesize the sensitivity and specificity of AI and machine learning programs in distinguishing between MCI and dementia as compared to traditional diagnostic protocols. A search of studies using EBSCOhost databases using the keywords: “artificial intelligence,” “machine learning,” “MCI,” and “dementia” retrieved a total of 127 studies. Excluded were 106 studies due to non-reporting of sensitivity and specificity data. In total, 21 studies were included in the present meta-analysis. Sensitivity and specificity data as well as the number of true-false categorizations were extracted and analyzed using OpenMeta[Analyst]. A bivariate correlation produced a summary point with sensitivity of 82% and specificity of 82%. A follow-up Rutter-Gatsonis multivariate correlation HSROC curve was created to correct for significant correlations (47%), and produced an adjusted mean specificity of 79% and sensitivity of 83%. Results suggest AI and machine-learning algorithms are effective in distinguishing MCI from dementia. AI procedures have potential in aiding clinical judgment given a larger body of empirical research.