AUTHOR=Gurevich Pavel , Stuke Hannes , Kastrup Andreas , Stuke Heiner , Hildebrandt Helmut 

TITLE=Neuropsychological Testing and Machine Learning Distinguish Alzheimer’s Disease from Other Causes for Cognitive Impairment

JOURNAL=Frontiers in Aging Neuroscience

VOLUME=Volume 9 - 2017

YEAR=2017

URL=https://www.frontiersin.org/journals/aging-neuroscience/articles/10.3389/fnagi.2017.00114

DOI=10.3389/fnagi.2017.00114

ISSN=1663-4365

ABSTRACT=With promising results in recent treatment trials for Alzheimer’s disease (AD), it becomes increasingly important to distinguish AD at early stages from other causes for cognitive impairment. However, existing diagnostic methods are either invasive (lumbar punctures, PET) or inaccurate (MRI). This study investigates the potential of neuropsychological testing to specifically identify those patients with possible AD among a sample of 158 patients with MCI or dementia for various causes. Patients were divided into an early stage and a late stage group according to their mini mental state examination score and labeled as AD or non-AD patients based on a post-mortem validated threshold of the ratio between total tau and beta amyloid in the cerebrospinal fluid (TB ratio). All patients completed the established CERAD-NAB test battery and two additional newly-developed neuropsychological tests (recollection and verbal comprehension) that aimed at carving out specific Alzheimer-typical deficits. Based on these test results, an underlying AD (pathologically increased TB ratio) was predicted with a machine learning algorithm. To this end, the algorithm was trained in each case on all patients except the one to predict (leave-one-out validation). In the total group, 82% of the patients could be correctly identified as AD or non-AD. In the early group with small general cognitive impairment, classification accuracy was increased to 89%. Neuropsychological testing thus seems to be capable of discriminating between AD patients and patients with cognitive impairment due to other neurodegenerative or vascular causes with a high accuracy, and may be used for screening in clinical routine and drug studies, especially in the early course of this disease.