Open AccessMACHINE LEARNING MODEL BASED ON A PANEL OF PLASMA PROTEINS PREDICTS PROGRESSION FROM MILD COGNITIVE IMPAIRMENT TO DEMENTIA DUE TO ALZHEIMER’S DISEASE WITHIN FOUR YEARS
Author(s) -
Daniella Castro Araújo,
Paulo Caramelli,
Nívio Ziviani,
Karina Braga Gomes,
Adriano Veloso,
Leonardo de Souza
Publication year - 2021
Publication title -
dementia and neuropsychologia
Language(s) - English
Resource type - Conference proceedings
SCImago Journal Rank - 0.54
H-Index - 21
ISSN - 1980-5764
DOI - 10.5327/1980-5764.rpda050
Subject(s) - dementia , disease , neuroimaging , cognitive impairment , alzheimer's disease , alzheimer's disease neuroimaging initiative , psychological intervention , cognition , medicine , pathological , cognitive decline , psychology , gerontology , psychiatry
Background: Alzheimer’s disease (AD) is a pathological process that begins many years prior to the emergence of symptoms. Thus, an effective risk model for AD should aim at detecting at-risk individuals in the prodromal stage, when treatments and lifestyle interventions are more likely to be successful, and should be minimally invasive and inexpensive to allow widespread applicability. Objectives: To develop a machine learning based blood panel in individuals with mild cognitive impairment (MCI) to predict increased risk for progression to AD dementia. Methods: We created over one billion models to predict the probability of conversion from MCI to dementia due to AD, and chose the best performing one. We used the Alzheimer’s Disease Neuroimaging Initiative (ADNI) data of 380 MCI individuals in the baseline visit, from which 177 converted to AD dementia. Results: The best performing model uses 12 plasma proteins (out of 146 possibilities), giving an average AUC of 0.91, accuracy of 0.92, sensitivity of 0.97 and specificity of 0.85. Conclusions: We were able to predict AD dementia conversion within four years in MCI individuals, a disease stage in which treatments and lifestyle interventions are more likely to be successful. Further studies in independent cohorts are needed to validate this panel.