A recently developed blood-based signature can help predict the status of an early Alzheimer’s disease risk indicator with high accuracy, investigators are reporting.
Courtesy NIH
By analyzing as few as four proteins, the machine learning-derived test can predict the status of cerebrospinal fluid (CSF) amyloid beta1-42 (Abeta1-42), according to Noel G. Faux, PHD, of IBM Australia and the University of Melbourne, and co-investigators.
While shifts in Abeta1-42 may signal the presence of disease long before significant cognitive decline is clinically apparent, collection of CSF is highly invasive and expensive, Faux and investigators said in their report.
By contrast, blood biomarkers could prove to be a useful alternative not only to invasive lumbar punctures, they said, but also to the positron emission tomography (PET) evaluation of Abeta1-42, which is expensive and limited in some regions.
“In conjunction with biomarkers for neocortical amyloid burden, the CSF Abeta1-42biomarkers presented in this work may help yield a cheap, non-invasive tool for both improving clinical trials targeting amyloid and population screening,” Dr. Faux and co-authors said in Scientific Reports.
Dr. Faux and colleagues used a Random Forest approach to build models for CSF Abeta1-42 using blood biomarkers and other variables.
They found that a model incorporating age, APOEe4 carrier status, and a number of plasma protein levels predicted Abeta1-42 normal/abnormalstatus with an AUC, sensitivity and specificity of 0.84, 0.78 and 0.73 respectively.
In a model they said was more suitable for clinical application, they narrowed down the variables to 4 plasma analytes and APOEe4 carrier status, which had an AUC, sensitivity, and specificity of 0.81, 0.81 and 0.64 respectively.
They validated the models on a cohort of individuals in the Alzheimer’s Disease Neuroimaging Initiative (ADNI), a large, longitudinal, multicenter study.
Patients with mild cognitive impairment with predicted abnormal CSF Abeta1-42 levels indeed did transition to a diagnosis of Alzheimer’s disease more quickly than those with predicted normal levels, according to investigators.
That helps provide “strong evidence” that the blood-based model is generalizable, robust, and could help stratify patients based on risk of progressing to Alzheimer’s disease, they said in their report.
Dr. Faux and colleagues declared no conflicts of interest related to the research.
SOURCE: Goudey B, et al. Sci Rep. 2019 Mar 10. doi: 10.1101/190207v3.