P4‐254: Predicting brain amyloidosis in MCI using clinical, cognitive, imaging and peripheral blood protein measures
Year: 2012
Type: article
Abstract: Randomized, double-blind clinical trials for the latent stages of Alzheimer's disease are now being planned. To identify enough amyloid positive cognitively normal subjects, these trials would need to screen tens of thousands of asymptomatic individuals using amyloid positron emission tomography (PET) imaging. Identifying people highly likely to be amyloid-positive without relying on amyloid PET could save time and resources. We investigated the accuracy of a novel unsupervised multimodal biomarker classifier for predicting brain amyloidosis in MCI subjects. By combining the predictive information of cognitive, imaging and peripheral blood protein biomarkers, we expected to achieve high accuracy in differentiating Pittsburgh compound B (PIB) positive (SUVR>1.6) from PIB negative (SUVR<1.6) subjects. Using the baseline cognitive, imaging and proteomic data of 60 ADNI MCI participants (mean age=75.2 ±8.1, mean MMSE=27.2 ±2.2) we developed automated unimodal and multimodal classifiers based on support vector machines. The pool of features provided to the classifiers included MMSE, Trails B, AVLT delayed recall, plasma levels of ApoE, BDNF, TNF alpha, IL13, IL6 and clusterin, and hippocampal volume. All classifiers included age and sex. Three separate classifiers were built to predict mean cortical, lateral parietal and precuneal SUVR as the outcome variable. All classifiers consistently ranked ApoE genotype as the most powerful predictor of brain amyloidosis. ApoE genotype alone predicted brain amyloidosis with 75–77% accuracy (AUC=0.67–0.75). When cognitive, imaging and peripheral blood protein variables were added, classifier performance improved. Greatest classifier accuracy in predicting PIB+ vs. PIB- subjects was seen for the lateral parietal (accuracy=85%, AUC=0.87) and precuneal regions (accuracy=85%, AUC=0.89). The variables selected by these classifiers included ApoE genotype, MMSE, Trails B, AVLT delayed recall, clusterin, TNF alpha and BDNF. The precuneal classifier also included IL13 and ApoE protein. The average cortical classifier used ApoE genotype, MMSE, Trails B, hippocampal volume, clusterin, BDNF and IL6 and achieved 80% accuracy (AUC=0.81).
Source: Alzheimer s & Dementia
Language: English
Subfield: Psychiatry and Mental health
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W2156937492