Authors: Fatemah A. Sakr, Martin Dybra, Markus Gräler, Stefan J. Teipel, Anja Braeuer
Published: 2020-12-07
DOI: 10.1002/alz.038950
Source: Full article
AbstractBackground“Precision medicine” is an emerging concept in the field of Alzheimer’s disease research, allowing for individual risk stratification based on the specific biological/molecular signature. Hence, extensive “Omics” studies exploring different biological molecules were run in order to retrieve molecular‐cellular pathways reflecting the full spectrum of the disease. Lipidomics seems to be one promising approach, as it reveals a number of bioactive molecules regulating patho‐physiological mechanisms. In this study, we explored Lipidomics‐derived data to identify relevant predictors of diagnostic sensitivity in the early phases of AD.MethodWe used data provided by ADNI and ran initial exploratory analyses followed by a logistic regression approach with elastic‐net regularization for selecting pertinent lipids discriminating between people from the AD spectrum and amyloid negative healthy controls. We defined three different models; 1‐ null model, only lipids were included as predictors 2‐ Fixed factors adjusted model, where Age, Gender and BMI were included as covariates and 3‐ APOE4 & Fixed factors adjusted model, which were further tested by adding continuous CSF‐pTAU values.ResultIn our analysis investigating CSF‐Aβ‐ NC (n=63) and CSF‐Aβ+ MCI (n=134), the best performance was achieved by the third model with an estimated cross‐validation accuracy, area under the curve (AUC), negative (NPV) and positive predictive value (PPV) of 0.76, 0.83, 0.67 and 0.80 respectively. Analogously, on investigating individuals at preclinical stage (CSF‐Aβ+ cognitively normal, n=39) against CSF‐Aβ‐ NC, the third model was the only model with acceptable performance with estimated CV accuracy, AUC, NPV and PPV of 0.66, 0.70, 0.68 and 0.64, respectively. This performance was further improved by combining the lipids with pTAU yielding 0.72, 0.74, 0.73 and 0.72 for the accuracy, AUC, NPV and PPV, respectively. The comparison between all models’ performance is summarized in Table. 1 for theNC versus preclinical analysis and displayed in terms of AUC values in Fig. 1 for NC versus MCI analysis. Additional results will be presented at the conference.ConclusionLipid profiling can be a useful tool to reflect the pathological changes in early phases of AD. We plan to replicate these findings on a locally recruited cohort from our specialized memory clinic.