A panel of blood lipids associated with cognitive performance, brain atrophy, and Alzheimer's diagnosis: A longitudinal study of elders without dementia

Abstract

Introduction: We sought lipid-metabolic biomarkers involved in the processes underlying cognitive decline and detected them in association with Alzheimer's disease (AD) phenotypes.

Methods: A least absolute shrinkage and selection operator logistic regression model was used to select lipids that best classified cognitive decline defined by a fast-annual rate of cognition. Lipid summary scores were constructed as predictors of cognitive decline by using this model. Multivariable-adjusted models tested the associations of risk score with AD phenotypes.

Results: A model incorporating 17 selected lipids showed good discrimination and calibration. The lipid risk score was positively associated with the baseline Alzheimer Disease Assessment Scale-13-item cognitive subscale (ADAS-Cog13) score and cerebrospinal tau protein level, and predicted cognitive diagnoses. Additional results showing that individuals with increased lipid risk scores had rapid change rates of ADAS-Cog13 and brain atrophy further corroborated the predictive role of lipids.

Discussion: A panel of blood lipids instead of individual lipid molecules could better diagnose and predict cognitive decline.

Keywords: biomarkers; cognitive decline; lipidomics; prediction.

FIGURE 1

Potential lipid predictor selection using least absolute shrinkage and selection operator (LASSO) logistic regression and the predictive accuracy of lipid signatures. A, Selection of the parameters in the LASSO model by 10‐fold cross‐validation based on minimum criteria. The vertical dotted line points to the optimal lambda value and the number of optimal predictors. B, The pathway of coefficients among all lipid signatures. The 17 lipid signatures were selected via LASSO regression. C and D, Receiver operating characteristic curves of the lipid signatures in the training and validation sets, respectively Abbreviation: AUC, area under the curve.

FIGURE 2

Seventeen lipid predictors of cognitive decline weighted by least absolute shrinkage and selection operator (LASSO) logistic regression. Color‐coded with the same classification annotated in the figure Abbreviations: CE, cholesteryl ester; LPE, lysophosphatidylethanolamine; PC, phosphatidylcholine; Cer, ceramide; dhCer, dihydroceramide; GM3, ganglioside.

FIGURE 3

Receiver operating characteristic curve analysis for prediction of cognitive decline Abbreviation: AUC, area under the curve; PHS, polygenic hazard score.

FIGURE 4

The boxplot of the lipid risk score in different diagnostic groups at baseline Abbreviations: sMCI, stable mild cognitive impairment; pMCI, progressive mild cognitive impairment; CN, cognitively normal; A+T+, amyloid abnormal using CSF Aβ and tau abnormal using CSF p‐tau; A–T–, amyloid normal using CSF Aβ and tau normal using CSF p‐tau.

FIGURE 5

The risk effect estimates among four diagnostic groups by multivariate logistic regression. Models are adjusted for age at baseline, sex, years of education, status of APOE‐ε4, and lipid medications. ^* The case is defined as participant with worse cognitive decline according to the rate of change in Mini‐Mental State Examination Abbreviations: sMCI, stable mild cognitive impairment; pMCI, progressive mild cognitive impairment; CN, cognitively normal; A+T+, amyloid abnormal using CSF Aβ and tau abnormal using CSF p‐tau; A–T–, amyloid normal using CSF Aβ and tau normal using CSF p‐tau.