[Study on the risk factors of development for mild cognitive impairment to Alzheimer's disease based on the competitive risk joint model].

Objective

To identify the risk factors for the progression from mild cognitive impairment (MCI) to Alzheimer's disease (AD) using a competing risks joint model.

Methods

This study was based on the Alzheimer's Disease Neuroimaging Initiative database. Three gradient boosting tree algorithm, namely CatBoost, XGBoost, and LightGBM, were employed to reduce the dimensionality of the high-dimensional lipoprotein and metabolite data (including 250 amino acids, lipids, and energy metabolism-related components, etc.). A random survival forest model (RSF) was used to screen out key demographic, cognitive function scores and metabolic variables. A competing risk joint model was constructed to identify the risk factors for the progression from MCI to AD.

Results

A total of 8 lipoprotein and metabolite variables were identified by the three algorithms [creatinine, lactate, glycine, large high-density lipoprotein phospholipids percent (L_HDL_PL_PCT), acetic acid, tyrosine, β-hydroxybutyric acid and valine]. The RSF model identified 14 main variables, including cognitive function indicators [Functional Activities Questionnaire (FAQ), Alzheimer's Disease Assessment Scale-13 items (ADAS13), Alzheimer's Disease Assessment Scale word recognition item 4 (ADASQ4), and Alzheimer's Disease Assessment Scale-11 items (ADAS11)], lipoprotein and metabolites (acetic acid, L_HDL_PL_PCT, β-hydroxybutyrate, glycine, and creatinine), and baseline characteristics [age, years of education, marital status, retirement status, and apolipoprotein E ε4 allele (APOE-ε4)]. The univariate competing risk joint model showed that the longitudinal changes of ADAS11, ADAS13, ADASQ4, FAQ, and glycine, as well as baseline age, marital status, and APOE-ε4, were positively associated with the progression of MCI to AD (P<0.05). The results of the multivariate competing risk joint model further indicated that the longitudinal changes of ADAS13, ADASQ4, FAQ, and glycine, as well as baseline age, marital status, and APOE-ε4 were positively associated with AD incidence (P<0.05).

Conclusions

Longitudinal increases in ADAS13, ADASQ4, FAQ, and glycine, as well as older age, unfavorable marital status, and APOE-ε4 carriage at baseline, were all risk factors for progression from MCI to AD.