LONGITUDINAL ANALYSIS OF COGNITIVE DECLINE PATTERNS IN PATIENTS WITH MILD COGNITIVE IMPAIRMENT

Abstract

Mild Cognitive Impairment (MCI) is a transition condition which is heterogeneous with variable rates of transition to Alzheimer Disease (AD). To improve clinical trial and/or intervene on case to case basis, people at risk of the rapid conversion should be properly diagnosed. The result of the analysis determines and validates a multi-modal machine learning system of of plasma biomarkers, neuroimaging, genetic and cognitive data that predicts 36-month MCI-to-AD progression. It was analyzed in terms of data of 412 MCI patients of the cohort of the Alzheimer Disease Neuroimaging Initiative (ADNI). They used penalized logistic regression model, which they erected by using baseline predictors, such as plasma p-tau217 concentration, hippocampal volume, entorhinal cortex thickness, FDG-PET ratio of standard uptake values, APOE 4, genotype and cognitive composite score. The AUC-ROC, calibration metrics and the decision curve analysis were used to evaluate the performance of the model. The longitudinal latent class mixed modeling (LCMM) was used to determine the different cognitive paths. The cost-effectiveness analysis was used to simulate the effects of p-tau217 prescreening on the size of research sample and cost of testing the biomarkers.Multi-modal model did a relatively good job in discriminative (AUC-ROC = 0.951, sensitivity = 0.903, specificity = 0.924) and the performance of the multi-modal model was also much high than the model of single-modality ( 0.0570.110, p < 0.001). Plasma p-tau217 proved to be the most predictive (odds ratio = 17.18, p < 0.001), with a cut-off point of >2.5 pg/mL (enrichment ratio = 1.95), and screen-failure rate (44.7%). This approach of prescreening also reduced the clinical trial samples needed by 58 percent and biomarker testing cost also reduced by 61 percent. Four out of the four tracks of thinking discovered by LCMM were a fast progressor (10.5% of cohort) with a conversion rate of 94.2. APOE ε4 subgroups and cognitive severity groups had a high integrity of the model with a AUC-ROC of 0.921-0.984. To further stratify patients at highest risk of progressing to AD, multi-modal and a combination of plasma p-tau217, neuroimaging and genetic biomarkers are needed. When coupled with this approach, many efficiencies are gained during clinical trial design such as a reduction in sample size by a large factor, a reduction in screening expenses and consequently more disease-modifying therapies are created.