<pre><FONT COLOR = BLACK>
<B>Using machine learning for predictive analytics for Alzheimer?s disease
 </B>
   Lead Investigator:    Ali Ezzati
   Institution      :    Albert Einstein College of Medicine
   E-Mail           :    ali.ezzati@einstein.yu.edu
   Proposal ID      :    1169
</FONT>
<BR>
   Proposal Description:
   <textarea name="description" rows="12" cols="85" wrap="virtual" readonly>Therapeutic failure in AD trials may arise, in part, from the biological heterogeneity of AD and from the frequent occurrence of mixed dementia pathologies. Many investigators have tried to find more homogeneous populations for their studies by applying strict inclusion and exclusion criteria. Using artificial cut offs (e.g. hippocampal volume or neuropsychiatric measures), specific diagnostic criteria (normal vs MCI vs AD), or having specific sets of genetic mutations are fairly common methods proposed in design of cohorts and clinical trials. However, such approaches are limited due to their susceptibility to diagnostic misclassification, failure to account for the pattern of association between variables, and substantial heterogeneity within groups. With the rapid growth of information on individuals? health and the increased availability of biomarkers, investigators have tried to use advanced statistical methods such as Bayesian modelling, latent class analysis (LCA) and machine learning (ML) techniques to improve quantitative risk prediction of cognitive decline and AD. The overall goal of this project is to use ML methods to identify specific subgroups in longitudinal cohorts of NACC, which differ in their rates of cognitive decline based on multimodal data.    Aim 1 tests the hypothesis that in comparison with traditional neuropsychiatric criteria and AD biomarkers, subgroups identified by ML methods based on baseline data (demographics, clinical and neurocognitive), will have higher accuracy for predicting longitudinal cognitive trajectories and cognitive events.   Aim 2 is to use subset of participant with MRI, and apply ML methods to both MRI and neuropsychological measures and evaluate the additive (or synergistic) value of each set of measures (indicators) in prediction of rate of cognitive decline as well as time to conversion to AD in preclinical AD stage.  Aim 3 is to use the subset of participants with neuropathology data and evaluate the longitudina </textarea>