Premium
P1‐271: CSF GLUCOSE TRACKS BRAAK STAGE TAU DEPOSITION
Author(s) -
Pappas Colleen,
McLimans Kelsey E.,
Klinedinst Brandon S.,
Plagman Alexandra K.,
Le Scott T.,
Anatharam Vellareddy,
Kanthasamy Anumantha,
Willette Auriel A.
Publication year - 2019
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1016/j.jalz.2019.06.826
Subject(s) - medicine , neuroimaging , precuneus , endocrinology , cerebrospinal fluid , standardized uptake value , positron emission tomography , glucose uptake , posterior cingulate , alzheimer's disease , psychology , dementia , neuroscience , disease , cognition , insulin
from 2005 to 2012 (for 8 years) fromKorean National Health Insurance Service database. EHR contained diagnoses (ICD-9/10), medications, laboratory values, annual health check-up database, and socio-demographics. [Event of interest] We used EHR from elders age above 65 years old: N1⁄4430,133. Our event of interest was the month of incidence of AD. We defined the incidence of AD based on diagnostic codes and prescription of anti-dementia (N1⁄4672). [Control samples] Total 100 sets of random bootstrap samples of controls with an equal number to AD cases were drawn repeatedly. [Modeling] We trained and validated random forest, support vector machine, and logistic regression to predict incidence of AD in 1,2,3, and 4 subsequent years. Data was split into training (60%), validation (20%), and test set (20%). Results: In predicting future incidence of AD, random forest showed the best performance in 1 year prediction with accuracy of 0.7660.03; in 2 year, 0.7360.02; in 3 year, 0.6760.03; in 4 year, 0.6660.04. Average duration of EHR was 1936 days in AD and 2694 days in controls. Features selected in logistic regression included well-known AD risk factors (age, smoking, essential hypertension), as well as relatively new, potential factors (waist, LDL and HDL-cholesterol, hemoglobin, urine glucose, amoxicillin. Conclusions: This is the first study reporting the promising utility of EHR in predicting future incidence of AD. The current prediction of future incidence of AD with maximum 76% accuracy (1 year) is surprising compared with other well-studied biomarkers. We expect that including EHRbased AD prediction will significantly improve individualized prediction of risk for AD. Towards EHR-based AD screening with practical utility, future research should test the generalizability of the model in independent data.