Logo image
Back
43 Random Forest Model Approaches to Build Prediction Models of Cognitive Impairment Using the National Alzheimer’s Coordinating Center database
Abstract   Open access   Peer reviewed

43 Random Forest Model Approaches to Build Prediction Models of Cognitive Impairment Using the National Alzheimer’s Coordinating Center database

Chooza Moon, Boxiang Wang, Sue Gardner, Joel Geerling and Karn Hoth
Journal of clinical and translational science, Vol.7(Suppl 1), pp.11-11
04/01/2023
DOI: 10.1017/cts.2023.135
PMCID: PMC10129561
url
https://doi.org/10.1017/cts.2023.135View
Published (Version of record) Open Access

Abstract

OBJECTIVES/GOALS: Our goal is to explore the complex, the non-linear interplay among chronic conditions collectively contributing to a greater detrimental impact on the progression of Alzheimer’s disease (AD) than a single chronic condition alone in individuals with normal cognition, MCI, and AD. METHODS/STUDY POPULATION: We used longitudinal data from National Alzheimer Coordinating Center (n = 41,437) and focused on individuals with normal cognition (n =16,884, mean age (SD) = 70.72 (9.7)). Random forest models were used to predict newly developed MCI or AD from baseline to the most recent visits. We used self-reported baseline data on 50 chronic conditions and comprehensive clinical and demographic information (e.g., age, sex, APOE status, mini-mental status exam (MMSE) scores, education, BMI, and depressive symptoms). A binomial random forest was used to identify significant interactions (with p-values RESULTS/ANTICIPATED RESULTS: Our model demonstrated an AUC of 0.708 and a classification error rate of 25.4%. Variables of importance for predicting MCI or dementia were age, coronary artery bypass, depression, APOE status, smoking, and depressive symptoms. Two-way interactions, such as age X MMSE score, age X depressive symptoms, and age X BMI, were significant. Three-way interactions, including age X depressive symptoms X MMSE score, or depressive symptoms X BMI X MMSE score, were significant. However, when we explored the random forest model using only the chronic condition data, we found an AUC of 0.602 and an error rate of 27.15%. We found that depression, anxiety, hypercholesterolemia, stroke, and the interaction between BMI and anxiety were significant. DISCUSSION/SIGNIFICANCE: Random Forest models indicate that not only known factors including age, baseline cognitive status, and APOE status, but also chronic conditions like depression, anxiety, hypercholesterolemia, and stroke may predict cognitive impairment.
 Biostatistics, Epidemiology, and Research Design

Details

Metrics

17 Record Views
Logo image