A novel electrocardiogram-based model for prediction of dementia-The Atherosclerosis Risk in Communities (ARIC) study

Deling Chen; Yuchen Yao; Ethan D Moser; Wendy Wang; Elsayed Z Soliman; Thomas Mosley; Wei Pan

doi:10.1016/j.jelectrocard.2024.153832

A novel electrocardiogram-based model for prediction of dementia-The Atherosclerosis Risk in Communities (ARIC) study

J Electrocardiol. 2024 Nov 21:88:153832. doi: 10.1016/j.jelectrocard.2024.153832. Online ahead of print.

Authors

Deling Chen¹, Yuchen Yao², Ethan D Moser³, Wendy Wang⁴, Elsayed Z Soliman⁵, Thomas Mosley⁶, Wei Pan⁷

Affiliations

¹ Division of Biostatistics and Health Data Science, University of Minnesota School of Public Health, Minneapolis, USA.
² Division of Biostatistics and Health Data Science, University of Minnesota School of Public Health, Minneapolis, USA; School of Statistics, College of Liberal Arts, University of Minnesota, Minneapolis, USA.
³ Department of Medicine, University of Minnesota Medical School, Minneapolis, USA.
⁴ Division of Epidemiology and Community Health, University of Minnesota School of Public Health, Minneapolis, USA.
⁵ Epidemiological Cardiology Research Center, Department of Cardiovascular Medicine, Wake Forest University School of Medicine, Winston Salem, NC, USA.
⁶ MIND Center, University of Mississippi Medical Center, Jackson, USA.
⁷ Division of Biostatistics and Health Data Science, University of Minnesota School of Public Health, Minneapolis, USA. Electronic address: [email protected].

PMID: 39612620
DOI: 10.1016/j.jelectrocard.2024.153832

Abstract

Aim: Create an ECG-based model to predict dementia and compare its performance with the existing Cardiovascular Risk Factors, Aging, and Dementia (CAIDE) model.

Methods and results: Participants without prevalent dementia in the Atherosclerosis Risk in Communities study were studied. Visit 4 (V4) (1996-98, mean age, 62 years) and V5 (2011-13, mean age, 75 years) were used as baselines. Incident dementia cases were adjudicated through 2019. We created parsimonious ECG models by using Cox regression with a backward selection method. C-statistic (95 % CI) of the ECG-based model (two or three ECG variables and age) was higher than the CAIDE model (seven variables) at V4 (0.72 [0.71-0.74] vs. 0.67 [0.66-0.68]) and V5 (0.70 [0.68-0.72] vs. 0.64 [0.62-0.66]). The ECG-based model was well calibrated, but the CAIDE model was poorly calibrated at V4 (P < 0.001).

Conclusion: For middle-aged and older adults, a novel ECG-based model has good discrimination that is superior to the CAIDE model in predicting dementia. Since ECG variables are readily obtainable, the ECG-based model will be easy to adopt clinically.

Keywords: Dementia; Electrocardiogram; Risk prediction.

Abstract

Grants and funding