Development of a risk prediction model for Alzheimer's disease based on the UK Biobank prospective study

Abstract

Background

Early prevention and intervention for Alzheimer's disease (AD) are critical due to the absence of effective therapeutic treatment. However, a widely accepted risk prediction model for AD has yet to be established.

Objective

To develop a novel risk prediction model for AD by leveraging recent advances in identifying risk factors, focusing on multi-omics data.

Methods

Genetic data from the UK Biobank were employed to calculate the polygenic risk score (PRS) using the clumping and thresholding (C + T) method. Univariate Cox regression and Elastic Net Cox models were utilized to identify significant predictors in the training cohort. Subsequently, a multivariate Cox regression model was developed to construct the prediction model, which was visualized using a nomogram. The performance of the model was evaluated through calibration curves, receiver operating characteristic (ROC) curves, and the Hosmer-Lemeshow test.

Results

Ten risk factors, including age, education, family history of dementia, diabetes, depression, hypertension, anemia, coronary heart disease (CAD), falls and PRS, were identified as significant predictors through Cox regression and Elastic Net Cox model. The model demonstrated strong predictive performance, with area under the curves (AUCs) of 0.864 [95% CI: (0.814, 0.911)], 0.860 [95% CI: (0.842, 0.876)], and 0.842 [95% CI: (0.819, 0.863)] at 5, 10, and 14 years, respectively, in the validation cohort.

Conclusions

Incorporating colocalized single nucleotide polymorphisms (SNPs) into the PRS derived using the C + T method significantly enhances predictive accuracy. This study highlights the importance of integrating multimodal patient data, including colocalized genetic information, to refine AD risk prediction.