Risk prediction of progression from normal cognitive function to Alzheimer's disease in elderly aged 65 and above based on deep learning methods

Abstract

Background

Alzheimer's disease (AD) is a severe neurological disorder for which a complete cure is not currently available. Therefore, predicting the risk of AD in elderly individuals with normal cognitive function is crucial for early prevention, treatment, and family-provided daily care preparation.

Objective

This study aimed to establish a risk prediction model for the progression from normal cognitive function to AD in elderly via deep learning (DL) methods to provide a reference for clinical decision-making and the development of screening tools for the early diagnosis of AD.

Methods

DeepSurv, DeepHit, and Cox models were constructed, and the consistency index (C-index), integrated Brier score (IBS), and area under the ROC curve (AUC) were used to evaluate the accuracy, calibration and discriminative power of the three prediction models.

Results

The overall predictive ability of the model was relatively stable, with concordance indices of 0.82 (DeepSurv), 0.83 (DeepHit), and 0.81 (Cox) and IBSs of 0.08, 0.07, and 0.05, respectively. From the perspective of the C-index indicator, the consistency of the deep learning model was better than that of the Cox model.

Conclusions

Risk prediction models for the progression from normal cognitive function to AD can be established using easily obtainable early-stage predictors, which are expected to be used for rapid screening of the risk of developing AD in elderly after clinical validation.