Clinical prediction models using artificial intelligence approaches in dementia.

Area of Science:

• Neurology
• Medical Informatics
• Machine Learning

Background:

• Nearly half of dementia cases are potentially preventable, highlighting the need for early detection and intervention.
• Artificial intelligence (AI)-enhanced clinical prediction models leverage machine learning (ML) to integrate diverse data for improved dementia diagnosis and prognosis.
• This systematic review assesses AI-based prediction models for dementia development, performance, and clinical use.

Purpose of the Study:

• To systematically review and evaluate the development, performance, and clinical applicability of AI-based prediction models in dementia.
• To assess the accuracy, bias, and generalizability of AI models for predicting dementia onset.

Main Methods:

• Systematic literature search of PubMed, Embase, and Web of Science up to October 2024 for AI models predicting dementia onset.
• Assessment of 21 included studies (over 1 million participants) using PROBAST for accuracy, bias, and generalizability.
• Data extraction followed TRIPOD and CHARMS frameworks, focusing on study design, demographics, predictors, and performance metrics.

Main Results:

• AI models demonstrated good predictive accuracy (mean AUC 0.845) across diverse datasets.
• Machine learning methods like random forests and support vector machines outperformed traditional models.
• Internal validation was common, but external validation was limited; calibration and generalizability remain challenges.

Conclusions:

• AI-based prediction models offer significant potential for early dementia detection and personalized care strategies.
• Clinical integration requires addressing external validation, data representativeness, and model interpretability.
• Future research should prioritize robust validation and ethical considerations for optimal dementia care implementation.