A scoping review on artificial intelligence-based tools for cardiovascular disease risk prediction

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Cardiovascular Disease Prediction

Background:

• Cardiovascular disease (CVD) is a leading global cause of mortality, underscoring the need for effective early risk prediction.
• Artificial intelligence (AI) models offer potential for enhanced CVD risk prediction, but concerns regarding interpretability, factor reliability, and validation persist.
• This scoping review critically examines AI-based prognostic models for predicting CVD risk in individuals without pre-existing CVD.

Purpose of the Study:

• To assess the scope, methodologies, and reporting quality of studies employing AI for CVD risk prediction.
• To identify gaps in the validation, calibration, and clinical utility reporting of AI-driven CVD risk models.
• To provide recommendations for improving the transparency, reproducibility, and clinical applicability of these models.

Main Methods:

• Systematic literature search across major scientific databases (PubMed, IEEE Xplore, Web of Science, Scopus, Google Scholar).
• Screening and data extraction adhering to PRISMA-ScR and JBI Guidelines for Scoping Reviews.
• Evaluation of included studies against the TRIPOD-AI reporting checklist; protocol registered on Open Science Framework (OSF).

Main Results:

• Thirty studies published post-2017 met inclusion criteria, primarily using established machine learning algorithms on unimodal clinical data.
• While multimodal AI studies showed strong performance, they were few and heterogeneous; comparisons with traditional scores yielded comparable or modest improvements.
• Crucially, only seven studies reported external validation, and none reported calibration; sensitivity was reported in only four studies, hindering clinical utility assessment.

Conclusions:

• Current AI-based CVD risk prediction tools exhibit significant deficits in validation, calibration reporting, and clinical utility assessment, precluding immediate clinical deployment.
• Future research must prioritize external validation across diverse populations and mandate reporting of calibration and sensitivity metrics alongside discrimination.
• Adoption of standardized reporting guidelines (e.g., TRIPOD+AI, PROBAST+AI) is essential for enhancing transparency and reproducibility in AI CVD risk modeling.