Machine learning for frailty assessment and outcome prediction in cardiovascular disease: A scoping review

Frailty is common among individuals with cardiovascular disease (CVD) and is a strong predictor of adverse outcomes. Machine learning (ML) methods have been applied independently in CVD research for risk prediction and in frailty research for detection and classification, yet applications combining both domains have not been systematically characterised. This review examined the scope, methodology, and performance of ML applications for frailty assessment and outcome prediction in populations living with CVD.

A scoping review following PRISMA-ScR guidelines was conducted on five databases (January 2014-May 2025). Eligible studies included frailty measures and applied ML methods in CVD populations. Data were extracted on study design, frailty measure, data sources, ML method, validation approach, and performance metrics.

Thirty-one studies published between 2016 and 2025 were included; most (65%) were published in 2023 or later. Most studies (71%) examined heart failure populations. Electronic health records (n = 27) and clinical assessments (n = 21) were the primary data sources. Thirteen frailty measures were used, with the Frailty Phenotype, Frailty Index, and Clinical Frailty Scale most common. Twenty-nine studies used supervised ML for prediction targeting mortality (n = 12), frailty status (n = 10), and hospitalisation including readmissions (n = 7). Logistic and penalized regression were most common. Frailty prediction performance ranged from AUC 0.80-0.96, mortality prediction from AUC 0.70-0.93. Only three studies conducted external validation and four reported calibration metrics.

The literature is characterised by supervised prediction on tabular data with minimal external validation. Methodological limitations and heterogeneity currently restrict generalisability and clinical translation.