Using machine learning to predict cardiovascular risk using self-reported questionnaires: Findings from the 45 and Up...

Area of Science:

• Cardiovascular disease risk prediction
• Machine learning in healthcare
• Epidemiology

Background:

• Machine learning (ML) models demonstrate superior performance over traditional statistical methods for risk prediction.
• Developing accurate ML-based risk prediction models for cardiovascular mortality and ischemic heart disease (IHD) hospitalisation using self-reported data is crucial.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting cardiovascular mortality and IHD hospitalisation.
• To assess the utility of self-reported questionnaire data in these risk prediction models.

Main Methods:

• Utilized data from the retrospective 45 and Up Study (New South Wales, Australia, 2005-2009).
• Included 187,268 participants without prior cardiovascular disease, linked to hospitalization and mortality records.
• Compared various ML algorithms including SVM, neural networks, random forests, logistic regression, and survival methods (Cox regression, random survival forest).

Main Results:

• The best model for cardiovascular mortality prediction was Cox survival regression (concordance indexes: Uno's 0.898, Harrel's 0.900).
• The optimal model for IHD hospitalisation prediction was also Cox survival regression (concordance indexes: Uno's 0.711, Harrel's 0.718).
• Models were developed using self-reported questionnaire data.

Conclusions:

• Machine learning models derived from self-reported data exhibit strong predictive performance for cardiovascular outcomes.
• These models hold potential for initial screening to identify individuals at high risk for cardiovascular events.
• Facilitates early detection and intervention, potentially reducing the need for extensive investigations.