Abstract 10437: “Explainable Artificial Intelligence (ai) in Cardiology”: A Tool to Provide Personalized Predictions on Cardiac Health States Among Older Adults Engaged in Physical Activity

Abstract

Introduction: As a prevention strategy, the effect of physical activity on older adults is likely heterogeneous against a background of individual ageing trajectories. While machine learning models that combine various systemic signals may aid in predictive modeling, the inability to rationalize predictions at a patient personalized level is a major shortcoming in AI. Methods: We applied a novel approach using Shapley Additive Explanations (SHAP) methodology on a dataset of older adults whose physical activity levels were studied in conjunction with changes in left ventricular (LV) structure. The SHAP approach provided intelligible visualization on the magnitude of the impact of the features in their physical activity levels on their LV structure. Results: As proof of concept, we studied n = 86 older adults (mean age 72±4 years). Using repeated K-cross validation on the train set (n = 68), we found the Random Forest Regressor with the most optimal hyperparameters, which achieved the lowest mean squared error. With the trained model, we evaluated its performance by reporting its mean absolute error and plotting the correlation on the test set (n = 18) (Fig 1). A collective force plot for individually numbered patients (horizontal axis) showing magnitude (i.e., effect) of physical parameters (higher in red; lower in blue) towards prediction of LV structure (Fig 2). For example, for patient number 6, basal metabolic rate was the predominant positive factor on LV structure, outweighing other weaker positive factors and negative factor contributed by body fat mass (Figure 3). In contrast, LV structure in patient number 11 was predicted jointly by several weaker positive factors without any single predominant factor (Figure 4). Conclusions: As a tool that identified specific features in physical activity that predicted cardiac structure on a per patient level, our findings support a role for explainable AI to be incorporated into personalized cardiology strategies.