Sensitivity Analysis of Stroke Predictors Using Structural Equation Modeling and Bayesian Networks

This study applies and validates causal graphical models for the task of assessing the risk of a stroke given the stroke patient data. A probabilistic causal (Bayesian) network is designed to evaluate the risk factors identified by a stroke expert. Structural Equation Modeling is applied on empirical data to provide a quantitative assessment of causal relationships among the variables in the Bayesian Network, thus statistically validating the expert's knowledge. Several scenarios of risk assessment using the inference mechanism on the Bayesian network are demonstrated.