Investigating the impact of multilevel treatment options on statin patient outcomes: a counterfactual analysis

Abstract

Statins are among the most commonly prescribed medications for lowering cardiovascular disease riskby reducing LDL-C levels. However, statin treatments can lead to several clinically significant side effects, such as myopathy and liver problems. Therefore, a machine learning (ML) system integrated with electronic health records (EHR) is essential for making personalized treatment recommendations. This system would consider medication side effects, treatment efficacy, and the risk of discontinuation, while addressing confounding factors inherent in observational data. Confounding occurs when conclusions like “Drug A is better than Drug B for alleviating symptoms” are drawn without fully accounting for factors that influence both treatment and outcomes. By mitigating these biases, ML models can assist doctors in making more informed prescription decisions, such as predicting what might happen if a patient receives a lower dosage. This thesis explores the combination of machine learning and causal inference techniques to develop counterfactual (CF) predictions for statin-related outcomes. Specifically, it focuses on predicting statinassociated symptoms (SAS) and other relevant outcomes in real-world settings. Additionally, the thesis includes a counterfactual outcome simulation for both baseline CF models and two-stage CF model, which is used to evaluate the accuracy of these predictions. Finally, the thesis presents discussions on the broader applications of counterfactual predictions in clinical decision-making.