Presentation (CTI)

OBJECTIVES: Statins are highly effective lipid-lowering drugs used for prevention of atherosclerotic cardiovascular disease (ASCVD). However, statin discontinuation or dose reduction due to statin-associated symptoms (SAS) remains a major challenge. The objective is to develop a machine learning model to suggest personalized optimal statin-dose combination to prevent the occurrence of SAS.
METHODS: Optum Research Database was used to identify ASCVD patients from Jan 2017 to Dec 2018 who initiated statin within 30 days of ASCVD event. Patients with continuous eligibility through 12 months pre- and 18 months post-index were included in the analysis. Patients with ASCVD event or statin prescription in the baseline period were excluded. Potential factors associated with statin discontinuation and risk of developing SAS were used as predictor variables for developing the model. An artificial neural network model was developed to predict the risk of discontinuation or SAS. An optimization method was applied to select the optimal statin-dose combination for a patient based on the risk predicted.
RESULTS: The predictive power of the model was evaluated based on ROC-AUC, PR-AUC, precision, recall and F1 score. The model achieved a ROC-AUC & PR-AUC of 86% and 82% respectively. The precision, recall and F1 score for the model were 84%, 80% and 82% respectively. To optimize the personalized statin therapy, the same model was applied on the statin-dose combination based on either single optimization (optimizing either SAS or discontinuation) or multi-optimization objective (both SAS & Discontinuation). A proof of concept will be developed for an individual based on the clinical and claim profile to recommend the ideal statin-dose combination optimized for both SAS development and statin discontinuation.
CONCLUSIONS: Artificial neural network models are a feasible way to prevent or minimize the risk of SAS and statin discontinuation. Further fine tuning is needed to increase the prediction power.