Background
Recently, with the increasing amount of data, rapid advancement of mathematical algorithms, and ease of access to large-scale computing resources, developing predictive analytics using machine learning technique has become a prevalent direction in many fields. In this work, we propose a novel big-data-based framework for patients that have a history of consuming selective serotonin reuptake inhibitors in order to measure their risk of stroke two-years into the future using information from three-years back.

Method
We construct this index using massive amount of medical insurance records in NHIRD. The index is derived by training gradient boosted decision tree with a 963 variables – characterizing a complete risk profile of a patient at the time of doctor's visit using machine learning approaches.

Result
We validate the framework's predictive power using unweighted average recall, and it achieves 76.9% accuracy. With this framework, we further uncover important risk factors, such as age, past history of stroke incidents, and even attributes related to the total amount of money involved in drug prescription. All of the risk factors are derived in a data-driven fashion. Furthermore, we demonstrate a simple use case of this index to systematically identify clusters of risk groups and compare the stroke risk of between cases with SSRI-prescription at the time of doctor-visit versus none.

Conclusion
This novel index provides a holistic characterization of SSRI-intake patient's risk of stroke that is currently lacking, and it offers a new perspective and opportunities for physicians to make informed clinical suggestions.