Highlighted Projects
![](https://tarheels.live/kosoroklab/wp-content/uploads/sites/1826/2022/08/white-survival_cho.png)
Dynamic treatment regimes for survival outcomes
By Hunyong Cho
With a goal of maximizing cancer patients’ survival time, deriving an optimal dynamic treatment regime (DTR) requires careful consideration of censoring. To this end, we developed a flexible Q-learning framework that uses random forests.
![](https://tarheels.live/kosoroklab/wp-content/uploads/sites/1826/2022/08/white-vlearning_kahkoska.png)
V-Learning and mHealth methods for management of type 1 diabetes
By Anna Kahkoska
In contrast to standard settings for dynamic treatment regime (DTR) estimation, mobile health (mHealth) applications provide a large number of observations per individual at a very fine granularity. One such example is type 1 diabetes, where DTRs can have significant implications on the day-to-day management of blood glucose levels. We have developed a reinforcement learning method, V-learning, which attempts to alleviate the difficulties of mHealth data by estimating an optimal policy without posing modeling assumptions on the data generating process.
![](https://tarheels.live/kosoroklab/wp-content/uploads/sites/1826/2022/08/white-knee_kim.png)
Precision medicine for knee osteoarthritis
By Siyeon Kim
The precision medicine approach has been actively applied to arthritis studies. We have developed a new tree-based method, Random Forest Informed Tree-based Learning, to identify underlying patient characteristics that impacted different improvements in patient outcome while incorporating the advantages of random forests.