Just wanted to introduce a potentially very impactful, but not super well known, research area: using reinforcement learning to develop data-driven treatment strategies in medicine.

In reinforcement learning, decisions are made based on a system’s understanding of its current state and its goal. In medicine, these decisions may include whether to prescribe a medication or order a test.

The state in reinforcement learning is determined by a set of factors, such as patient characteristics.

Reinforcement learning algorithms are driven by a reward signal, which in medicine might be based on patient outcomes like mortality, morbidity, and quality of life.

Reinforcement learning algorithms can be trained through trial and error, adjusting their decisions based on feedback. While this approach is not practical in medicine, alternative methods, such as causal inf, can be used to train new treatment strategies based on observed data.

Reinforcement algorithms have promise to ultimately help healthcare provided and patients make better decisions, but these algorithms will not replace human decision makers.

To learn more, see eg https://pubmed.ncbi.nlm.nih.gov/25401119/ or the reinforcement learning book by Sutton et al.