Causal machine learning for extracting insights from observational radiotherapy data

Determining true side effects of treatment can be challenging because many side effects can take years to develop or be rare enough that they require a large clinical cohort to detect 1 , 2 , 3 . Thus, while clinical trials would be the most rigorous way to quantify such side effects, the length of study and resources required to detect some side effects might be prohibitively expensive 4 , 5 . Many treatment side effects end up being uncovered correlationally from retrospective studies. But while retrospective studies are helpful for suggesting potential side effects, without a controlled study, it can often be hard to determine if the side effects are truly caused by the treatment itself 2 , 3 , 6 .