My research focus is on understanding the impact clinical decision-making and interventions have on patient outcomes. In a clinical environment, it can be difficult to know if decisions such as ICU triage are benefitting patients as expected. These decisions are not made randomly, and patients more likely to go to the ICU are also more likely to be severely ill and, thus, more likely to experience negative outcomes. Therefore, the data alone would make it seem as though ICU is not beneficial. I use causal inference-based approaches to gain clarity on whether such interventions are beneficial and, specifically, what types of patients would benefit from certain actions.
Another focus is monitoring the performance of ML/AI clinical decision support tools after deployment. Models are often trained on retrospective data and then deployed if they perform well. However, clinical practices will change and evolve over time which can alter the features and/or outcomes of the model. Furthermore, the presence of the model itself can change clinical workflows and patient outcomes. My work involves being able to identify and report these changes, and strategies to mitigate risks associated with model deployment.