Opioid therapy is the cornerstone of management of pain in the ICU. However, opioids present numerous side effects and are highly addictive. In fact, it is estimated that over 130 Americans die every day from an opioid overdose. Adequate opioid therapy, personalized to each patient's needs, is therefore essential. Unfortunately, ICUs are frenetic environments and clinicians are often unable to make optimal decisions or to continuously adapt therapy in real time based on the evolving patient physiological state.
To address many of these issues and augment physicians' decision making with information about what an optimal therapeutic approach may look like, we propose to leverage the latest advancements in artificial intelligence. Specifically, we focus on deep reinforcement learning, which can learn optimal state-action policies using training data that does not represent optimal behaviors. We are therefore able to train the machine learning model to recommend optimal opioid interventions using training data that does not contain optimal decisions.
Opioid analgesia in the ICU is a complex decision problem influenced by multiple factors, and extensive work will be required to develop systems that can be deployed in real clinical environments.