Model-based optimization of pain management in surgical patients

点击翻译按钮获取中文摘要

PROJECT SUMMARY AND ABSTRACT Chronic opioid usage after surgery is a major contributor to the opioid epidemic, which poses a major crisis in public health. 51 million patients undergo surgery each year in the United States. Between 9-13% of surgical patients continue chronic use of opioids, leading to opioid use disorder in 8-12% of cases of chronic use. However, under half of surgical patients report adequate postoperative pain relief, which hinders recovery, increasing mortality, and length of stay. Best outcomes require personalization of treatment from patient to patient, accounting for the detrimental effects of both excessive opioid administration and uncontrolled pain. However, current clinical guidelines on pain management do not provide clear guidance on how best to adjust courses of treatment. Moreover, assessment of pain is reliant upon patient self-report, and is hindered when patients are sedated or have altered mental status. This project seeks to quantitatively understand the relationships which govern the efficacy of post- operative pain management strategies, and to characterize how different real-time physiological measures may be used to assess pain and opioid requirements. This will be accomplished through the analysis of a large dataset of electronic health record data from over 100,000 surgical procedures performed at Massachusetts General Hospital, as well as intraoperative electroencephalogram (EEG) recordings for a subset of several thousand of these procedures. Aim 1 of this project is to model analgesic response to opioids, identifying cases of excessive as well as inadequate opioid usage. We propose to model pain evolution over time using neural ordinary differential equation models, and to use learned dynamics to compute optimal treatment policies. Aim 2 of this project is to identify cases where can be improved through usage of non- opioid treatment modalities. This can also be accomplished through modeling of pain dynamics, or through statistical analyses of the outcomes of cohorts of patients receiving different treatment modalities. Aim 3 of this project is to compute intraoperative correlates of postoperative pain state from EEG data. Signal processing methods as well as deep learning will be used to extract features from EEG data related to sedation, loss of consciousness, and pain. We will also study the relationship between intraoperative interventions and postoperative outcomes. This project has the potential to reduce excess opioid usage and improve pain management, improving post-surgical clinical outcomes and reducing the incidence of opioid abuse disorder. Our results will also provide the ability to objectively assess pain and treatment requirements.