Title : Predicting reductions in acute pain and opioid consumption with non-opioid analgesics: A machine learning analysis of randomised controlled trials (OPERA study)
Abstract:
Background: Acute pain is one of the most common complications following surgery and has a detrimental effect on patients’ quality of life. It also poses significant healthcare and economic burdens. The role non-opioids play in managing post-surgical pain is increasing, either as a sole agent or in combination with other analgesics. Advances in machine learning (ML) presents opportunities to use data from randomised controlled trials (RCTs) in developing algorithms that can guide treatment recommendations. This study is part of a wider systematic review and aims to use ML to develop an algorithm that aims to predict reductions in acute pain and opioid consumption with non-opioid analgesics.
Methods: We included studies in adult patients undergoing a surgical procedure, under general or regional anaesthesia, which evaluated non-opioid analgesia commonly used to treat acute pain. Other procedures, dental surgery and ophthalmic procedures were excluded. The risk of bias was assessed using the Cochrane Risk of Bias tool Version 2 by two study authors (and a third if consensus could not be reached). Predictors include baseline risk (control group mean), age, sex, risk of bias domains, drug within class, route of administration, type of surgery country of origin, year of publication, postoperative doses given, concurrent regional anaesthesia or non-opioids and pre-emptive doses.
Results: and Conclusions We have identified 1375 RCTs for potential inclusion. Data extraction is ongoing. The implications of our results are:
1) Allow clinicians to predict in what clinical circumstances analgesics are more effective
2) Present a method for meta-analyses to present results
3) Demonstrate how the transitivity assumption can be violated in network meta-analyses
