Author: Michael Vlessides
The first-ever validated predictive model for patients undergoing spine surgery not only elucidates factors that determine postoperative pain, but also opens up the possibility of real-time input from machine learning algorithms. Among other things, the researchers demonstrated that the use of one nonopioid medication during surgery did not result in clinically meaningful pain relief, whereas two or more did so.
Although several perioperative factors have been found to consistently correlate with the severity of postoperative pain, the interrelatedness of these factors—along with patient selection bias—make it difficult to identify modifiable factors that influence postoperative pain, said Quentin J. Baca, MD, PhD, an instructor in anesthesiology, perioperative and pain medicine at Stanford University, in California. What’s more, few validated predictive models of postoperative pain have been developed. “We could make a big difference in the clinical outcomes of our patients if we could predict who is going to have a tough time, then change our interventions based on those expectations,” said Dr. Baca, who is also the lead author of the study.
To help develop the model, the researchers turned to the PAIN-OUT initiative, an international registry that assesses and analyzes clinical and patient-reported outcome data on postoperative pain. A total of 1,008 patients who underwent spine surgery under general anesthesia, between 2011 and 2013, comprised the study population.
“In a controlled environment, you make one intervention—such as giving patients ketamine—and measure the outcome,” he continued. “The machine learning algorithm, on the other hand, works best with messy real-world data. So the algorithm considers that ketamine may be part of the equation, but also considers other factors like chronic pain or gender. It allows us to look at all those things together.”
The performance of the predictive model was then validated in an independent patient cohort, and its stability was characterized through bootstrap modeling. Linear regression analysis of the key factors selected by the model identified perioperative factors that may influence postoperative pain.
Remifentanil a Surprise Predictor
Reporting at the 2018 Joint World Congress on Regional Anesthesia and Pain Medicine and annual meeting of the American Society of Regional Anesthesia and Pain Medicine (abstract 5527), Dr. Baca said that of 30 potential factors, bootstrap modeling identified eight robust predictive factors of postoperative pain in spine surgery patients:
“Those same eight factors came out almost every time,” he said. “Interestingly, age ends up not being there all the time, although it’s often reported as predicting pain outcomes.” The study’s most clinically significant finding was the relationship between postoperative pain and the number of classes of nonopioid analgesics used during surgery (P=1.1×10-10).
“It doesn’t matter what type of nonopioid it was, just how many different classes the patient got,” Dr. Baca said. “There was a statistically significant difference between zero and one nonopioid pain medication, but only a small reduction in patients’ reported pain scores. However, when patients received two or more nonopioid medications intraoperatively, there was a larger, nearly 25% reduction in their pain scores.”
The use of intraoperative remifentanil was found to be strongly associated with increased pain scores (P=1.7×10-6). “Is it that these patients are having opioid-associated hyperalgesia?” Dr. Baca said. “Or is it that we treat their pain while they’re on the remifentanil but we don’t do a great job of transitioning off? This study doesn’t answer that, but helps illustrate the problem.”
The identification of factors such as these, the researchers concluded, present opportunities for interventions to improve pain control in the future. “I think this is a nice way to start thinking about how we build predictive models and use machine learning algorithms as tools for looking at our own clinical practices or hospital systems.
“And as electronic databases have become more and more accessible, this becomes easier and easier to do,” Dr. Baca added. “Eventually, we may be able to see, in real time, the things that matter for any patient in any setting.”
Hiroyuki Yoshihara, MD, PhD, an assistant professor of orthopedic surgery and rehabilitation medicine at SUNY Downstate Medical Center, in New York City, was encouraged by the researchers’ efforts to determine predictors of postoperative pain in spinal surgery. “If such factors are validated by other researchers, the protocol should be standardized across the country,” he said. “Then each patient’s specific protocol can lead to not only good patient care, but also a red uction of unnecessary medication use and cost reduction.”
Yet as Dr. Yoshihara pointed out, therapeutic innovations may have changed the treatment landscape in these patients in recent years. “This study was performed based on data from 2011 to 2013. Recently, surgical site infiltration is frequently performed during spinal surgery with liposomal bupivacaine. Based on my experience, this medication works well for about two days after the surgery. Therefore, I hope this medication is also included as a variable in the future studies.”