Among the most prevalent surgical issues is postoperative discomfort that endures long after the surgical cut has mended, affecting between 10% and 35% of the approximately 300 million individuals globally who undergo surgical procedures annually.
The cause of this postoperative pain remains ambiguous. The web of risk factors can be challenging to decipher. Discomfort arises not only from surgical injury but also from a multifaceted interaction among the peripheral and central nervous systems, the immune system, and an individual’s emotional and cognitive capabilities to interpret pain.
This is where machine learning becomes relevant. Utilizing data gathered before surgery, machine-learning algorithms can discern the numerous factors involved to foresee who may endure ongoing postoperative pain.
Past clinical studies aimed at preventing this pain have proven ineffective when attempting to mitigate specific risk factors across a highly heterogeneous group of surgical patients.
“Chronic postoperative pain is exceedingly intricate,” remarked Simon Haroutounian, an anesthesiology professor at Washington University School of Medicine in St. Louis. There is no singular formula for calculating a person’s risk, he noted.
“It’s not just a straightforward 1 + 1 kind of situation, where we gather a few metrics and create an accurate risk profile,” Haroutounian explained. “This is where we genuinely hope that machine learning can present an edge, uncovering some of those finer contributors to an individual’s risk.”
Haroutounian is involved in a multidisciplinary team at WashU investigating this issue, including Chenyang Lu, director of the AI for Health Institute and the Fullgraf Professor of Computer Science and Engineering at the McKelvey School of Engineering.
In research published in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, Lu and the team illustrate how machine learning can assist doctors aiming to avert chronic postoperative pain. Most critically, the system not only anticipates who is likely to experience that pain but also provides uncertainty assessments for each prediction.
Effectively communicating uncertainty can play a vital role in informing physician decisions. Lu and the team aimed not only to estimate patient risk but also to indicate how confident the AI is regarding that risk estimate, hence they developed an “uncertainty-aware” machine-learning model.
“It empowers the models to express, ‘I don’t know,’ and quantify that uncertainty,” stated Ziqi Xu, a PhD candidate in the Lu lab and the primary author of the paper.
A common challenge in AI-driven clinical decision support systems is their tendency to deliver a binary answer without clarifying how confident the machine is in that response, Lu pointed out. He likened it to employing generative AI applications like ChatGPT: the machine can be “certain” in its responses and prompts, even if it leads to inaccuracies.
However, clinicians need to understand the level of uncertainty associated with predictions so they can apply their own expertise to make informed decisions. Humans and machine-learning systems are designed to collaborate, and “when uncertainty isn’t conveyed in a precise way, it can lead to complications,” Lu added.
To generate those assessments, the team recruited 782 patients to participate in their study. They requested participants to complete a series of daily survey questions sent to their smartphones in the days to weeks leading up to surgery. Not all patients took the time to fully respond to the surveys, so missing data was included in their uncertainty assessments.
Then Lu integrated the survey outcomes alongside clinical data such as a patient’s medical history, laboratory results, and more. His team designed a new model that will provide an uncertainty estimate partly based on how much information the patient supplied and individual risk factors.
The model might convey: Patient X has a 30% chance of experiencing lingering pain, but there is a 50% likelihood of “uncertainty” in that estimation. In such scenarios, doctors will need to delve deeper and utilize their clinical knowledge to assist patients in making the optimal choice for managing their pain.
Conversely, the model may indicate that patient Y has a 10% probability of developing chronic pain, and the model is 80% certain of that assessment. In this case, the physician can more confidently rely on the predicted likelihood of persistent pain risk.
When compared to other prediction algorithms, the team discovered that their model achieves superior performance and provides the best model for “calibration performance,” indicating that those uncertainty estimates are both significant and accurate.
From data to the doctor
Incorporating the model into the clinical decision support process is the subsequent phase for the research, Lu stated.
Physicians aim to predict who will face chronic postoperative pain using data but, importantly, “we also seek to comprehend why,” Lu added. “Understanding causality is crucial, as this knowledge can aid in developing interventions.”
Machine learning can facilitate this discovery process by identifying the variables most significantly linked with persistent pain, which can inform better clinical trials.
For some individuals, the determinants for the risk of postoperative pain are more behavioral, and cognitive behavioral therapy (CBT) interventions could propose solutions.
However, other patients might endure pain due to a dysregulated immune response to surgery, in which case CBT approaches may be inadequate. The emphasis may need to shift toward interventions aimed at modifying the immune or inflammatory responses to surgery, Lu said.
This ongoing endeavor — focused on refining the model and revealing the causes of chronic postoperative pain — is backed by a $5 million grant from the National Institutes of Health (NIH). As the team continues validating their predictive algorithm, the next goal will be to develop tailored interventions based on each patient’s risk profile.
Understanding what factors contribute to susceptibility or resilience to postoperative pain — and exploring strategies to address these risks — could ultimately significantly affect the number of individuals suffering from pain, Haroutounian noted.
Ziqi Xu, Jingwen Zhang, Simon Haroutounian, Hanyang Liu, Zihan Cao, Gabrielle Rose Messner, Harutyun B Alaverdyan, Saivee Ahuja, Rahual Koshy, Joel Hanns, Madelyn Frumkin, Thomas L. Rodebaugh, and Chenyang Lu. Integrating Uncertainty into Predictive Models Utilizing Mobile Sensing and Clinical Data: A Case Study on Persistent Post-Surgical Pain. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 9, 2, Article 58 (June 2025) https://doi.org/10.1145/3729488.
This research was funded by a CDMRP grant from the U.S. Department of Defense to Dr. Simon Haroutounian, with additional support from the NIH grant 1RM1NS135283-01 to Dr. Simon Haroutounian and Dr. Chenyang Lu (alongside Drs. Meaghan Creed, Pratik Sinha, Thomas Rodebaugh, and Andrew Shepherd), as well as backing from the Fullgraf Foundation to Dr. Chenyang Lu.
Originally published on the McKelvey Engineering website
The post Predicting pain with machine learning appeared first on The Source.