HSS Presents New Research Leveraging AI to Uncover Insights Related to Pain Risk and Anesthesia Education at ASRA Annual Meeting

At this year's American Society of Regional Anesthesia and Acute Pain Medicine (ASRA) annual meeting, investigators at Hospital for Special Surgery (HSS) presented significant studies leveraging artificial intelligence (AI) to provide insights into long-term pain risk after surgery and what patients want to know about anesthesia. These insights may ultimately help guide anesthesiologists' consultations with patients scheduled for surgery.

What follows are highlights from these studies:

HSS Study Uses Machine Learning to Predict Risk of Long-Term Pain After Knee Replacement Surgery

A new study led by researchers at HSS used a type of AI known as machine learning (ML) to identify key clinical and biological factors that raise a person's risk of having persistent pain after total knee arthroplasty (TKA). Risk factors included elevated levels of certain inflammatory cytokines (proteins) in the blood, severe preoperative pain, and a longer period of tourniquet use in the operating room.

"These findings highlight the importance of incorporating biological markers like cytokine levels with patient-specific pain profiles and what the surgeon does during the operation to more accurately predict the risk of long-term pain after surgery," says Meghan Kirksey, MD, PhD, an anesthesiologist at HSS and senior author of the study.

ML is a specialized approach that uses algorithms and statistical models to analyze patterns in large amounts of data to learn, predict, and make recommendations.

"Machine learning is allowing us to look at patient and clinician information in new ways," says Alexandra Sideris, PhD, Director of the HSS Pain Prevention Research Center and a coauthor of the study. "It gives us a multidimensional approach to understanding patients' pain experience that we didn't have in our arsenal even five or 10 years ago."

One in five people has significant knee pain months after having TKA, also known as total knee replacement surgery. "The lingering pain greatly affects their daily activities and quality of life, so that's why it's an important focus for us," explains Dr. Sideris.

Persistent postoperative pain (PPP) is typically flagged when a patient has lasting pain at the site of the operation that is above a four on a scale of zero to 10 and severely impacts their activities of daily living three to six months after surgery.

The researchers used four different ML models to analyze data from a previously published study that collected comprehensive clinical information and blood samples from 160 patients before and after TKA at HSS. The new study identified key predictors associated with PPP beyond risk factors that were already known such as sex (women tend to have a higher risk), preexisting pain, and mental health issues like anxiety and depression.

The findings showed that having high blood levels of an inflammatory marker called TARC immediately after surgery raises the risk of PPP. "This molecule hasn't been extensively studied in pain, but the evidence shows that it was consistently associated with persistent pain six months after surgery across all four ML models we tested," notes Dr. Kirksey.

Other top predictors of PPP that emerged from the analysis included a higher preoperative pain score at rest, longer tourniquet time (a tourniquet is a device that squeezes the leg to help clear the area of blood flow during surgery), and higher blood levels of other inflammatory cytokines right after surgery.

In this study, researchers entered 318 clinical and biological characteristics collected from patients in the older study and asked each of the ML models to figure out the most important features associated with pain after TKA. The researchers also evaluated the accuracy of the ML models they used and found that XGBoost was the most informative.

"To my knowledge, this is the first study that looked at all of this information and tried to make sense of the best ML approach to use," says Dr. Sideris. "What is exciting to us is that there was one feature - TARC - that consistently popped up across all four models and it wasn't on anybody's radar beforehand…this gives us hope that ML can help us identify with high integrity targets that haven't been studied before."

The researchers note that more research is needed to know if their findings can be used to impact clinical care. "Our goal is to be able to use these tools and data to tailor pain management strategies, prevent long-term complications, and personalize treatment decisions," says Dr. Kirksey.

HSS AI Analysis Reveals What Patients Are Googling About Local Anesthesia

A new study by researchers at HSS used AI to systematically evaluate the types of questions that patients are Googling related to regional (local) anesthesia, identify websites that are frequently presented in search results, and assess the quality of the information provided.

"We knew that patients frequently search for information about anesthesia online, but we wanted to know exactly what they were looking for so that we could proactively address these topics and concerns in our conversations in the clinic and our patient education materials," says Jashvant Poeran, MD, PhD, Director of Research in the Department of Anesthesiology, Critical Care and Pain Management at HSS and lead author of the study.

The analysis found that most patients' questions focused on risks, complications, and details surrounding medications, awareness during sedation, nerve block duration, and the recovery process. The study also found that while the overall quality of information accessed during Google searches was accurate, the source of that information was not always clear.

Anesthesiologists see patients in the clinic for a short period to prepare them for surgery and manage expectations.

"There's so much information being conveyed during that limited time that sometimes patients forget what to ask, or they don't even know what they should be asking during the visit," explains Dr. Poeran. "Our study results will help us anticipate some of their questions and give us a starting point when we sit down with them so it's not so much of a guessing game."

The researchers entered seven search terms into Google Web Search: "regional nerve block," "regional anesthesia," "peripheral nerve block," "pain block," "neuraxial anesthesia," "epidural anesthesia," and "spinal anesthesia." The top 200 questions in the "People Also Ask" section and its associated websites were collected, totaling 1,400 question and website combinations.

The authors then used AI to categorize themes and assess website quality. They found that most questions pertained to facts around risks and complications, comparisons between different techniques and approaches, technical details, and indications.

"We were expecting questions around risks and complications, but it was surprising that so many patients were looking at technical details, specifically around sedation," notes Dr. Poeran. "They were not always aware that you can be awake for a peripheral nerve block, for example."

Because patients' questions are linked to specific websites, researchers also wanted to know where patients were being referred to and how reliable the information presented was. The AI analysis found that 55% of websites were academic, 19% were government, and 11% were public/social media sources. Information on government and academic/hospital websites scored the highest in terms of accuracy, and medical practice websites scored the lowest.

Dr. Poeran cautioned that some patients' questions are nuanced, and online information can bias a person in the wrong direction.

"For example, if you ask whether regional anesthesia is better than general anesthesia you will be able to get generic information about those approaches online, but you need to talk to your doctor to get a more personalized recommendation based on your specific circumstances," says Dr. Poeran.

While the study revealed important questions, it's not all-encompassing. "There will be questions that aren't captured by this study, but anesthesiologists can use this data to guide their consultations with patients scheduled for surgery, provide more-informative patient education materials, and refer them to the most reliable websites for more information," notes Dr. Poeran.

He plans to continue leveraging AI in his future research endeavors.

"Based on the information we gathered, we will update our educational materials with the questions we now know patients ask most often and may even provide it in different languages and reading levels," says Dr. Poeran. "We can then use AI to see how that information is perceived and understood by patients, study differences in search terms entered in different languages, etc."